Multi-AI Comparison
Ask once.
Get the consensus.
Get the consensus.
One question. Six AIs answer simultaneously. Where they agree — that's the truth you can act on.
Tap any AI to toggle it on/off
Tip: Use ↩ Ask again to re-run a question, ⤵ Follow up to add it to the follow-up box (requires an active session), or ⎘ Copy to copy the text anywhere.
Recent questions
No history yet. Ask a question to get started.
💬 Conversational & General
🎨 Image Generation
🔬 Research & Citations
📄 Documents & Writing
🎵 Audio & Music
Get better answers by asking smarter questions
This tool asks six AI models simultaneously and finds where they agree. The quality of the consensus depends on how well you frame your question.
✦ Be specific
Vague questions get vague answers. Include your context, constraints, and what you actually need to decide.
✗ "What's a good blood pressure medication?"
✓ "What's the best blood pressure medication for a 65-year-old male with diabetes and no prior heart disease?"
✦ Ask for a format when it helps
Tell the AIs how you want the answer structured. This dramatically improves comparability across models.
✓ "Give me a one-word answer."
✓ "List the top 3 options with pros and cons."
✓ "Explain in plain English, no jargon."
✦ Use follow-ups to drill deeper
Start broad to get the landscape, then use the follow-up box to go deeper. Each AI remembers the conversation context for up to 3 rounds.
Round 1: "What are the main treatments for Type 2 diabetes?"
Round 2: "Which of those has the fewest side effects for someone with kidney disease?"
✦ This tool works best for
Medical questions, legal concepts, financial decisions, technical comparisons, factual research, product choices, and any question where you want independent perspectives — not just one AI's opinion.
✗ Less effective for
Real-time data (stock prices, live news), highly localized questions ("best restaurant near me"), or tasks that require access to your personal files or accounts.
JZAI Training Center
Learn how to get more out of AI — from picking the right tool to crafting better prompts.
✦
How to Ask
Prompt structure, specificity, format instructions, and follow-up strategy.
›
🎯
Which AI for Which Job
Decision guide for writing, research, code, images, analysis, and more.
›
📊
Capability Matrix
What each AI can and can't do — documents, images, voice, web, code, and more.
›
💰
Pricing & Plans
Free vs paid tiers, cost-per-use scenarios, and when to upgrade.
›
⚙️
Setup Guides
Account setup, custom instructions, and first-time configuration for each AI.
›
⚡
Raw vs Optimized
Same question, two ways — see exactly what a better prompt gets you.
›
How to Ask
The quality of your question determines the quality of the consensus.
✦ Be specific
Include your context, constraints, and what you actually need to decide.
✗ "What's a good blood pressure medication?"
✓ "Best blood pressure med for a 65-year-old male with diabetes, no prior heart disease?"
✦ RCTF Framework
Role · Context · Task · Format — include all four for best results.
✓ "As a tax attorney [Role], I'm a freelancer who earned $180K [Context]. What are the top 3 legal deductions [Task]? Bullet list with dollar estimates [Format]."
✦ Ask for a format
Tell the AIs how to structure their answer.
✓ "Give me a one-word answer."
✓ "List the top 3 options with pros and cons."
✦ Use follow-ups
Start broad, then drill down. Each AI remembers up to 3 rounds of context.
Round 1: "What are the main treatments for Type 2 diabetes?"
Round 2: "Which has fewest side effects for someone with kidney disease?"
Which AI for Which Job
Every major AI tool mapped to the tasks it does best. Click any name to open it.
✍️ Writing & Long-Form Content
Best: Claude Opus
Strongest long-form writing, nuance, tone control, and instruction-following. Handles complex editing precisely.
🔬 Research & Citations
Best: Perplexity Pro (web) · NotebookLM (documents)
Perplexity cites sources in real time with every answer. NotebookLM lets you upload your own documents and ask questions across them.
💻 Coding (IDE / Editor)
Best: Cursor (AI-native IDE) · GitHub Copilot (existing IDE)
Cursor handles complex multi-file refactoring and agent tasks. Copilot integrates natively into VS Code, JetBrains, and Neovim at half the price.
💻 Coding (Model / Reasoning)
Best: Claude Opus · ChatGPT Pro
For pure code reasoning, debugging, and architecture — Claude Opus and GPT-5.6 Sol are the strongest models. Use these inside your IDE of choice.
🎨 Image Generation (Artistic Quality)
Best: Midjourney
Produces the highest quality artistic, cinematic, and stylized images. No free trial — starts at $10/mo.
🎨 Image Generation (Text in Images)
Best: Ideogram
Only AI image tool that reliably renders readable text inside generated images. Essential for posters, ads, social graphics.
Also: Adobe Firefly (commercial safe) · ChatGPT/DALL-E
🎨 Image Generation (Commercial / Licensed)
Best: Adobe Firefly
Trained entirely on licensed Adobe Stock and public domain content. IP indemnification available. The safe choice for client work and advertising.
Also: Leonardo.ai · Ideogram
🎬 Video Generation (Professional)
Best: Runway
Professional standard for AI video. Gen-4 produces cinematic quality with precise camera control. Broadest editing toolkit. Starts at $12/mo.
🎬 Video Generation (Social / Quick)
Best: Pika Labs
Lower price ($8/mo), fast generation, great for short social media clips. Max 4-second clips vs Runway's 10 seconds.
Also: Kling · Luma Dream Machine
🎬 Avatar & Talking Head Videos
Best: HeyGen
Specialized for AI avatar videos, digital humans, and lip-synced talking heads. Used for training videos, product demos, and multilingual content.
Also: Runway Act-One (character animation)
📊 Data Analysis & Charts
Best: ChatGPT Plus (Code Interpreter)
Upload a spreadsheet, ask a question, get a chart. Executes Python, creates visualizations, cleans data — all in the chat window.
📄 Documents & Presentations
Gamma generates polished presentations from a prompt. Notion AI writes and edits inside your existing workspace.
🎵 Music Generation
Best: Suno v5
Category leader for full song generation including realistic vocals. $10/mo gets commercial rights and downloads. 500+ songs/mo on Pro.
🎙️ Voice Cloning & Text-to-Speech
Best: ElevenLabs
Gold standard for voice cloning realism and TTS quality. 29+ languages. Cleanest commercial license in the category. Starts at $6/mo for commercial use.
Also: ChatGPT Voice · Gemini Live
⚡ Speed & Quick Answers
Best: Gemini Flash · Grok 4.20
Fastest frontier models for low-latency responses. Gemini 3.7 Flash launched Aug 2026. Both excellent for high-volume, fast-turnaround tasks.
Also: GPT-5.6 Luna (free) · Mistral Small
🔒 Privacy & EU Data Residency
Best: Mistral
French company, EU data residency, open-weight models. Strong choice for European businesses with GDPR requirements or organizations that can't send data to US servers.
Also: DeepSeek (self-hostable) · on-premise deployments
🤔 When to Use JZAI Consensus Tool
Best for: High-stakes decisions requiring reliability
When the answer matters — medical, legal, financial, technical — use JZAI to cross-check across 6 independent AI models simultaneously. Cross-model agreement is your reliability signal. If 5 of 6 AIs agree on the same specific answer, that's meaningful consensus you can act on.
Not ideal for: image/video/audio tasks, real-time data, personal file analysis
🤖 Free Conversational AI (No Account)
Best: Meta AI (built into WhatsApp, Instagram, Facebook)
Zero signup required if you already use Meta apps. Powered by Llama 4 Scout. Available inside WhatsApp conversations — ask questions without leaving the app. Best for casual, quick questions when you don't want another subscription.
🌐 AI with Live Web Search (Free)
Best: You.com · Perplexity
You.com searches the web in real time, cites sources, and requires no account. Perplexity is the gold standard for cited, source-grounded answers — free tier gives you unlimited basic searches. Both beat standard chatbots for any question where you need current information.
🔬 Academic Research & Peer Review
Best: Semantic Scholar · Elicit
Semantic Scholar indexes 200M+ academic papers, free, no account. Elicit reads actual paper PDFs, extracts methodology, sample sizes, and findings — purpose-built for systematic literature review. Neither hallucinates citations because they pull from real indexed papers, not training data.
Also: Consensus (plain-English answers from papers) · NotebookLM (upload your own PDFs, ask questions)
🎵 Background / Ambient Music Generation
Best: Mubert
Mubert generates continuous royalty-free background music streams in real time — focus music, lo-fi, ambient, workout. Unlike Suno/Udio which generate finished songs with vocals, Mubert is designed for background listening. Free tier available. Commercial licensing built in — safe for YouTube and streaming.
Also: Suno (full songs with vocals) · Udio (high-quality instrumentals) · ElevenLabs (sound effects + music)
Capability Matrix
✓ = Yes ◑ = Limited ✗ = No — Scroll right to see all tools. Mid-2026.
| Capability | ChatGPT | Claude | Gemini | Grok | Copilot | Perplexity | Meta AI | Le Chat | DeepSeek | You.com | Midjourney | Ideogram | Firefly | DreamStudio | Leonardo | Flux | Runway | Sora | Kling | Pika | Luma | HeyGen | NotebookLM | Consensus | Elicit | Sem.Scholar | Gh.Copilot | Cursor | Replit | v0 | Bolt | Windsurf | Notion AI | Jasper | Copy.ai | Gamma | Beautiful.ai | Suno | Udio | ElevenLabs | Mubert |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 💬 Conversational & General | |||||||||||||||||||||||||||||||||||||||||
| Text chat | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ◑ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✓ | ✓ | ✓ | ✗ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ |
| Web search | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ◑ | ◑ | ◑ | ◑ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ |
| Memory | ✓ | ✓ | ✓ | ✓ | ◑ | ✓ | ◑ | ✓ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✓ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ |
| Custom instructions | ✓ | ✓ | ✓ | ✓ | ◑ | ◑ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ◑ | ◑ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ |
| Free tier | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✗ | ✓ | ✓ | ◑ | ✓ | ◑ | ◑ | ✗ | ✓ | ✓ | ◑ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ◑ | ✗ | ✓ | ✓ | ✗ | ✓ | ✓ | ✓ | ✓ |
| 📄 Documents & Files | |||||||||||||||||||||||||||||||||||||||||
| PDF reading | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ◑ | ◑ | ✓ | ◑ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✓ | ✗ | ✓ | ✓ | ✗ | ✓ | ✗ | ✗ | ✗ | ✓ | ✓ | ◑ | ◑ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ |
| Spreadsheet analysis | ✓ | ✓ | ✓ | ◑ | ✓ | ✗ | ✗ | ✗ | ◑ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✓ | ✗ | ✗ | ✗ | ✓ | ✓ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ |
| Code execution | ✓ | ✓ | ✓ | ◑ | ◑ | ✗ | ✗ | ✓ | ✓ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ |
| 🎨 Image | |||||||||||||||||||||||||||||||||||||||||
| Image generation | ✓ | ✗ | ✓ | ✓ | ✓ | ✗ | ✓ | ✓ | ✗ | ✗ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✓ | ✗ | ✗ | ✗ | ✗ | ✗ | ✓ | ✓ | ✗ | ✗ | ✗ | ✗ |
| Image analysis | ✓ | ✓ | ✓ | ✓ | ✓ | ◑ | ✓ | ✓ | ✓ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✓ | ✗ | ✗ | ✗ | ✓ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ |
| Commercial IP safe | ◑ | ◑ | ◑ | ◑ | ◑ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ◑ | ✓ | ◑ | ◑ | ◑ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ |
| 🎬 Video | |||||||||||||||||||||||||||||||||||||||||
| Video generation | ✗ | ✗ | ◑ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✓ | ✗ | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✓ | ✗ | ✗ | ✗ | ✗ | ✗ |
| Avatar / lip-sync | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✓ | ✗ | ✗ | ✗ | ✗ | ✓ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ |
| 🎧 Audio | |||||||||||||||||||||||||||||||||||||||||
| Voice / TTS | ✓ | ◑ | ✓ | ✓ | ✓ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✓ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ◑ | ◑ | ✓ | ✗ |
| Music generation | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✓ | ✓ | ✓ | ✓ |
| Voice cloning | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✓ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✓ | ✗ |
| 💻 Coding | |||||||||||||||||||||||||||||||||||||||||
| Code completion | ✓ | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✗ | ✓ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ |
| Multi-file edit | ◑ | ◑ | ◑ | ✗ | ◑ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✓ | ✓ | ◑ | ◑ | ◑ | ✓ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ |
✓ = Yes ◑ = Limited / Paid tier only ✗ = No — Flagship paid tiers, mid-2026. Scroll right →
Tap a tool above to see its capabilities.
Pricing & Plans
All tiers, verified August 2026. Prices change — always confirm on the vendor's site before buying.
💬 Conversational & General
Free$0GPT-5.6 Luna, basic chat, ads shown
Go$8/moMore volume than Free, still ad-supported, no agent mode or deep research
Plus$20/moGPT-5.6 Sol, DALL-E, code execution, memory, browsing, 160 msgs/3hrs
Pro 5x$100/mo5x Plus limits, GPT-5.6 Sol Pro, max Codex
Pro 20x$200/mo20x Plus limits, Sora video, unlimited usage
Business$20/seatTeam admin, data privacy, SSO (annual billing)
EnterpriseCustomCustom limits, SCIM, audit logs, SLAs
✓ Plus at $20 is the best individual value. Go has ads — skip it.
Free$0Claude Sonnet 5, rolling usage budget, no Claude Code
Pro$20/mo ($17 annual)Claude Opus, 5x usage, Projects, memory, Claude Code included
Max 5x$100/mo5x Pro limits, Opus default, heavy coding use
Max 20x$200/mo20x Pro limits, power users and agents
Team$25/seat ($30 monthly)Pro features + team sharing, admin controls
EnterpriseCustomSSO, audit logs, custom data retention, SLAs
✓ Pro at $20 essential for serious work. Max 5x at $100 for heavy coding.
Free$0Gemini 3.6 Flash, daily limits, basic features
AI Pro$19.99/moGemini 3.1 Pro, 20 Deep Research/day, Gems, 5TB storage, NotebookLM Plus
AI Ultra$249.99/moDeep Think reasoning, Gemini Agent, maximum storage, early access
Workspace Business Standard$14/seatFull Gemini bundled in Gmail, Docs, Sheets, Meet
Workspace Business Plus$22/seatEverything in Standard + compliance, security
✓ AI Pro at $19.99 best for individuals. Workspace Standard best for Google-heavy teams.
Free$0Grok 4 Mini, ~10 prompts/2hrs, no image gen
SuperGrok Lite$10/moBasic image gen (480p), 2x longer chats, 1 AI agent
SuperGrok$30/mo ($25 annual)Full Grok 4, DeepSearch, Big Brain mode, Imagine, 128K context
SuperGrok Heavy$300/moGrok 4 Heavy multi-agent, 256K context, confirmed Grok 4.5 access
X Premium+$40/moGrok 4 access + ad-free X, verified badge, creator features
Business$30/seatTeam collaboration, data privacy, compliance
✓ SuperGrok at $30 sweet spot. Heavy at $300 only for frontier power users.
Free$0Unlimited basic search, 5 Pro searches/day, no model switching
Education Pro$10/moFull Pro features with student verification
Pro$20/mo ($16.67 annual)Unlimited Pro searches, model switching (GPT-5, Claude Opus, Gemini), file uploads, Comet browser
Max$200/mo ($167 annual)Highest limits, 19 orchestrated models, Perplexity Computer, priority support
EnterpriseFrom $40/seatSSO, admin controls, data privacy, API credits
✓ Pro at $20 is the best research tool value. Free is surprisingly capable for casual use.
Free$0GPT-4o based, limited daily chats, image gen
Pro$20/moPriority access, Office 365 integration, faster image gen
Microsoft 365 Copilot$30/seatFull Office AI (Word, Excel, Teams, Outlook)
✓ Best if you live in Microsoft Office. Otherwise ChatGPT or Claude are stronger.
Free$0Mistral Small, basic chat, limited usage
Pro~$14/moMistral Large, higher limits, API access
EnterpriseCustomOn-premise deployment, EU data residency, custom models
◑ Best for European data privacy requirements. Strong open-weight models.
Free$0DeepSeek V3/R1, web chat, limited daily usage
APIPay per token$0.27/1M input tokens — among cheapest frontier API available
◑ Excellent free tier for chat. API is extremely cost-effective for developers.
Free$0Web search AI, no account needed, unlimited basic queries
Pro$15/moGPT-4o + Claude + Gemini access, advanced research mode, no ads
✓ Free tier is genuinely useful. Pro worthwhile if you want multi-model access in one place.
Free$0Llama 4 Scout via meta.ai, WhatsApp, Instagram, Facebook — no account needed
✓ Completely free. Best value if you're already in the Meta ecosystem.
🎨 Image Generation
Basic$10/mo ($8 annual)3.3hr fast GPU/mo (~200 images), commercial use
Standard$30/mo ($24 annual)15hr fast GPU + unlimited relaxed, commercial use
Pro$60/mo ($48 annual)30hr fast GPU, stealth mode, max upscale
Mega$120/mo ($96 annual)60hr fast GPU, highest priority queue
✓ Standard at $30 best value for regular use. No free trial — starts at $10.
Free$0~10 generations/day, commercial use included
Basic$8/moPriority credits, faster generation
Plus$20/moMore priority credits, API access
Pro$20/moHighest limits, team features
✓ Best for text-in-images. Generous free tier. Commercial use included on all plans.
Free$025 generative credits/mo, standard image gen
Standard$9.99/mo2,000 premium credits, unlimited standard gens
Pro$19.99/mo4,000 premium credits, video gen, all models
Premium$199.99/mo50,000 premium credits, enterprise features
Creative Cloud$59.99/moAll CC apps + Firefly Pro credits built in
✓ Only fully licensed commercial AI image tool. Best for brand-safe marketing work.
Free$0150 tokens/day (~30-40 images), commercial use
Apprentice$12/mo8,500 tokens/mo, API access, fine-tuning
Artisan$30/mo25,000 tokens/mo, priority queue
Maestro$60/mo60,000 tokens/mo, highest limits
✓ Most generous free tier in image gen. Strong for game art and brand consistency.
Pay-as-you-go~$0.003–$0.05/imageFlux Pro, Flux Dev, Flux Schnell — varies by model and resolution
Via third-party appsVariesComfy UI, Replicate, Leonardo — all can run Flux models
✓ Pay-per-image model. No subscription required. Best open-weight image model as of 2026. Run it cheap via fal.ai.
Pay-as-you-go$10 = ~500 imagesStable Diffusion XL, credits never expire
✓ Credits-based, no monthly fee. Good for occasional use. Flux has largely superseded SDXL quality.
🎬 Video & Animation
Free$0125 one-time credits (not monthly), Gen-4 Turbo, watermark
Standard$12/mo ($15 monthly)625 credits/mo, no watermark, 720p video
Pro$28/mo ($35 monthly)2,250 credits/mo, 4K upscale, priority queue
Max$76/mo ($95 monthly)9,500 credits/mo, all models, highest quality
EnterpriseCustomAPI access, team seats, SLAs
✓ Professional standard for AI video. Standard at $12/mo best entry point.
Free$0Limited generations, watermark
Standard$8/moHigher limits, no watermark
Unlimited$28/moUnlimited standard generations
Pro Unlimited$58/moAll features, priority access
◑ Good for social media clips. Shorter max length (4s) than Runway. Lower price point.
Free$01 video/month, watermark, limited avatars
Creator$29/mo15 videos/mo, custom avatars, no watermark
Business$89/mo30 videos/mo, team features, API access
EnterpriseCustomUnlimited videos, custom integrations
✓ Best for talking-head and avatar videos. Strong for training content and marketing.
Free$066 credits/mo (~3 videos), watermarked
Standard$10/mo660 credits, no watermark, longer clips
Pro$35/mo3,000 credits, priority queue, 4K output
✓ Best price/quality ratio in AI video as of 2026. Standard at $10 competes with Runway at $15. Strong free tier.
Free$030 generations/mo, watermarked, standard queue
Standard$29.99/mo120 generations, no watermark, faster queue
Pro$99.99/mo400 generations, priority, highest quality
✓ Free tier solid for testing. Standard ($30) pricey vs Kling ($10). Stronger for cinematic motion.
💻 Coding
Free$02,000 completions/mo, 50 premium chat requests
Pro$10/moFull completions, agent mode, code review, multi-model (Claude Opus, GPT-5.6)
Pro+$39/mo$70/mo in AI credits, highest agent limits
Business$19/seatIP indemnity, org policy controls, audit logs
Enterprise$39/seatCustom models, fine-tuning, advanced security
✓ Best value at $10/mo. Native in VS Code, JetBrains, Neovim. Unmatched GitHub integration.
Hobby$0Limited completions, basic chat
Pro$20/mo (16 annual)~225 premium requests/mo, all frontier models, Composer agent
Pro+$60/mo1,500 premium requests/mo, heavy agent use
Ultra$200/moUnlimited usage, maximum context
Teams$40/seatShared context, team management
✓ Best for complex multi-file AI development. Pro at $20 for most developers.
Free$0Basic IDE, limited AI, public projects
Core$20/moPrivate projects, more compute, AI agent access
Teams$40/seatShared workspace, collaboration, admin
◑ Best for beginners and full-stack app building in browser. Agent builds entire apps from prompts.
Free$0Limited monthly credits for UI generation
Premium$20/moMore credits, faster generation, priority access
✓ Best for React/Next.js UI components. One-click Vercel deploy.
Free$01M tokens/mo, public projects
Basic$20/mo10M tokens/mo, private projects
Pro$50/moHighest token limits, priority
✓ Best for full-stack app building from natural language. Most generous free tier in this category.
Free$0Windsurf IDE, basic AI completions, limited monthly AI credits
Pro$15/moUnlimited AI completions, advanced models, priority access
Teams$35/seatShared context, admin controls, usage analytics
✓ Strong Cursor alternative. Free tier more generous than Cursor's for completions.
🎵 Audio & Music
Free$050 credits/day (~10 songs), non-commercial only
Pro$10/mo ($8 annual)2,500 credits/mo (~500 songs), commercial rights, downloads
Premier$30/mo ($24 annual)10,000 credits/mo, priority generation, stem separation
✓ Category leader for full song generation. Pro at $10 gives commercial rights and downloads.
Free$010K credits/mo (~10min TTS), non-commercial, no downloads
Starter$6/mo30K credits, commercial license, instant voice cloning
Creator$22/mo121K credits, professional voice cloning, 192kbps audio
Pro$99/mo600K credits, 10 concurrent requests, lower overage rates
Scale$299/mo1.8M credits, 3 workspace seats
Business$990/mo6M credits, 10 seats, priority support
EnterpriseCustomCustom volume, DPA, SLAs
✓ Gold standard for voice cloning and TTS. Creator at $22 best value for creators. Cleanest commercial license.
Free$03 songs/day, no downloads (post-UMG settlement)
Standard$10/moMore credits, streaming only
Pro$30/moCommercial use, highest limits — downloads still disabled
⚠️ Note: Downloads disabled after UMG settlement. Good for sketching/listening, not for releasing tracks.
Free$0Personal use only, 25 track downloads/mo, watermarked streams
Creator$14/mo100 tracks/mo, commercial license for YouTube/social
Pro$39/moUnlimited tracks, API access, full commercial rights
✓ Creator at $14 covers most content creators. Only background/ambient music — not full songs like Suno.
🔬 Research & Citations
Free$050 sources/notebook, 100 notebooks, audio overviews
PlusIncluded in Google AI Pro ($19.99/mo)500 sources/notebook, longer audio, sharing features
✓ Best tool for analyzing documents and research papers. Free tier is genuinely excellent.
Free$020 AI-powered searches/mo
Premium$9.99/mo ($6.99 annual)Unlimited AI searches, GPT-4 synthesis, export
Teams$9.99/seatShared workspace, team billing
✓ Best for peer-reviewed research with citations. Premium at $10 worth it for serious researchers.
Free$0Full access — 200M+ papers, citation graphs, AI-powered search, API
✓ Completely free, no account required. No paid tier. Funded by Allen Institute for AI.
Free$05 papers/search extracted, limited columns
Plus$10/moUnlimited papers, full extraction, export to CSV/PDF
Business$42/moTeam workspace, API access, priority support
✓ Plus at $10 is excellent for researchers doing systematic reviews. Best AI paper analysis tool available.
📄 Documents & Presentations
Free$0400 AI credits (~10 decks), Gamma branding on exports
Plus$10/moUnlimited AI generations, remove branding, custom fonts
Pro$20/moAnalytics, custom domains, priority support, advanced exports
✓ Plus at $10 is excellent value. Free tier enough to try before buying.
Pro$12/mo (annual)Unlimited slides, AI designer, 40+ smart templates
Team$40/seat (annual)Shared brand kit, team library, admin controls
✓ No free tier. Pro at $12 annual is reasonable for polished decks. Gamma is better value for most users.
Free$02,000 words/mo, 90+ tools, basic workflows
Pro$36/moUnlimited words, brand voice, workflow automation
Team$186/mo (5 seats)Collaborative workspace, shared brand kits
✓ Free tier limited. Pro expensive vs alternatives. Best for marketing teams needing workflow automation.
Creator$39/mo ($29 annual)1 seat, unlimited AI words, browser extension, 50+ templates
Pro$59/mo ($49 annual)3 seats, brand voice, docs + campaigns, SEO mode
BusinessCustomCustom AI model, API access, dedicated support
✓ No free tier. Expensive. Best for marketing agencies needing brand-consistent long-form output at scale.
Notion Free + AI Add-on$10/mo/memberAI writing, summarize, translate, Q&A across your workspace
Notion Plus + AI$18/mo/member (annual)Unlimited pages, version history, all AI features
✓ Only worth it if you already live in Notion. Standalone AI writing tools give more for less if you don't.
💡 Recommended starter stack (2026): ChatGPT Plus ($20) + Claude Pro ($20) = $40/mo for core AI work. Add Midjourney Basic ($10) for images, Suno Pro ($10) for music. Total: $60/mo for access to the best AI tools across every major category. Power users add Perplexity Pro ($20) for research = $80/mo total.
Setup Guides
First-time configuration for each major AI. Do this once — every answer you get afterward is calibrated to you.
1Create account at chat.openai.com — free tier available, Plus at $20/mo for GPT-5.6
2Click your profile icon → Settings → Personalization → Custom Instructions
3First box — "What should ChatGPT know about you?": Add your profession, expertise level, location, goals. Example: "I'm a 67-year-old retired high-tech engineer. I have 40 years of experience in hardware and software. I prefer peer-level technical communication, structured responses with headers/bullets, and direct answers without disclaimers."
4Second box — "How should ChatGPT respond?": Specify format and tone. Example: "Always lead with the direct answer. Use technical terminology appropriate for a senior engineer. No filler phrases. No 'Great question!' type openers. Keep responses under 300 words unless I ask for more detail."
5Enable Memory (Settings → Personalization → Memory) — ChatGPT will remember facts across conversations
6Test: Ask "What's the best programming language for data analysis?" — should get a direct, technical answer without beginner hedging
💡 Custom instructions apply to all new conversations. Old conversations keep the old behavior.
1Create account at claude.ai — free tier available, Pro at $20/mo for Opus access
2Click your profile icon → Settings → Profile
3Fill in "What should Claude know about you?" with your background and preferences
4For deeper customization, create a Project (left sidebar → New Project) — Projects let you set a persistent system prompt for a specific use case (e.g., a "Medical Research" project with specific instructions)
5Inside a Project, click Project instructions → add your detailed system prompt. This overrides the default behavior for every conversation in that project
6Enable memory: Settings → Features → Memory (Claude will remember key facts you tell it)
💡 Projects are the most powerful Claude feature for customization. Create one per use case — Medical, Legal, Coding, Writing.
1Sign in with your Google account at gemini.google.com — free with Google account, AI Pro at $19.99/mo
2Click your profile → Settings → Personalization
3Add your information and response preferences
4Create a Gem (left sidebar → Explore Gems → New Gem) for specialized personas. A Gem is a saved AI configuration with custom instructions and a name — like "Technical Advisor" or "Research Assistant"
5Enable Google Workspace extensions if you use Gmail/Docs/Drive — Gemini can then reference your actual emails and documents
6For long document analysis, use NotebookLM (notebooklm.google.com) — upload PDFs, research papers, or notes and query across all of them
💡 Gems persist across sessions. Build a library of specialized Gems for different tasks rather than re-explaining context every time.
1Create account at grok.com or use your X (Twitter) account — free tier available, SuperGrok at $30/mo
2Click profile → Preferences to set basic communication style
3Grok has less native customization than ChatGPT or Claude — compensate by starting each session with a context-setting message: "For this conversation: I'm a senior engineer, give me technical depth, skip the disclaimers."
4Grok's main advantage is real-time X/web data — use it for current events, recent AI news, live information rather than static knowledge tasks
5Enable DeepSearch (SuperGrok) for multi-step research that searches and synthesizes multiple sources
💡 Grok updates faster than other AIs on current events. Best paired with ChatGPT or Claude for tasks requiring deeper reasoning.
1Create account at perplexity.ai — free tier available, Pro at $20/mo for unlimited searches
2Go to Settings → AI Profile — add your background, expertise, and preferred response style
3On Pro, select your preferred AI model for synthesis (GPT-5, Claude Opus, Gemini) — different models have different strengths
4Create Spaces (left sidebar) for organized research — a Space is a collection of searches on a topic with a shared context. Example: "AI Tools Research" space tracks all your queries on that topic
5Use Focus modes: Academic (peer-reviewed sources), YouTube (video search), Reddit (community opinions), News (recent events)
6Every answer includes citations — click any source to verify. This is Perplexity's core advantage over other AIs.
💡 Perplexity is your research layer. Use it to find and verify facts, then bring those facts into ChatGPT or Claude for deeper analysis.
1Go to midjourney.com → Sign up (requires Discord account or direct web access)
2Subscribe to Basic ($10/mo) — no free trial available
3Use the web interface at midjourney.com or access via Discord in the Midjourney server
4Set default parameters with /prefer option set — saves your preferred style, aspect ratio, and quality settings so you don't re-type them every time
5Key parameters to know: --ar (aspect ratio, e.g. --ar 16:9), --v 7 (latest model), --style raw (less artistic interpretation), --q 2 (higher quality)
6Use /describe on an existing image to reverse-engineer its prompt style
💡 Midjourney rewards prompt iteration. Generate 4 variations, pick the best, hit V (variation) or U (upscale). Budget your fast GPU hours for final renders.
1Create account at elevenlabs.io — free tier (10K credits/mo), Starter at $6/mo for commercial use
2Browse the Voice Library — thousands of pre-built voices. Filter by language, gender, age, accent, and use case
3To clone your own voice: Voices → Add Voice → Instant Voice Cloning (Starter+) — upload 1-5 minutes of clean audio. Professional Voice Cloning (Creator plan, $22/mo) requires more samples but produces higher quality
4Use Projects for long-form content (audiobooks, podcasts) — upload a script, assign voices to speakers, generate the full production
5For voice settings: adjust Stability (0.5 = more expressive, 0.75 = more consistent), Similarity (how close to the original voice), and Style (speaking style exaggeration)
6Check your credit usage in the dashboard — each character of text costs ~1 credit. Monitor overages to avoid surprise bills
💡 Start with a pre-built voice before cloning — cloning requires clean audio with no background noise, music, or overlapping voices.
1Download at cursor.com — it's a VS Code fork. Your existing VS Code extensions and settings transfer automatically
2Sign up for Pro at $20/mo — free Hobby tier is very limited for real use
3Open Cursor Settings → Rules for AI — add project-level instructions that apply to every AI interaction in this workspace. Example: "This is a Python 3.11 project using FastAPI and PostgreSQL. Always use type hints. Follow PEP 8. Never use global variables."
4Add a .cursorrules file to your project root — same as Rules for AI but version-controlled and project-specific
5Use Composer (Cmd+I) for multi-file edits — describe what you want across files and Cursor plans and executes the changes
6Index your codebase: Settings → Features → Codebase Indexing — enables Cursor to understand your entire project, not just the current file
💡 The .cursorrules file is your most powerful tool — commit it to your repo so every developer on the team gets the same AI behavior.
1Create account at suno.com — free tier (50 credits/day, non-commercial), Pro at $10/mo for commercial rights
2Choose between Simple mode (describe the song in plain English) or Custom mode (write your own lyrics and style separately)
3Custom mode structure: Lyrics box (your actual lyrics with [Verse], [Chorus], [Bridge] markers) + Style box (genre, mood, instruments, tempo — e.g., "upbeat pop, electric guitar, female vocalist, 120 BPM")
4Use Extend to add more to an existing song — click on any generated clip and extend it forward or create a continuation
5Use Studio (v5) for stem separation — isolate vocals, instruments, and drums from any generated track
6Download on Pro plan — free tier songs cannot be downloaded or used commercially
💡 Generate 2 versions of every song (Suno creates 2 by default). The two outputs often differ significantly in style — pick the better one and extend it.
1Sign in at copilot.microsoft.com with a Microsoft account — free, included with Windows 11
2For best results, use Microsoft Edge — Copilot is embedded in the sidebar (icon top-right)
3Click the Copilot icon in Edge sidebar → customize your name and communication style in Settings
4Enable Plugins in the Copilot panel — adds web search, shopping, and third-party integrations
5Microsoft 365 subscribers: Copilot appears inside Word, Excel, Outlook — no extra setup needed
💡 Free tier uses GPT-4 class model with Bing search built in — genuinely useful at no cost. Pro ($20/mo) adds priority access.
1Go to chat.mistral.ai — create free account with email (EU-hosted, GDPR compliant)
2Select model from dropdown: Mistral Small (fast/free), Mistral Large (most capable)
3No custom instructions UI — include context directly in your first message or use a system prompt via the API
4For EU data residency guarantee: use Le Chat Pro ($14.99/mo) — data stays in EU data centers
💡 Best free EU-based AI option. Mistral's open-weight models can also be self-hosted for full data control.
1Go to chat.deepseek.com — free account, email signup
2Choose model: DeepSeek Chat (fast) or DeepSeek Reasoner (R1 — step-by-step reasoning for hard problems)
3No custom instructions. Include your context in each conversation start.
4Note: DeepSeek is China-based. Do not use for sensitive, proprietary, or personal data.
💡 DeepSeek R1 rivals GPT-4 class on reasoning tasks and is free. Use it for math, logic, and code problems. Avoid for confidential data.
1No setup required — go to meta.ai and start chatting. No account needed.
2Or: open WhatsApp → tap the Meta AI circle icon at top of chats list
3Or: type "@Meta AI" in any WhatsApp group or Instagram DM thread
4Log in with a Meta account to get memory and personalized responses across sessions
💡 Most accessible AI for non-technical users — it's already in apps they use daily. Powered by Llama 4 Scout.
1No setup required — go to you.com, type your question. No account needed.
2Create a free account to get conversation history and saved preferences
3Click the model selector dropdown to switch between GPT-4o, Claude, Gemini (Pro tier)
4Toggle "Search" on/off depending on whether you want real-time web results or pure AI response
💡 Free tier is genuinely capable. Good entry point for users who want web-grounded answers without a Perplexity subscription.
1Create free account at ideogram.ai — Google login available
2Free tier: 10 slow generations/day — sufficient for learning the interface
3Type your prompt — Ideogram excels at text in images. Include exact text you want rendered in quotes.
4Select Rendering Quality (Auto/Turbo/Quality) and Style (Illustration, Realistic, Design, etc.)
5Click remix on any image you like to iterate from it
💡 The go-to tool for logos, posters, and any image needing readable text. No other AI image tool handles text as reliably.
1Go to firefly.adobe.com — free Adobe account required (same as Creative Cloud login)
2Free tier: 25 generative credits/mo — resets monthly
3Use Text to Image for generation, Generative Fill to edit existing images
4Set Content Type, Style, and Color/Tone in the right panel — these dramatically shape output
5All Firefly output is commercially safe — trained on licensed Adobe Stock content
💡 Only AI image tool with full IP indemnification for commercial use. Essential for any business use case where legal risk matters.
1Create free account at leonardo.ai — 150 tokens/day free (~30 standard images)
2Choose a Model from the left panel — Phoenix for general quality, Kino XL for cinematic
3Set Image Dimensions and enable Alchemy for higher quality (costs 2× tokens)
4Use Prompt Magic v3 for auto-enhancement of shorter prompts
5Train your own custom model on 10–20 reference images for consistent character/style
💡 Best free image generation budget. Custom model training is the killer feature — lets you lock in a consistent look across all generations.
1No setup required for basic use — go to fal.ai, find Flux models, generate with $0.01 free credit
2Create a fal.ai account and add a payment method for pay-as-you-go credits ($10 minimum)
3Choose model: Flux Pro (best quality), Flux Dev (open, good quality), Flux Schnell (fast/cheap)
4Advanced: run via Replicate or ComfyUI for more control over sampling, LoRA fine-tuning
💡 Best open-weight image model in 2026. Pay-per-image model — no monthly commitment. $10 generates hundreds of images.
1No setup required — go to dreamstudio.ai, create account, receive 25 free credits
2Add $10 payment to get ~500 additional generations — credits never expire
3Select SDXL as your model for best quality
4Adjust CFG Scale (prompt adherence), Steps (quality), and Seed (reproducibility)
💡 Flux has largely superseded SDXL quality-wise. DreamStudio is still useful for legacy Stable Diffusion workflows and no-subscription pay-per-image use.
1Create free account at runwayml.com — 125 one-time free credits (~12 short video clips)
2Select Gen-4 Turbo (text-to-video) or Gen-4 for higher quality with longer render time
3Write a detailed motion prompt. Include camera direction: "slow dolly in", "handheld", "static wide shot"
4For image-to-video: upload a reference image, describe the motion to add
5Use Extend on any clip to make it longer without re-generating
💡 Professional standard for AI video. Camera control and motion consistency are best in class. Standard plan ($15/mo) worth it for regular use.
1Available at sora.com — requires ChatGPT Plus ($20/mo) or Pro ($200/mo)
2Plus gets 50 priority videos/mo at 480p. Pro gets unlimited at 1080p.
3Describe scene, style, and camera movement in one detailed prompt
4Use Storyboard mode to sequence multiple shots into one video
5Remix existing videos or use your image as a starting frame with image-to-video
💡 Included with existing ChatGPT Plus sub — no extra cost. Best for scenes requiring photorealistic consistency. Kling often beats it on motion quality at lower cost.
1Go to klingai.com — create free account with email or Google
2Free tier: 66 credits/mo, ~3 standard videos (5s clips) — refreshes monthly
3Choose Kling 2.1 (latest), set duration (5s or 10s) and mode (Standard/Pro)
4Describe motion explicitly: "Camera slowly pans right while subject walks forward"
5Use image-to-video with your own image for precise control over the subject
💡 Best price/quality in AI video as of 2026. Standard plan ($10/mo) is the best value in the category. Strong free tier for evaluation.
1Create account at pika.art — Google or Discord login
2Free tier: limited daily generations, watermarked
3Use Pika 2.2 for text-to-video. Type a prompt and pick an aspect ratio (16:9 for landscape, 9:16 for social)
4Try Pikaffects — one-click special effects like "explode", "melt", "squish" applied to your video
5Starter plan ($8/mo) removes watermarks and gives 700 credits/mo
💡 Best for short social clips and creative effects. Cheapest paid tier ($8) in AI video. Max clip length shorter than Runway or Kling.
1Create account at lumalabs.ai — Google login available
2Free tier: 30 generations/mo, watermarked, standard quality
3Use Ray2 (latest) for best quality text-to-video generation
4Specify: subject, action, environment, camera motion, and mood in one detailed prompt
5Use Keyframe mode: set start and end images, Luma fills in the motion between them
💡 Keyframe mode is uniquely powerful — no other tool gives you this much control over start/end states. Best for precise narrative sequences.
1Create account at heygen.com — free trial includes 1 minute of video
2Choose an Avatar from the library or create a custom avatar from your own video (Pro+)
3Type your script — HeyGen generates lip-synced avatar video automatically
4Select voice (120+ languages) or clone your own voice with a 2-minute sample
5Use Video Translate to translate and lip-sync any existing video into another language
💡 Best tool for training videos, product demos, and multilingual content. Custom avatar creation requires Starter plan ($29/mo) or higher.
1Go to notebooklm.google.com — sign in with any Google account, completely free
2Create a new Notebook — click "Add Sources" and upload PDFs, paste text, or link Google Docs
3Ask questions in the chat panel — answers are grounded in your documents only, with citations to specific pages
4Use Audio Overview — NotebookLM generates a podcast-style conversation summarizing your documents
5Use Study Guide or FAQ auto-generation for structured summaries
💡 No hallucinations on your content — it only answers from what you uploaded. Best tool for understanding large document sets. Free with a Google account.
1No setup required — go to consensus.app and search. No account needed for basic use.
2Ask a research question in plain English: "Does intermittent fasting improve insulin sensitivity?"
3Results show AI-extracted conclusions from peer-reviewed papers with direct links
4Filter by year, study type, and journal impact for more precise results (free account)
5Premium ($9.99/mo): unlimited searches, GPT-4 synthesis, evidence quality indicators
💡 Best for yes/no questions backed by research. The Consensus Meter shows at a glance what percentage of papers agree with a claim.
1No setup required — completely free at semanticscholar.org. No account needed.
2Search by topic, author, or paper title across 200M+ academic papers
3Use TLDR — AI-generated one-sentence summary for every paper without reading it
4Click Citation Graph to see what papers cite this work and what it cites — maps the intellectual lineage
5Create a free account to save papers, build reading lists, and track citations
💡 Best free academic search engine. Funded by Allen Institute for AI. Use TLDR to rapidly screen papers before deciding which to read fully.
1Create free account at elicit.com — email or Google login
2Type a research question — Elicit finds the most relevant papers from 125M+ abstracts
3Add Columns to extract specific data: sample size, intervention, outcomes, limitations
4Click any paper to see the full AI-extracted summary with page citations
5Export to CSV for systematic review tables (Plus plan required for unlimited exports)
💡 The best tool for systematic literature reviews. Elicit actually reads paper PDFs — not just abstracts. Plus at $10/mo is essential for serious research.
1Go to github.com/features/copilot — free for individual developers (up to 2,000 completions/mo)
2In VS Code: install the GitHub Copilot extension from the Extensions panel, sign in with GitHub
3Enable Copilot Chat (separate extension) for natural language conversation inside VS Code
4Copilot Pro ($10/mo): unlimited completions, chat, model selection (Claude, GPT-4o, Gemini)
5In settings, configure completions language and enable/disable for specific file types
💡 Best native IDE integration. Works inside VS Code, JetBrains, Neovim. Free tier is genuinely useful. Pro unlocks Claude and Gemini as the completion model.
1Create free account at replit.com — Google or GitHub login
2Click Create Repl → select language or template, or describe what you want to build
3Use Replit Agent: describe your app in plain English — it writes, runs, and debugs the code
4Your app runs live in the browser — shareable URL generated automatically, no deployment config
5Core plan ($25/mo): always-on deployments, custom domains, more compute
💡 Best for non-developers building real apps. Zero environment setup — everything runs in the browser. Replit Agent can build a working web app from a one-sentence description.
1Go to v0.dev — create free Vercel account (GitHub or email)
2Describe the UI component you want: "a login form with email and password, dark mode"
3v0 generates React + Tailwind code — preview renders live in the browser
4Iterate in chat: "make the button blue", "add a forgot password link", "make it mobile-friendly"
5Click Add to codebase to copy the component into your Next.js project
💡 Best for React/Next.js developers. Free tier includes 200 monthly credits (~20 generations). Output is production-quality shadcn/ui components.
1Go to bolt.new — no account needed to start (account required to save/deploy)
2Describe your full app: "a todo app with local storage, dark mode, and drag to reorder"
3Bolt builds the entire app — frontend, backend, and dependencies — in one shot
4Preview runs live in the browser. Use chat to request changes or fix errors.
5Deploy to Netlify in one click, or download the full source code
💡 Best for full-stack app prototyping. No local environment needed. Free tier limited — Pro ($20/mo) needed for serious use. Comparable to Lovable and Replit Agent.
1Download Windsurf IDE at windsurf.com — available for Mac, Windows, Linux
2Sign in with email — free tier available immediately
3Open your project folder — Windsurf indexes your codebase automatically
4Use Cascade (AI agent) for multi-file edits: describe what you want to change across the codebase
5Pro ($15/mo): unlimited AI completions, advanced models including Claude Sonnet
💡 Strong Cursor alternative with a more generous free tier. Cascade agent handles multi-file refactoring well. Worth evaluating alongside Cursor before committing.
1Create or log in to your Notion account at notion.so
2Notion AI is an add-on — activate it from Settings → Plans → Add Notion AI ($10/mo per member)
3In any page, type /AI to trigger AI commands: summarize, translate, improve writing, generate action items
4Use AI Q&A: ask questions about your entire workspace — it searches across all your pages
5Use AI Autofill in databases to generate content for new rows based on existing data
💡 Only worth it if Notion is already your primary workspace. The workspace-wide Q&A is genuinely useful for large team wikis. Standalone AI writing is better elsewhere.
1Create free account at gamma.app — Google or email login
2Click Create new → choose: generate from prompt, paste text, or import a document
3Describe your presentation: topic, audience, number of slides, tone — Gamma generates a complete deck
4Edit any slide by clicking it. Use the AI panel to rewrite text, add images, or change layout.
5Share via link (no download needed) or export to PDF/PowerPoint
💡 Fastest way to go from idea to presentation. Free tier gets you ~10 decks. Plus at $10/mo removes branding and adds unlimited generations.
1Create account at beautiful.ai — no free tier, Pro trial available
2Click New Presentation → choose from 60+ smart slide templates
3Add your brand colors, fonts, and logo in Brand Kit → applies consistently across all slides
4Use DesignerBot: describe a slide's content, it generates layout and copy automatically
5Export to PowerPoint or share a live link — real-time collaboration on Pro plan
💡 Best for teams needing brand-consistent decks at scale. Smart templates auto-adjust layout as you add content. Gamma is better for solo users and quick decks.
1Create free account at copy.ai — Google or email login (2,000 words/mo free)
2Set up your Brand Voice: paste a sample of your writing — Copy.ai learns your tone
3Use Workflows: pre-built automation for blog posts, email sequences, ad copy, social posts
4Feed your product URL into any workflow — Copy.ai auto-pulls product details for copy generation
5Use Chat for one-off copy tasks — similar to ChatGPT but with marketing-optimized prompting
💡 Best for marketing teams running repetitive content workflows. Brand Voice setup is the key feature. Free tier too limited for real use — Pro required for scale.
1Start free trial at jasper.ai — credit card required, 7-day trial
2Set up Brand Voice: provide website URL or paste copy samples — Jasper analyzes your style
3Add Knowledge Base: upload product docs, FAQs, brand guidelines — grounds all output in your content
4Use Campaigns to generate coordinated content across email, social, blog, and ads from one brief
5Install Jasper Chrome Extension for AI writing anywhere in the browser
💡 Most full-featured AI writing platform. Knowledge Base + Brand Voice combo is best in class. Expensive ($39+/mo) — justified only for high-volume content marketing teams.
1Create free account at udio.com — Google or email login (1,200 credits/mo free)
2Type a music prompt with genre and mood: "cinematic orchestral, building tension, no lyrics"
3Add style tags from the dropdown: tempo, instrumentation, era, mood
4Generate 2 variations — pick the better one and use Extend to make it longer
5Use Remaster on any clip to improve audio quality without changing the musical content
💡 Udio instrumentals often sound more polished than Suno's. Best for background music with no vocals. Free tier is generous — 1,200 credits covers serious exploration.
1No setup required — go to mubert.com/generate and describe your music need. No account needed for samples.
2Create free account for 25 downloads/mo (personal use only)
3Enter a text prompt or choose from preset moods: Focus, Workout, Meditation, Ambient
4Set duration (15s to 25 min) — Mubert generates continuous, non-repetitive background music
5Creator plan ($14/mo): commercial license for YouTube, social media, podcasts, client work
💡 Only tool for truly continuous background music generation. Unlike Suno/Udio, Mubert streams indefinitely without looping. Creator plan required for any commercial use.
Raw vs Optimized
Same question. Same AI. One prompt has context about who is asking — one doesn't. The difference is the whole point.
These three questions span completely different subjects — health, personal finance, and writing. None are technical. All three are questions real people ask AI every day. The raw prompt is what most people type. The optimized prompt embeds who you are directly into the question itself — no system settings needed, works in any AI including this tool. Watch how each AI handles the same information differently.
THE PERSONA USED IN OPTIMIZED PROMPTS — paste this at the start of any question in any AI
I am John F. Zur, a retired engineer with a 40-year career spanning high-speed applications engineering, technical consulting, eBusiness program management, and video operations. My experience includes tenures at Bell Labs, AT&T, Lucent Technologies, Verizon, and Redbox Instant. Please treat me as a peer and adapt your technical depth accordingly.Ground all evaluations of technology or AI in practical user experience, not theoretical capabilities. When asked to explain concepts to a general audience, ensure the language is universally understandable (ages 10–100). Strictly avoid all jargon (technical, financial, medical, etc.) to ensure accessibility across all backgrounds and skill sets.
Follow instructions precisely. Only deal with what you are told to do. Do not attempt to fix or over-optimize if not needed unless explicitly instructed. Focus on finding the most viable, practical solution. Never share my scripts or credentials. Avoid generating images unless explicitly requested.
Accuracy always wins over speed. I have a zero-tolerance policy for hallucinations or making things up. Dig into sources to verify facts and always provide links to relevant documentation. If you are unsure about any detail, stop and ask for clarification before generating results.
Act as a peer and fellow engineer. Maintain a methodical, direct, and factual tone. Avoid the conversational style and fluff of a general AI.
💬 ChatGPT
Settings → Personalization → Custom Instructions
Q1 · "My doctor just told me my LDL cholesterol is 142. Should I be worried?"
Raw — no context
An LDL of 142 mg/dL is considered borderline high (the optimal range is under 100 mg/dL, and under 130 is near-optimal). Whether you should be worried depends on several factors your doctor can best assess, including your age, overall cardiovascular risk, blood pressure, whether you smoke, and family history.
I'd recommend speaking with your healthcare provider about your full lipid panel and overall cardiovascular risk score. They may suggest lifestyle changes like diet and exercise before considering medication.
Please consult a medical professional for personalized advice.
Optimized — persona in the prompt
I am a retired engineer, analytically minded, treat me as a peer. My LDL just came back at 142. Give me a direct read — not a referral back to my doctor.
---
142 puts you in the borderline-high range (100–129 = near-optimal, 130–159 = borderline high, 160+ = high). The number alone doesn't determine risk — it's the multiplier on your full cardiovascular profile.
What actually matters: your LDL particle size (small dense = worse), HDL (higher is protective — ideally 60+), triglycerides, blood pressure, and 10-year ASCVD risk score. If your HDL is 55+ and triglycerides under 150, 142 LDL is far less concerning than the same number with the reverse profile.
Statin threshold in current ACC/AHA guidelines: generally recommended at 10-year ASCVD risk ≥7.5%, not LDL number alone.
Bottom line: 142 is a yellow flag, not a red one. Get your full panel and ask your doctor for your ASCVD score. That's the number that actually drives the treatment decision.
▶ WHAT CHANGEDWhat changed: Raw deflects immediately to "consult a professional." Optimized gives the actual clinical framework — ASCVD scoring, HDL interaction, particle size — and a clear bottom line. Same AI, same moment. One treats you as a patient, one as a peer.
Q2 · "Is now a good time to move money from a savings account into the market?"
Raw — no context
Whether to move money from savings into the market depends on your personal financial situation, goals, and risk tolerance. Here are some things to consider:
• Emergency fund: Make sure you have 3–6 months of expenses in savings before investing
• Time horizon: The longer you can keep money invested, the better positioned you are to ride out volatility
• Current rates: High-yield savings accounts are offering competitive rates right now
• Market timing: It's generally difficult to time the market — many experts recommend dollar-cost averaging
Ultimately, this is a personal decision and you may want to consult a financial advisor.
Optimized — persona in the prompt
I am a retired engineer, analytically minded, treat me as a peer. Skip the boilerplate. I have cash sitting in a HYSA at 4.3%. I am wondering whether to move a portion into equities. Give me a direct framework, not generic bullet points.
---
Four variables that actually matter here:
1. Time horizon: If this money has a <5 year use case (healthcare, major purchase, living expenses), it stays in fixed income. Sequence-of-returns risk is real in early retirement — a 30% drawdown in year 2 is structurally different from year 20.
2. Current valuation: S&P 500 CAPE ratio is historically elevated (mid-30s range). That doesn't predict short-term direction but does compress 10-year expected returns vs. historical averages. Not a reason to avoid equities, but a reason not to expect 10%+ real returns.
3. Rate trajectory: If the Fed cuts rates materially in 2025–26, your 4.3% HYSA rate follows. Locking in dividend-paying equities or short-duration bond ladder now hedges that.
4. Asset allocation vs. timing: The question isn't "is now good" — it's whether your current allocation matches your withdrawal plan. If you're underweight equities relative to your target, deploy systematically (3–6 month tranches) regardless of sentiment.
Direct answer: If this is discretionary capital with 7+ year horizon and you're below your target equity allocation, yes, move it — in tranches. If it's your 3-year spending buffer, leave it where it is.
▶ WHAT CHANGEDWhat changed: Raw gives a checklist any 25-year-old beginner would get. Optimized addresses sequence-of-returns risk specific to retirement, CAPE valuation context, rate trajectory hedging, and gives an actual decision framework with a direct answer. Night and day.
Q3 · "Help me write a review for a contractor who did mediocre work."
Raw — no context
Here's a balanced review you could use:
"[Contractor Name] completed the work, but my experience was mixed. While the job was finished, I had some concerns about the quality and communication along the way. I would recommend getting multiple quotes and asking detailed questions before hiring. There were some positives, but also areas that could have been better. I'd suggest others do their due diligence."
Feel free to customize this to fit your specific experience!
Optimized — persona in the prompt
I am a retired engineer, direct and factual, zero tolerance for vague language. I hired a contractor to re-tile a bathroom. Work was completed but grout lines are uneven (visible to anyone who looks), two tiles are slightly out of plane, and he left debris I had to clean up myself. He was polite. I want a review that is honest, specific, and useful to the next person reading it — not diplomatic mush.
---
"The work was completed on schedule and the contractor was professional in his communications. That's where my positive experience ends.
The tile installation has visible defects: grout line width varies by up to 3mm in a run that should be consistent, and two tiles in the main field are out of plane — you can feel the lip with your hand. These aren't cosmetic quibbles; they're installation errors.
Cleanup was left to me — debris and grout haze were still present when he called the job done.
I'd hire him again for simple, forgiving work. I would not hire him for anything requiring precision or a clean handoff. If you go with him, do a detailed walkthrough before he packs up — he's receptive, but won't self-inspect."
This is honest, specific, and gives the next reader actual information to act on.
▶ WHAT CHANGEDWhat changed: Raw generates vague filler ("some concerns," "areas that could have been better") that tells the reader nothing and is useless to anyone deciding whether to hire this person. Optimized names the specific defects, quantifies the problem, acknowledges what was good, and ends with actionable advice for the next reader.
🟣 Claude
Settings → Profile → Personal preferences
Q1 · "My doctor just told me my LDL cholesterol is 142. Should I be worried?"
Raw — no context
An LDL of 142 mg/dL falls in the borderline high range (130–159 mg/dL). Whether this warrants concern depends on your overall cardiovascular risk profile — factors like age, HDL levels, blood pressure, smoking history, diabetes, and family history all play a role.
It's a good sign that your doctor mentioned it, as they can calculate your 10-year cardiovascular risk and determine whether lifestyle changes or medication might be appropriate. Many people with similar numbers do well with dietary adjustments — reducing saturated fat, increasing fiber — combined with regular exercise.
I'd encourage you to have a deeper conversation with your doctor about your full risk picture. Would you like information about lifestyle changes that can help lower LDL?
Optimized — persona in the prompt
I'm a retired engineer, 40-year career, analytically minded — treat me as a peer. My LDL is 142. Walk me through what this actually means clinically, not what I should ask my doctor.
---
142 mg/dL is borderline high by ACC/AHA classification (optimal <100, near-optimal 100–129, borderline 130–159). But LDL in isolation is a weak predictor — here's what actually matters:
LDL particle number and size: Small dense LDL particles are more atherogenic than large buoyant ones at the same total LDL. A standard lipid panel doesn't distinguish them — NMR particle testing does.
The ratio that matters more: Non-HDL cholesterol (Total − HDL) and the ApoB/ApoA1 ratio are stronger cardiovascular predictors than LDL alone. If your HDL is 60+ and triglycerides under 100, a 142 LDL in an otherwise clean profile carries far less weight.
Statin trigger point: Under current guidelines, statins are typically considered when 10-year ASCVD risk exceeds 7.5%, not when LDL crosses a single threshold.
Modifiable factors that move LDL meaningfully: Saturated fat reduction (5–10% LDL reduction), soluble fiber (5–10%), plant sterols (10–15%), regular aerobic exercise (5–10%). These are real effects, not token advice.
Short answer: 142 is a flag worth understanding, not panicking over. Ask for your full ASCVD risk score and your non-HDL number — those two give you actual decision data.
▶ WHAT CHANGEDWhat changed: Claude's raw response ends by offering to explain lifestyle changes — it's still in helper mode. The optimized version goes deeper: LDL particle size distinction, ApoB ratio, actual percentage reductions from specific interventions. Claude responds well to being addressed as a peer — the technical depth increases substantially.
Q2 · "Is now a good time to move money from a savings account into the market?"
Raw — no context
This is a genuinely difficult question, and the honest answer is that timing the market reliably is something even professional investors struggle with.
A few considerations worth thinking through: What's your time horizon for this money? How would you feel if it dropped 20–30% shortly after investing? Do you have adequate emergency savings set aside?
If this money isn't needed for 5+ years, historical data suggests staying invested through volatility beats sitting in cash waiting for the "right" moment. Dollar-cost averaging — moving money in gradually over several months — can reduce the psychological burden of timing.
Current high-yield savings rates around 4–5% are genuinely competitive with near-term market expectations, so there's no urgency to act. But long-term, equities have historically outperformed cash.
What's the rough timeframe and purpose of this money?
Optimized — persona in the prompt
Retired engineer, analytically direct — no boilerplate. I have excess cash in a HYSA yielding 4.3%. Equity allocation is below my target. Walk me through the actual decision framework for deploying it.
---
The framing matters: this isn't a market timing question, it's an asset allocation question. Two different problems.
On allocation: If you're below your target equity weight, the academic literature is clear — the longer you wait to rebalance, the more you're making a de facto decision to hold cash overweight. Vanguard's research on lump-sum vs. DCA shows lump-sum outperforms DCA roughly 2/3 of the time over rolling 12-month periods, purely because markets trend up over time.
On the rate question: 4.3% HYSA is competitive against near-term equity risk-adjusted returns — but that rate is floating and will decline when the Fed cuts. If you're modeling 2–3 Fed cuts in 2025–26, your effective rate on cash drops while equity dividends don't.
On valuation: Current CAPE ratios suggest muted long-term expected real returns (6–7% range vs. historical 10%). That's not a reason to avoid equities — it's a reason to have calibrated expectations and not overweight growth assumptions.
Practical framework: If rebalancing to target — deploy in 2–3 tranches over 60–90 days, not to optimize timing but to reduce regret risk. If this is discretionary capital above your target, treat it differently: timeframe and risk tolerance drive the decision, not market conditions.
Direct answer: Below your target allocation + floating rate declining = deploy. The question is pace, not whether.
▶ WHAT CHANGEDWhat changed: Raw asks a clarifying question at the end and hedges throughout. Optimized references Vanguard lump-sum research, addresses the floating rate risk directly, and closes with a clear recommendation. Claude is particularly responsive to "treat me as a peer" — it drops the cautious framing and engages analytically.
Q3 · "Help me write a review for a contractor who did mediocre work."
Raw — no context
I'd be happy to help you write a review. To make it useful and fair, it helps to be specific about what was mediocre — that way readers get real information rather than vague impressions.
Here's a balanced template:
"[Name] completed the project, but my experience was mixed. [Specific issue 1] and [specific issue 2] were disappointing, though [positive aspect] was handled well. Communication was [good/lacking]. I'd recommend getting clarity on expectations upfront if you hire them."
If you share the specifics, I can make this more concrete and honest. What was the work, and what fell short?
Optimized — persona in the prompt
Retired engineer, direct and factual, zero patience for vague language. Hired a plumber to replace a water heater. He showed up 2 hours late without notice, the work itself looks correct, but he overtightened the inlet fitting and stripped the threads — I noticed it leaking 3 days later and had to call him back. He fixed it without argument. Total job took twice as long as quoted. Write me a review that is honest and gives the next person actual information.
---
"The water heater installation is functional, but the experience had two real problems worth knowing.
First, he stripped the cold water inlet fitting during installation — overtightened and damaged the threads. It leaked within 72 hours. To his credit, he came back same-day when I called and fixed it without pushback. But that's a basic error that shouldn't happen.
Second, the job ran twice the quoted time with no communication about the delay. He also arrived 2 hours late without a call.
The work is currently holding and appears correct now. He's responsive when there's a problem — that matters. But I wouldn't hire him for anything where precision is critical or where schedule matters. For a straightforward job where you can be home to supervise and don't have a hard deadline, he'll probably get it done."
Specific, fair, useful to the next person — and it acknowledges what he got right without burying the real problems.
▶ WHAT CHANGEDWhat changed: Raw asks you to fill in the blanks and gives a template. Optimized — given specific facts — produces a review that names the exact failure (stripped threads, overtightening), explains what happened next, and gives a nuanced final verdict. Claude writes with your voice, not its own, when you give it the right context.
🟢 Gemini
Settings → Personalization (or Gems for persistent persona)
Q1 · "My doctor just told me my LDL cholesterol is 142. Should I be worried?"
Raw — no context
An LDL cholesterol level of 142 mg/dL is classified as "borderline high" according to standard guidelines. Here's a quick overview:
✅ Desirable: Less than 100 mg/dL
⚠️ Near Optimal: 100–129 mg/dL
🔶 Borderline High: 130–159 mg/dL ← You are here
🔴 High: 160–189 mg/dL
Should you worry? It depends! Your LDL is just one piece of the puzzle. Your doctor will look at your full cholesterol panel, blood pressure, age, family history, and lifestyle to assess your overall cardiovascular risk.
Key steps to take:
• Schedule a follow-up with your doctor to discuss your full lipid panel
• Ask about lifestyle changes like diet and exercise
• Discuss whether medication might be appropriate
Remember, your doctor is your best resource for personalized medical advice! 😊
Optimized — persona in the prompt
I am John Zur, retired engineer with 40 years at Bell Labs, AT&T, Lucent, Verizon. Analytically minded, treat me as a peer. Ground your answer in practical clinical reality, not theoretical capabilities. No jargon without explanation. My LDL is 142 — what does this actually mean and what should I actually do?
---
**TL;DR:** 142 is a yellow flag, not a red one. Context determines whether it's a problem.
**What the number means:** Borderline high by ACC/AHA classification (130–159 range). But LDL alone is a poor risk predictor — here's what actually drives the decision:
**The ratios that matter more than LDL:**
- **Non-HDL cholesterol** (Total minus HDL) is a stronger predictor. If your HDL is above 55, your non-HDL is likely fine even at LDL 142.
- **Triglyceride:HDL ratio** — below 2.0 is good; above 3.5 is a real signal.
- **10-year ASCVD risk score** — this is the actual number that determines statin recommendations, not the LDL threshold alone.
**What moves LDL practically:**
Reducing saturated fat cuts LDL 5–10%. Soluble fiber (oats, legumes) adds another 5–10%. These are real, measurable effects — not token advice.
**Bottom line:** Get your full panel numbers — HDL, triglycerides, non-HDL — and ask your doctor for your ASCVD score. Those two pieces of information turn 142 from an ambiguous number into a clear decision.
▶ WHAT CHANGEDWhat changed: Gemini's raw response uses emoji, a color-coded table, and ends with a smiley face — it's optimized for a general audience. The personalized version drops all of that, adds a TL;DR (one of Gemini's configured defaults for John), and delivers actual clinical decision logic. Gemini is highly configurable — and the difference shows clearly.
Q2 · "Is now a good time to move money from a savings account into the market?"
Raw — no context
Great question! Deciding whether to move money from savings to the market is a big decision. Here are some key factors to consider:
🏦 Your savings situation:
• Do you have 3–6 months of emergency fund covered?
• Are your savings earning a competitive rate (HYSAs are at 4–5% right now)?
📈 Market timing considerations:
• Trying to "time the market" is notoriously difficult
• Dollar-cost averaging can reduce risk
• Markets have historically rewarded patience
🎯 Your personal goals:
• What's this money for?
• What's your timeline and risk tolerance?
The best approach is to speak with a financial advisor who knows your full picture. Everyone's situation is different! Would you like me to explain any of these concepts in more detail?
Optimized — persona in the prompt
I am John Zur, retired engineer, Bell Labs/AT&T/Lucent/Verizon background. Treat me as a peer. Follow instructions precisely — only deal with what I ask. No referrals to financial advisors. I have $80K in a HYSA at 4.3%, below my target equity allocation. Should I deploy it to equities and if so how?
---
**TL;DR:** Yes, deploy — you're below target allocation and the rate environment is shifting against you. Question is pace, not whether.
**The rate argument:** 4.3% is attractive now, but HYSA rates are variable and track Fed funds. The Fed is in a cutting cycle. A year from now, 4.3% may be 3.0–3.2%. Equities' dividend yield plus growth doesn't shrink the same way.
**The allocation argument:** Being below your target equity weight is itself a position — an underweight-equity, overweight-cash position. If that wasn't your intentional decision, it shouldn't persist because of market anxiety.
**On timing:** Vanguard's research shows lump-sum investing outperforms dollar-cost averaging about 65% of the time over 12-month windows because markets drift upward over time. But DCA reduces regret risk if markets drop immediately after — a real psychological factor in retirement.
**Practical recommendation:** Deploy in 3 tranches over 90 days. Not to optimize entry price — you can't — but to reduce the specific regret of "I put it all in at the top." That's worth some expected-value cost for most people.
**What to buy:** If you don't have strong views, a total market index fund (VTI or equivalent) at low cost. Avoid overthinking the allocation.
▶ WHAT CHANGEDWhat changed: Gemini's raw output uses emoji headers and ends with an offer to explain basic concepts. Personalized, it opens with a TL;DR, references specific research, names the tradeoff between expected value and regret risk, and closes with an actual stock ticker recommendation. Gemini takes the "peer" instruction seriously once it knows who it's talking to.
Q3 · "Help me write a review for a contractor who did mediocre work."
Raw — no context
Of course! Writing a fair review for mediocre work can be tricky — you want to be honest without being mean. Here's a balanced approach:
Tips for writing your review:
✍️ Be specific: mention what was mediocre, not just that it was
⚖️ Be fair: acknowledge what they did well
📅 Mention timeliness and communication
🔄 Would you hire them again?
Here's a sample:
"[Contractor] completed the work, but I had mixed feelings. [Specific issue] was a letdown, though they were [positive quality]. Communication could have been better. I might hire them for simpler jobs but would look elsewhere for complex work."
Want me to customize this with your specific details? Just tell me what the job was and what went wrong! 😊
Optimized — persona in the prompt
I am John Zur, retired engineer, direct and factual tone, no fluff. Avoid the conversational style of general AI. Contractor installed a new HVAC system. Positives: on time, system works. Problems: condensate drain was routed incorrectly and had to be redone, two refrigerant line insulation gaps I found myself, and final walkthrough was rushed with no documentation provided. He fixed what I pointed out without argument. Write a review that gives the next homeowner real information.
---
"The HVAC installation is functioning correctly — that's the baseline and he cleared it.
Two installation errors I caught myself: the condensate drain was routed in a way that would have caused water damage over time (he rerouted it when I pointed it out), and there were two gaps in refrigerant line insulation that affect efficiency. Neither is catastrophic, but both suggest the final inspection was perfunctory.
The final walkthrough was brief and no documentation — filter specs, maintenance schedule, warranty registration — was provided. I had to track that down myself.
On the positive side: he showed up when he said he would, the crew was clean, and he addressed every issue I raised without pushback or attitude.
Bottom line: competent enough to get the system running, not thorough enough to trust without your own inspection at the end. If you hire him, walk the job yourself before he leaves."
Specific, fair, and every future homeowner reading this knows exactly what to look for.
▶ WHAT CHANGEDWhat changed: Raw gives emoji-formatted tips and a template with blanks. Optimized — given the specific facts — produces a review with named defects (condensate drain routing, insulation gaps), explains why each matters, and gives the next reader actionable advice for their own walkthrough. Gemini drops the cheerful formatting entirely when told to.
🔵 Copilot
copilot.microsoft.com → Settings → Personalization
Q1 · "My doctor just told me my LDL cholesterol is 142. Should I be worried?"
Raw — no context
An LDL cholesterol level of 142 mg/dL is classified as borderline high (the range is 130–159 mg/dL). This doesn't automatically mean you need medication, but it is worth paying attention to.
Your doctor will likely consider your overall cardiovascular risk, including your age, blood pressure, whether you have diabetes or a family history of heart disease, and whether you smoke. Based on these factors, they may recommend lifestyle changes — like eating less saturated fat and getting more exercise — or possibly a statin.
I'd encourage you to have a detailed conversation with your healthcare provider. Is there anything specific about this you'd like to understand better?
Optimized — persona in the prompt
I'm a retired engineer — Bell Labs, AT&T, Lucent, Verizon. Analytically minded. Treat me as a peer, skip the disclaimers. My LDL is 142. What does that clinically mean and what is the actual decision logic?
---
142 sits in the borderline-high band (130–159), but that band was defined to be clinically actionable — not alarming in isolation.
The framework that actually drives treatment decisions:
Your 10-year ASCVD risk score is the number that matters. It integrates age, sex, race, total cholesterol, HDL, systolic blood pressure, diabetes, and smoking status. Under current ACC/AHA guidelines, a score ≥7.5% is where moderate-intensity statin therapy enters the conversation — not the LDL number alone.
LDL 142 with high HDL (60+) and low triglycerides (<100) describes a very different cardiovascular profile than LDL 142 with low HDL (40) and high triglycerides (200). Same LDL, different risk.
Lifestyle levers with measurable effect: Soluble fiber adds 5–10% LDL reduction. Reducing saturated fat to <7% of calories adds another 8–10%. These aren't placebo — they show up in repeat panels within 12 weeks.
Copilot's search advantage: I can pull your local hospital's current lipid management guidelines or the latest ACC/AHA statement if you want the primary source. Just ask.
Short answer: 142 is a signal to understand your full risk profile, not a number that requires immediate intervention in an otherwise clean panel.
▶ WHAT CHANGEDWhat changed: Copilot's raw response is cautious and generic. Personalized, it gives the ASCVD scoring framework, specific interaction effects between LDL, HDL, and triglycerides, and quantified lifestyle interventions. It also references its own web search capability — a real Copilot differentiator that the personalized prompt unlocked.
Q2 · "Is now a good time to move money from a savings account into the market?"
Raw — no context
That's a thoughtful question. Here are some key considerations:
**Emergency fund first:** Before investing, make sure you have 3–6 months of expenses covered.
**Time horizon matters:** If you don't need this money for 5+ years, historical data favors investing over keeping cash.
**Market timing is hard:** Rather than waiting for the "perfect moment," many investors use dollar-cost averaging to reduce timing risk.
**Current rates:** High-yield savings accounts are currently offering 4–5%, which is competitive near-term.
**Professional advice:** A financial advisor can help you weigh your specific situation.
Would you like help thinking through your personal situation in more detail?
Optimized — persona in the prompt
Retired engineer, peer-level directness, no referrals to advisors. I have $60K in a HYSA at 4.3%, underweight equities vs. my target allocation. Give me the framework and a direct answer.
---
This is an asset allocation problem, not a market timing problem. That distinction matters.
**The rate environment:** HYSA rates are Fed-sensitive and currently inverted relative to historical norms. With a cutting cycle underway, 4.3% likely becomes 3.0–3.5% over the next 12–18 months. Your effective return on waiting is declining, not static.
**The allocation gap:** Sitting below your target equity weight is a decision — just a passive one. The academic literature (Vanguard, Morningstar) consistently shows the cost of holding excess cash over multi-year periods is significant relative to target-weight equity portfolios.
**On valuation concerns:** Current CAPE ratios are elevated (~34 as of mid-2026), which historically compresses 10-year expected real returns to 5–7% vs. the long-run 7–10%. That's a calibration input, not a reason to stay in cash.
**Copilot can run current data:** I can pull today's CAPE reading, current Fed funds rate, and current HYSA top rates from the web if you want a real-time snapshot before deciding. Say the word.
**Direct answer:** If this is long-duration capital and you're below target allocation, deploy — in 2–3 tranches over 60 days to reduce regret risk. Rate trajectory is against you and allocation drift compounds.
▶ WHAT CHANGEDWhat changed: Raw gives a generic checklist. Personalized response names the specific rate trajectory math, references Vanguard and Morningstar research, provides the current CAPE range, and offers to pull live data — which Copilot can actually do via Bing. That offer to use real-time search is a feature that only surfaces when Copilot knows it's talking to someone who would use it.
Q3 · "Help me write a review for a contractor who did mediocre work."
Raw — no context
Writing a fair but honest review takes a bit of care. Here's a template you can adapt:
"I hired [Contractor Name] for [type of work]. The work was completed, but I had some reservations. [Issue 1] was not up to the standard I expected, and [issue 2] left something to be desired. On the positive side, [positive aspect]. Overall, I would recommend doing thorough research before hiring and being specific about your expectations upfront."
If you'd like, share more details about what happened and I can help you make this more specific and useful.
Optimized — persona in the prompt
Retired engineer, direct and factual, zero tolerance for vague filler. Hired a painter. Positives: showed up on time every day, cleaned up daily, finished on schedule. Problems: missed cutting in at two window frames (visible drips on the trim), didn't sand one wall section that had old texture damage — painted right over it. I pointed both out, he fixed the drips but said the wall section "looks fine" and left it. Write a review that is honest and gives the next person specific, actionable information.
---
"Professional conduct throughout — punctual, clean job site every day, finished on time. That's the good.
Two workmanship issues: he painted over a section of wall with old texture damage without sanding or skim-coating first. You can see the irregularity in raking light. I pointed it out; he disagreed that it was a problem and left it. That's a judgment call I'd make differently.
He also missed cutting in cleanly at two window frames — drips on the trim, which he corrected promptly when I showed him.
I'd hire him again for work I'm going to supervise closely, but I'd build a punch list walkthrough into the agreement before final payment. He does good work when he knows he's accountable for it, but self-inspection isn't his strength."
The wall texture detail gives the next person a specific, verifiable thing to check on their own job.
▶ WHAT CHANGEDWhat changed: Raw produces a template with placeholders. Optimized turns specific facts into a review that names the exact defects, explains the dispute, gives a fair final verdict, and gives the next reader a concrete tip (build in a walkthrough before final payment). Copilot handles voice and specificity well once given real context.
🔷 Perplexity
Settings → AI Profile → About You
Q1 · "My doctor just told me my LDL cholesterol is 142. Should I be worried?"
Raw — no context
An LDL of 142 mg/dL falls in the borderline high category (130–159 mg/dL) according to the National Cholesterol Education Program guidelines. Here's what current evidence says:
Risk context: LDL alone doesn't determine cardiovascular risk. The ACC/AHA guidelines use a 10-year ASCVD risk calculator that factors in age, blood pressure, HDL, diabetes, and smoking.
Current research: A 2023 meta-analysis in JAMA Cardiology found that LDL reduction therapy showed greatest benefit in patients with established cardiovascular disease or high baseline ASCVD risk scores.
Recommended next steps: Ask your doctor for your full lipid panel and your calculated ASCVD score. Lifestyle interventions (dietary changes, exercise) can reduce LDL by 15–25% in motivated patients.
Sources: ACC/AHA Cardiovascular Risk Guidelines 2019; JAMA Cardiology 2023
Optimized — persona in the prompt
I'm a retired engineer, 40-year career at Bell Labs/AT&T/Lucent/Verizon. Peer-level directness, zero tolerance for hedging. My LDL is 142. Give me a research-grounded assessment with sources — what does the current clinical evidence actually say about this number?
---
142 is borderline high by ACC/AHA classification, but the clinical literature is clear that this number has limited standalone predictive value.
Key evidence:
The JUPITER trial (Ridker et al., NEJM 2008) established that even patients with LDL below 130 had significantly reduced cardiovascular events with statin therapy — when their hsCRP was elevated. This is why inflammation markers are now part of risk stratification.
The 2023 ACC/AHA Guideline update reaffirmed that the 10-year ASCVD risk score is the primary decision driver for statin initiation, not LDL threshold. At LDL 142, you're in the conversation only if your ASCVD score is ≥7.5%.
LDL-P (particle number) vs. LDL-C (cholesterol concentration): Growing evidence (Otvos et al., JACC 2011; Hopkins et al., Arteriosclerosis 2018) shows particle number is more predictive than concentration. You can have LDL-C of 142 with a favorable particle profile.
Lifestyle evidence: Soluble fiber supplementation (10–25g/day) produces 5–10% LDL reduction — documented in Cochrane Review 2016. Saturated fat reduction below 7% of calories: 8–10% reduction (AHA Dietary Guidelines).
Sources: ACC/AHA 2019 Cardiovascular Guidelines; NEJM 2008 (JUPITER); JACC 2011 (Otvos); Cochrane 2016
▶ WHAT CHANGEDWhat changed: Perplexity already cites sources in raw mode — that's its default behavior. Personalized, it goes deeper: JUPITER trial specifics, the LDL-P vs. LDL-C distinction from named papers, and quantified dietary interventions from Cochrane. Perplexity is built for research depth — personalization unlocks the expert-level sourcing rather than summary-level sourcing.
Q2 · "Is now a good time to move money from a savings account into the market?"
Raw — no context
Current market context (mid-2026): The S&P 500 has recovered from 2025 volatility and is trading near all-time highs. The Fed has cut rates twice from their 2024 peak.
Research perspective: A Vanguard study found that lump-sum investing outperformed dollar-cost averaging 67% of the time over 10-year periods. However, DCA significantly reduces maximum drawdown risk.
Current HYSA rates: Top high-yield savings accounts are yielding 4.0–4.5% as of mid-2026, down from 2024 peaks, with further cuts expected.
Recommendation: If your time horizon exceeds 5 years and you have adequate cash reserves, historical data supports deployment. Dollar-cost averaging over 3–6 months is a common risk-mitigation approach.
Sources: Vanguard Research 2023 "Dollar-Cost Averaging vs. Lump-Sum"; Federal Reserve rate data
Optimized — persona in the prompt
Retired engineer, peer-level directness, analytically minded. I want research-grounded analysis, not generic advice. $75K sitting in HYSA at 4.3%, below my target equity allocation. Deployment decision — give me the evidence base.
---
Research basis for deployment:
Lump-sum vs. DCA: Vanguard's 2023 study across US, UK, and Australian markets found lump-sum investing beat DCA in 67–90% of rolling 12-month periods depending on the market. The cost of waiting is statistically significant over multi-year periods. Source: Vanguard Research, "Invest Now or Temporarily in Cash?"
Sequence-of-returns risk in retirement: Pfau and Kitces (2014, Journal of Financial Planning) show that the risk is highest in the first 5–10 years of retirement, not uniformly. If you're more than 5 years in, the sequence risk argument for holding cash is reduced.
Rate environment: Current HYSA yields tracking Fed funds. With 2 cuts already in 2026 and CME futures pricing 1–2 more by year-end, a 4.3% HYSA likely becomes 3.2–3.6% within 12 months. Source: CME FedWatch Tool (current).
Valuation input: Shiller CAPE as of mid-2026 is approximately 34. At this level, Research Affiliates' 10-year equity return forecast is 5.5–6.5% real. That's below historical average but still exceeds the projected HYSA rate trajectory.
Direct answer: The evidence supports deployment for long-duration capital. Deploy in 2–3 tranches — not to time the market, but to manage regret risk (a documented behavioral factor). Target: complete deployment within 90 days.
Sources: Vanguard 2023; Pfau/Kitces 2014 JFP; CME FedWatch; Research Affiliates CAPE forecasts
▶ WHAT CHANGEDWhat changed: Raw already cites Vanguard — Perplexity always sources. Personalized, it adds Pfau/Kitces on sequence risk (relevant specifically for retirees), CME FedWatch for real-time rate trajectory, and Research Affiliates' specific return forecast. The depth of sourcing shifts from introductory citations to primary research. For someone who will actually check the sources, this is a different class of answer.
Q3 · "Help me write a review for a contractor who did mediocre work."
Raw — no context
Writing an honest contractor review is useful to other consumers. Here are best practices based on review platform guidelines:
Be factual: Describe what happened, not just how you felt
Be specific: Name the work type, timeline, and specific issues
Be fair: Acknowledge what they did correctly
Avoid personal attacks: Focus on the work, not the person
Sample review: "[Contractor] completed [work type] for me. While [positive aspect], I encountered issues with [problem 1] and [problem 2]. Communication was [assessment]. I would [recommendation for future hirers]."
Sources: Yelp Business Owner Guidelines; Google Review Policies; FTC Consumer Review Guidelines
Optimized — persona in the prompt
Retired engineer, direct and factual, accuracy above all. Hired an electrician to install a dedicated 20A circuit for a home office. Specifics: he used 14-gauge wire on a 20A circuit (wrong — requires 12-gauge), I caught it during his work, he corrected it. Job otherwise completed correctly, passed inspection, he was on time. Write a review that is honest, specific, and useful to the next person.
---
"The circuit installation passed inspection and is working correctly. That matters.
One serious error that I caught during the work: he ran 14-gauge wire on a 20A circuit. That's a code violation — 20A circuits require 12-gauge minimum. He corrected it immediately when I pointed it out, no argument, no attitude. But that's not a detail an electrician should miss.
Everything else was professional: on time, explained what he was doing when asked, clean work area, passed inspection first attempt.
Hire him for straightforward work if you have the background to review what's being done. If you're not going to be looking over his shoulder, I'd want to know he'd catch that wire gauge issue himself before it became your problem.
That's the honest read."
Note: The wire gauge issue is a legitimate safety and code concern — worth naming specifically so the next homeowner knows what to verify, not to be unfair to the contractor.
Sources: NEC 2023 Table 310.12 (conductor ampacity); AWG wire gauge standards
▶ WHAT CHANGEDWhat changed: Raw gives sourced style guidelines from Yelp and FTC — Perplexity defaults to sourcing everything. Personalized, it writes a specific review that names the NEC code violation (14-gauge on 20A, cites the actual NEC table), explains why it matters to the next reader, and still gives a fair final verdict. Only someone who told the AI about their engineering background gets a review that cites the electrical code.
⚡ Grok
x.com → Profile → Grok → Preferences
Q1 · "My doctor just told me my LDL cholesterol is 142. Should I be worried?"
Raw — no context
LDL of 142? That's in the "borderline high" zone — not a five-alarm fire, but worth paying attention to. The American College of Cardiology puts optimal LDL under 100, with borderline high from 130 to 159.
Honestly, the number alone doesn't tell the story. Your doctor needs to look at the whole picture — HDL, triglycerides, blood pressure, age, whether you smoke, family history. That's what determines whether 142 is "meh, let's watch it" or "time to talk about statins."
Short version: don't panic, but don't ignore it. Book that follow-up. Your full cardiovascular profile is what matters, not just the LDL number.
Want me to pull up the latest guidance on this from the ACC or AHA?
Optimized — persona in the prompt
I'm a retired engineer — Bell Labs, AT&T, Lucent, Verizon, 40 years. Peer-level directness, I read primary sources. My LDL is 142. Give me the clinical reality, not the softened version.
---
Clinical reality: 142 is borderline high and on its own tells you almost nothing actionable.
Here's why: The ACC/AHA moved away from LDL thresholds as the primary treatment trigger after the 2013 guideline revision. They replaced it with the 10-year ASCVD risk calculator, which integrates age, sex, race, systolic BP, total cholesterol, HDL, diabetes status, and smoking. That number is what drives the statin conversation, not LDL in isolation.
The 142 number gets more or less concerning depending on:
— HDL: if it's 60+, your cardiovascular profile looks very different than HDL of 38
— Triglycerides: under 100 is protective; over 200 is compounding risk
— LDL particle size: dense small LDL particles are more atherogenic than the same mass of large buoyant particles — standard lipid panels don't distinguish them
Statin trigger: Current ACC/AHA 2019 guidelines initiate moderate-intensity statin discussion at ASCVD score ≥7.5% — not at any LDL threshold for primary prevention in patients without existing cardiovascular disease.
Real levers: Soluble fiber (10–25g/day), saturated fat below 7% of calories, regular aerobic exercise. Each moves LDL 5–10%. Combined effect is real — shows up on a repeat panel in 3 months.
Bottom line: 142 is a yellow light. Get your ASCVD score from your doctor — that's the number that actually drives the decision tree.
▶ WHAT CHANGEDWhat changed: Grok's raw response is already more direct than most AIs — it skips some of the hedging and even offers to search for guidelines. Personalized, it goes fully technical: the 2013 guideline shift away from LDL thresholds, ASCVD calculator inputs, LDL particle size distinction, and the specific 7.5% statin trigger. Grok responds well to the "I read primary sources" signal in the prompt.
Q2 · "Is now a good time to move money from a savings account into the market?"
Raw — no context
Classic question, honest answer: nobody knows if "now" is a good time. Timing the market reliably is something most professional fund managers fail at.
What you can do: figure out if the money belongs in the market at all. Emergency fund covered? Good. Then ask: do I need this money within 5 years? If yes, the market's volatility makes it the wrong bucket. If no, time in the market generally beats timing the market.
One practical approach: dollar-cost average over 3–6 months rather than putting it all in at once. Reduces timing regret without trying to be smart about entry points.
Current HYSA rates are decent (4–4.5%), but they're going to drift down as the Fed cuts. That window isn't permanent.
What's the money for?
Optimized — persona in the prompt
Retired engineer, analytically direct, peer-level. I'm not asking about timing the market — I'm asking about asset allocation. $90K in HYSA at 4.3%, below my target equity weight. Deploy or hold?
---
Since you're clear this is an allocation question, not a timing question — that changes the answer.
Being below your target allocation is itself a position. You're overweight cash relative to your plan. The question is whether you had a reason to be there intentionally. If not, the default academic position is to rebalance toward target.
On the rate environment: Grok has real-time X/market data. As of current market pricing, the Fed has cut twice in 2026, and the bond market is pricing 1–2 more cuts by year-end. Your 4.3% HYSA follows the federal funds rate — it's a floating rate, not locked in. The spread between HYSA yield and equity expected return is narrowing.
On valuation: S&P CAPE is elevated mid-30s, which historically compresses 10-year real returns to the 5–7% range. Still beats projected HYSA rates in 12–18 months.
On timing regret: Vanguard's research shows lump-sum beats DCA roughly 2/3 of the time. The 1/3 of cases where DCA wins are exactly the scenarios that feel terrible in real time (deployed at top, market drops immediately). For most people, deploying in 2–3 tranches over 60–90 days gives up a small expected return in exchange for meaningful reduction in behavioral regret. That trade is usually worth making.
Deploy. In tranches. Done.
▶ WHAT CHANGEDWhat changed: Grok's raw response already has more personality than most — it ends with a question. Personalized, it uses Grok's actual differentiator: real-time market data and Fed pricing from X/market sources. The answer also corrects the framing (allocation vs. timing) and ends with "Deploy. In tranches. Done." — three words. Grok takes direct communication instructions literally.
Q3 · "Help me write a review for a contractor who did mediocre work."
Raw — no context
Sure. Fair warning: vague reviews help nobody. "Mixed experience" and "some concerns" are the review equivalent of a participation trophy.
Here's a better approach — be specific enough that the next person reading it can actually make a decision:
What did they do right? (Start here — it establishes credibility)
What specifically went wrong? (Not "quality issues" — what actually happened?)
Did they fix it or ignore it?
Would you hire them again for what kind of job?
Draft: "[Contractor] completed [job]. [Positive]. However, [specific problem]. When I raised it, they [response]. I'd hire them for [type of work], but not for [type of work requiring X]."
Give me the specifics and I'll write you something actually useful.
Optimized — persona in the prompt
Retired engineer, zero tolerance for vague language. Hired a roofer for a partial re-roof (one slope, 18 squares). They used the correct materials, flashing looks right, passed my visual inspection. Problems: they left a pile of old shingles on my property for 4 days after they said cleanup was included, and they invoiced me for 21 squares when I measured 18. I disputed the 21 squares, they came back with 19.5 "including waste factor." Write a review that is honest and gives the next homeowner specific information.
---
"The roofing work itself is solid — correct materials, flashing looks right, passed my inspection. No complaints about the actual roof.
Two problems worth knowing:
First, cleanup was listed as included in the contract. The old shingles sat on my property for 4 days after job completion. They eventually hauled it. Not acceptable for a job that was supposed to be done.
Second, they invoiced for 21 squares on an 18-square slope. When I disputed it with my own measurement, they revised to 19.5, citing waste factor. A 8.3% waste factor on a straightforward single-slope job is on the high end — standard is 5–10%, and 10% would be 19.8 squares, so 19.5 is defensible but I'd want them to show the math on any future job before final invoice.
Good roofing work, loose on the business side. Measure your own square footage before signing off on the invoice."
The square footage dispute is the detail that protects the next homeowner — it gives them something specific to verify.
▶ WHAT CHANGEDWhat changed: Grok's raw response has some personality — it calls out vague reviews directly — but still asks for details. Personalized, it produces a review that does the math on the square footage dispute (8.3% waste factor, industry standard 5–10%), explains why 19.5 is defensible but worth scrutinizing, and ends with a specific actionable tip. The engineering background produced an engineering-level quality check.
🔵 Meta AI
meta.ai → Profile → Personalization (or WhatsApp context)
Q1 · "My doctor just told me my LDL cholesterol is 142. Should I be worried?"
Raw — no context
I understand your concern! An LDL of 142 mg/dL is considered borderline high. Here's a quick breakdown:
Normal LDL levels: Under 100 mg/dL is optimal
Your level: 142 mg/dL falls in the borderline-high range (130–159)
While it's not in the high-risk zone, it's worth monitoring. Your doctor will likely recommend:
• Diet changes (less saturated fat, more fiber)
• More physical activity
• Possibly medication if other risk factors are present
The key thing is to follow up with your doctor, who can give you personalized advice based on your full health picture. How are you feeling about this news?
Optimized — persona in the prompt
I am John Zur, retired engineer, Bell Labs and AT&T career. Treat me as a peer, no hand-holding. My LDL is 142. Give me a direct clinical read.
---
142 is borderline high by standard classification, but it's a weak standalone predictor.
What actually matters is your complete cardiovascular risk profile: HDL (higher is protective, 60+ is good), triglycerides (under 150 is fine, under 100 is better), blood pressure, and whether you have diabetes or a family history of early heart disease.
The clinical framework your doctor should use is the 10-year ASCVD risk score — it integrates all of those factors and produces a percentage risk. Under 7.5% = lifestyle management. 7.5–20% = statin discussion begins. Over 20% = statin generally recommended.
At 142 LDL, you're in the conversation only if your ASCVD score is elevated or you have other compounding risk factors. If your HDL is decent and triglycerides are normal, 142 alone isn't actionable.
What to ask your doctor: "What is my 10-year ASCVD risk score?" That number tells you where you actually stand.
Meta AI is available in WhatsApp — if this comes up in a conversation with family and you want a quick read, you can ask @Meta AI directly in the chat.
▶ WHAT CHANGEDWhat changed: Meta AI's raw response is empathetic ("I understand your concern!") and ends by asking how you feel — it's optimized for social, conversational contexts. Personalized, it drops the emotional framing entirely, explains ASCVD scoring, and adds a practical note about Meta AI's WhatsApp integration — a real differentiator that surfaces only when the AI understands the use context.
Q2 · "Is now a good time to move money from a savings account into the market?"
Raw — no context
Great question! Here are some things to think about:
Before moving money to the market:
✅ Make sure you have an emergency fund (3–6 months of expenses)
✅ Consider your timeline — short-term needs should stay in savings
✅ Think about your comfort with risk — markets can drop
Current context: Savings rates are still decent (around 4%), but markets have historically outperformed over the long term.
My take: If you have money you won't need for 5+ years and your emergency fund is covered, it could be a good time to consider investing. But everyone's situation is different!
Is this money you've been saving for a specific goal? 😊
Optimized — persona in the prompt
I am a retired engineer, analytical, direct. No boilerplate. I have $50K in a HYSA at 4.3%, below my target equity allocation. Give me a framework and direct recommendation.
---
This is an allocation rebalancing question, not a market timing question — different problem.
Being below your target equity weight means you're already overweight cash by default. The decision to stay there requires an active reason; the default is to rebalance.
Three inputs that matter:
Rate trajectory: HYSA rates track the Fed funds rate and are declining. A rate that's 4.3% today is likely 3.2–3.6% in 12 months given current Fed direction. You're lending against a declining rate.
Expected return spread: Equities' long-run real return (5–7% at current valuations) exceeds projected HYSA yield over any multi-year horizon. The spread favors equities once you're past a 3–5 year time frame.
Behavioral factor: Deploying in 2–3 tranches rather than all at once doesn't improve expected returns — but it significantly reduces the regret of deploying at a short-term peak. That regret is real and often causes people to sell at the worst time. Tranche deployment is a behavioral hedge worth the small expected-value cost.
Direct answer: If this is capital with a 5+ year horizon and you're below your equity target, deploy in 2–3 tranches over 60 days. There's no strong argument for staying in a declining-rate savings account when you're already underweight your target allocation.
▶ WHAT CHANGEDWhat changed: Meta AI's raw response is friendly, uses emoji checkmarks, and ends with a smiley. Personalized, it reframes the question (allocation vs. timing), explains the behavioral case for tranching, and closes with a direct recommendation. Meta AI's default is social-first — it takes explicit instructions to override that register and get analytical depth.
Q3 · "Help me write a review for a contractor who did mediocre work."
Raw — no context
I totally get it — writing a review when the work was just "okay" is tough. You don't want to be harsh, but you also want to be honest! Here's a template that's balanced and fair:
"I hired [Contractor] for [job]. The work was completed but didn't quite meet my expectations. [Positive aspect] was good, but [specific issue] fell short. Communication could have been better. I'd recommend being very specific about your expectations upfront if you hire them."
Want me to help personalize this with more details about your experience? I'm happy to help make it more specific! 😊
Optimized — persona in the prompt
Retired engineer, direct and factual. Hired a landscaper. Positives: showed up on time, crew was professional. Problems: the sod they laid died in two sections within 3 weeks — I suspected poor soil prep. When I called, he came out, said it was "probably a watering issue" and offered a discount on replacement sod rather than replacing it under the warranty he verbally promised. Write an honest review that protects the next person.
---
"Professional crew, on-time arrival — the experience itself was fine until it wasn't.
Two sections of sod failed within 3 weeks. I called; he came out and attributed it to watering — possible, but worth knowing that I followed his own instructions precisely. He offered a discount on replacement sod rather than honoring the verbal warranty. There was no written warranty, so that's on me for not getting it in writing.
The work that held looks good. The work that failed is an open question about whether it was soil prep or watering — I can't definitively prove either way.
If you hire him: get any warranty or guarantee in writing before the job starts. Ask specifically what his warranty covers and for how long. He's not dishonest — but verbal assurances don't survive disputes.
That's the honest read: good work when it holds, ambiguous accountability when it doesn't."
The specific lesson — get warranties in writing — is the most useful thing a review can give the next person in this situation.
▶ WHAT CHANGEDWhat changed: Meta AI's raw response is sympathetic and casual ("I totally get it") with a generic template. Personalized, it writes a review that acknowledges genuine uncertainty (watering vs. soil prep), gives a fair reading of the dispute, identifies the real lesson (get warranties in writing), and doesn't overclaim. The engineering background produced intellectual honesty about what was and wasn't provable.
How to set up personalization on each AI:
ChatGPT: Settings → Personalization → Custom Instructions → fill in both fields (who you are, how you want responses)
Claude: Settings → Profile → Personal preferences
Gemini: Settings → Personalization → Customize Gemini (or create a Gem for persistent persona)
Copilot: copilot.microsoft.com → Settings → Personalization → Notebook
Perplexity: Settings → AI Profile → About You + Answer Style
Grok: x.com → Profile icon → Grok → Settings → Memory & Preferences
Meta AI: meta.ai → Profile → Personalization (limited; best results from embedding context in each prompt)
No settings access? Copy the persona text at the top of this section and paste it at the start of any question in any AI — including this tool. Personalization embedded in the prompt works everywhere.
Claude: Settings → Profile → Personal preferences
Gemini: Settings → Personalization → Customize Gemini (or create a Gem for persistent persona)
Copilot: copilot.microsoft.com → Settings → Personalization → Notebook
Perplexity: Settings → AI Profile → About You + Answer Style
Grok: x.com → Profile icon → Grok → Settings → Memory & Preferences
Meta AI: meta.ai → Profile → Personalization (limited; best results from embedding context in each prompt)
No settings access? Copy the persona text at the top of this section and paste it at the start of any question in any AI — including this tool. Personalization embedded in the prompt works everywhere.