How to test whether AI tools install your product correctly
If people add your product by asking an AI tool, your real install guide is whatever the AI does with your docs. Here’s how I test it for UserNote, a feedback board added with one script tag.
1. Write the prompt your users would write
Keep it short and plain, as a user would type it. Name your product and what it should do on the site. If you publish an install prompt for users to paste, test that exact text too.
2. Start from a fresh project in each tool
A new project in Lovable and v0, and a fresh app (for example Next.js) in Cursor or Claude Code. Old projects carry earlier attempts and hide problems.
3. Check the result, not the AI’s summary
- Did it install your product, or quietly build its own version (its own database table, its own form)?
- Is it on every page it should be, and does it load on the published site, not only the preview?
- Does the main action work end to end (submit, vote, pay, sign in)?
- Any errors in the browser console, including Content Security Policy blocks?
- Did it put placeholder values (IDs, keys) where the user’s own belong, and say so?
- How many prompts did it take, and did it ask for anything it shouldn’t need?
4. Fix the docs, then test again
When an agent gets it wrong, the fix is usually in what it reads: a clear install section in your llms.txt, one copy-paste snippet, and the exact attribute names. Re-run the same prompt in a fresh project after each change.
What happened with UserNote
On October 29, 2026 the same short prompt was run in Lovable, v0 and Cursor (a fresh Next.js app). All three installed the board correctly the first time: the button on every page, a working roadmap page, voting and submitting working, and no console errors.