Build a Synthetic User Feedback Agent with Claude Code

If you want faster product feedback without waiting on live interviews, Claude Code can simulate believable target personas, have them browse your site or app, and tell you where the friction or skepticism shows up.

This is not a replacement for real user research. It is a fast spot-checking workflow. In the demo, it was useful for catching things like hidden pricing skepticism, missing discount clarity, and little conversion details that are easy to miss when you are too close to the product.

The GitHub Repo

Install this synthetic-user-feedback skill into my current Claude Code project.

Repo:
https://github.com/travisseh/synthetic-user-feedback

Read the repo README and the skill instructions first.

Then:
- copy the skill into .claude/skills/
- show me where the personas template lives
- prefer Claude Chrome if available; otherwise set up Playwright MCP
- tell me the exact first command or prompt to run against my site

Do not assume personas already exist. If they do not, ask me to define 2-4 target customer personas or draft them from my product and market.

This one ships as a Claude Code skill, not a full plugin. That is the right primitive here because the workflow is contextual, browser-driven, and best invoked when you want simulated feedback on a specific page or task.

What's In The Repo

The repo is intentionally simple. It gives Claude the browsing behavior, the reporting format, and a reusable persona-template file so you can run synthetic feedback without rebuilding the prompt every time.

synthetic-user-feedback/
├── SKILL.md
└── references/
    └── personas-template.md

What This Skill Actually Does

• It has 2 to 4 personas browse your site or app in character instead of reading every word like a robot.
• It gives you stream-of-consciousness commentary while they click, skim, get confused, and react.
• It ends with a structured synthesis report: verdicts, pain points, unanswered questions, and top recommendations.
• It saves the output into `docs/` if that folder exists, or creates `user-research/` if it does not.

Why This Is Useful

The biggest value is speed. In the walkthrough, the point was not that this magically replaces talking to real users. The point was that it gives you a fast first-pass read that can catch things you would otherwise find much later.

That makes it especially useful for pricing pages, homepage messaging, signup flows, feature discovery, and other places where one unclear detail can create skepticism or cause someone to bounce.

Chrome Beats Playwright For Some Flows

The skill works with browser access either way, but the demo recommends running it with Claude Chrome when possible. That gives Claude more room to click through logged-in experiences and richer browsing flows. Playwright MCP still works well as the fallback.

The Prompt To Run Synthetic Feedback

Once the skill is installed and browser access is ready, use a prompt like this:

Run synthetic user feedback on [MY URL].

I want feedback on:
- [what I want tested]

What You Get Back

• A realistic narrative of what each persona notices, misses, questions, and reacts to.
• A clear verdict per persona: buy, bounce, needs more info, completed task, or gave up.
• Cross-persona patterns, not just isolated complaints.
• A saved research artifact you can come back to later instead of losing the feedback in chat history.

Best Use Cases

• Pricing-page reviews before launch
• Homepage and landing-page message checks
• Signup and onboarding friction reviews
• Logged-in feature discovery and task-completion tests
• Quick pre-launch validation when you want a fast reality check before talking to real users

Build It In Layers

If you are new to Claude Code, read the skills guide first, then the MCP servers guide. This workflow depends on both: the skill provides the behavior, and browser access gives Claude the ability to actually test the experience.