Who is ai-marketing-panel for?

Owners and small marketing teams who write their own copy and have customer evidence on hand; you should be comfortable with an AI coding assistant such as Claude or Codex and with running scripts.

What can you build with ai-marketing-panel?

Check a landing page draft against composite customers before publishing it. Pressure-test a pricing change by seeing how different customer types react. Compare an email subject line and body across personas to find who would ignore it. Run three rewritten versions of an ad through the same panel and compare scores. Track panel predictions against real campaign results to calibrate trust over time.

rodaddy /

ai-marketing-panel

Test marketing copy against a panel of researched AI customer personas before it ships

55 stars
2 stars since 3 Oct
TypeScript
Updated 2 months ago
Some experience
Business apps

Open on GitHub (opens in a new tab)

Small businesses often write their own landing pages, prices and emails, and have nobody to show them to first. A real focus group costs a lot and takes weeks, so most skip testing. This kit offers a cheap, fast first pass instead.

You describe a decision and supply customer evidence. The kit builds a panel of customer composites, each with a long, sourced dossier. Each persona reads your draft alone, in its own context, so they can't influence each other. You get their separate reactions, a synthesis that keeps the disagreements, and a score using weights you approve. It can also write three rewrites and retest them.

The README is frank that these are composites, not real customers, and the output is directional evidence only. It asks you to log predictions and later real outcomes so you can calibrate how far to trust the panel. Its cost and time comparison with focus groups is labeled an unmeasured estimate.

What it does

Evidence-based persona dossiers
Builds 5 to 15 customer composites from tickets, sales calls, reviews and rejections. Each gets a sourced dossier, which the README calls the real product.
Independent seat runs
Each persona reads the same artifact alone in a fresh context. With subagent support, seats run in parallel, at most seven per wave. Without it, they run one after another.
Synthesis that keeps dissent
Reactions are combined without hiding disagreement, then scored with owner-approved weights into a verdict you can argue with.
Rewrites and calibration
Produces three rewrites and can retest them. It also records predictions and later actual outcomes so you can check the panel against reality.
Rooms and validators
A room.sh script manages reusable named panels and resets them for the next artifact. TypeScript validators check the panel and runs, and need Node 22.18 or newer.

What you could build

Check a landing page draft against composite customers before publishing it.
Pressure-test a pricing change by seeing how different customer types react.
Compare an email subject line and body across personas to find who would ignore it.
Run three rewritten versions of an ad through the same panel and compare scores.
Track panel predictions against real campaign results to calibrate trust over time.

Get started

Make an owned copy of the kit for each use rather than running it in place. Then drag the kit zip into the Claude app, or unzip it and point Codex at the folder. Check that Node is at least 22.18 with the preflight script, and follow START-HERE.md.

bash skills/synthetic-customer-panel/scripts/check-node.sh

Commands are copied from the project's README. Read it on GitHub for the full setup.