The skill
Install once. Bring a decision and your real audience data and get ranked hypotheses, confidence flags, a bias check, and a validation plan.
What's inside
Three files, one job: hypotheses to test, not answers to trust
Built for growth, product and research teams, not data scientists.
Install guide
A two-minute setup, plus what to bring for grounding so the pass reflects your buyers instead of a vanilla model.
Plan B prompt
No skills access? Paste the included prompt into any Claude chat with your data. Same rules, same guardrails, zero setup.
How it works
Three steps, from a decision to hypotheses with a validation gate.
Step 1State the decision and bring your data
Say what you are trying to decide and paste your real signal, interviews, tickets, win-loss notes, survey results, or CRM segments.
Step 2Get hypotheses, not findings
The skill runs a grounded persona pass and returns ranked hypotheses with confidence flags and the assumption each rests on, plus a bias and homogenization check.
Step 3Validate before you spend
Every decision-driving hypothesis comes with the real-world check that must confirm it first. The synthetic pass tells you what to test, reality tells you what is true.
It refuses red-line uses like predicting conversion, sizing a market, or picking a test winner, and it never presents synthetic output as evidence. That is the feature, not a limitation.
How to use synthetic personas without being misled
Synthetic personas can simulate your buyers on demand, but their answers are fluent and confident whether or not they are right. This free Claude skill uses them the safe way, as a hypothesis engine grounded in your real data, with a bias check and a validation gate in front of every decision, so a plausible answer never gets mistaken for proof.
Why synthetic answers need a rule
An AI panelist can hand you a neatly worded reason your offer will land even when a real buyer would disagree. Models also collapse toward stereotypes even when asked for diversity, and they reflect stated attitudes rather than revealed behavior. Useful for exploration, dangerous as evidence. The rule that keeps them safe is simple. treat every output as a hypothesis to test, never a finding.
What the skill produces
It grounds a persona pass in the real audience data you supply, then returns a grounding summary, a ranked list of hypotheses each with a confidence flag and its underlying assumption, a bias and homogenization check, and a validation plan. Every hypothesis that could drive spend comes with the specific real-world check and the metric that would confirm or kill it.
Green light and red light
Green-light uses are exploration, generating angles, surfacing objections, pre-screening a long list of variants, stress-testing an ICP assumption before spend. Red-light uses are anything you would cite as proof, predicting conversion or retention, declaring a test winner, sizing a market, or modeling a niche audience a general model has not truly seen. The skill refuses red-line questions and reframes them toward the real research they need.
Why now
Synthetic audiences have crossed from novelty to standard kit, with new B2B launches on top of a market research industry worth around 140 billion dollars. The upside is real and so is the failure mode. The full background, including what the research says on both sides, is in the source article, AI Can Now Simulate Your Customers on Demand. Treat the Answers as Hypotheses, Not Evidence.
Who it is for
Growth and performance marketers pre-testing positioning, angles, and objections before spend, product and lifecycle teams exploring concepts, and anyone who wants the speed of synthetic research without mistaking it for the real thing.