| name | synthcheck |
| description | Test offers, ad copy, lead forms, landing pages, pricing, names, and positioning on simulated buyer cohorts before spending money. |
| version | 1.0.0-portable |
| author | Hermes Agent |
| license | MIT |
SynthCheck
Use this when the user wants to test ad copy, offers, landing pages, lead forms, pricing, positioning, or campaign ideas before spending money.
Non-negotiable method
Do not ask simulated people for a numeric rating directly. Ask for reactions in words first, then map those reactions to a score using Semantic Similarity Rating.
Workflow
- Define the decision and ship/kill threshold before testing.
- Build 4-6 commercially relevant cohorts. Include a skeptic cohort and premium cohort.
- Ground cohorts in real competitors, real prices, reviews, comments, sales-call objections, or CRM context when available.
- Force heterogeneity: income, urgency, skepticism, purchase fatigue, life situation, and pain intensity.
- Show each person one option when testing price. Rotate variants for copy, offers, names, or landing pages.
- Collect natural-language reactions: what stops them, what they believe, what they remember, what they would click, and what they would ignore.
- Score reactions against anchors. Report distributions, not only averages.
- Apply an optimism haircut and recommend the smallest real-money smoke test.
Paid ads outputs
- Winning hook or angle
- Objection map
- Believability risks
- CTA friction
- Lead-form friction
- Offer confusion
- Verbatim lines worth stealing
- Recommended real-money test
Guardrails
SynthCheck is directional. It ranks options and surfaces objections, but it does not replace live spend data.
