How to Verify an AI Money-Making Claim in 10 Minutes
Step 1 — assign every number a tier
| Tier | What qualifies | What it proves |
|---|---|---|
| verified 🟩🟩🟩 | Primary documents (filings, contracts, platform dashboards you can audit) — or data you measured yourself | The number itself, for the period shown — nothing about your ability to repeat it |
| reported 🟨🟨⬛ | Named, independent press or institutions with a date | That the event (a funding round, a policy) occurred — not that users profit |
| self-reported 🟥⬛⬛ | Everything that traces only to whoever profits from the claim — screenshots included | Only that the seller says so |
Step 2 — check the five red flags
Each one marks a claim unverifiable, not necessarily false — which is precisely why the burden of proof stays on the seller:
- Screenshots as proof. Trivially faked, never auditable, and even genuine ones show a moment, not a track record.
- Revenue without the denominator. "$10k/month" means nothing without the traffic, ad spend, or hours that produced it.
- Backtests sold as track records. Try enough variants on the same history and a spectacular in-sample result is guaranteed — the graded evidence.
- The seller earns from the tool, not the method. The picks-and-shovels tell: their revenue arrives whether or not yours does.
- Testimonials where a ledger should be. A method that works produces a ledger; a method that sells produces testimonials.
Step 3 — the decisive test
Ask for the falsifiable live ledger: entries timestamped before outcomes, all results graded including losses, misses never deleted, methodology fixed in advance, net of costs. It is cheap to publish and impossible to fake retroactively — which is why its absence is informative. A seller who meets it deserves your attention; a seller who offers anything else instead has answered your question.
The same test binds us: every verdict on our ledger carries a dated evidence tier and a written flip condition, and wrong calls stay published. Method without self-application is marketing.
Apply it in two minutes — free, no sign-up
The Claim Grader walks any claim through the tiers, forces a written flip condition, and outputs a paste-anywhere verdict card. One self-contained file; works offline; renamed copies encouraged.
Open the grader → Need a formal claim audit? →FAQ
Trace every number in the pitch to its source and assign a tier: verified (primary documents or data you can measure yourself), reported (named independent press), or self-reported (traces only to whoever profits). Then apply the one decisive test: does the seller publish a falsifiable live ledger — results timestamped before outcomes, losses never deleted, methodology fixed in advance? As of 2026, in the categories our ledger tracks, no retail AI-income vendor has published one.
Five recur across the graded record: screenshots as proof (unverifiable and trivially faked); revenue quoted without the denominator (traffic, ad spend, or hours that produced it); backtests sold as track records (trying enough variants always produces a spectacular in-sample result); the seller earning from the tool rather than the method (the picks-and-shovels tell); and testimonials in place of a ledger. Each flag marks a claim as unverifiable, not necessarily false — which is exactly why the burden of proof stays on the seller.
No — and this is the most common confusion in the AI gold rush. Funding proves investor appetite for the picks-and-shovels business: selling tools to would-be earners, a revenue stream that arrives whether or not users profit. User returns are a separate, almost never audited claim. Our ledger grades them separately for exactly that reason.
Yes, and it is free with no sign-up: the Claim Grader is a single self-contained file that walks any claim through the tiers, forces a written flip condition, and outputs a paste-anywhere verdict card plus a Claim Ledger Protocol entry. Save the file and it works offline; adapted, renamed copies are encouraged.