Evidence before hype
Every product makes claims. We test them.
Independent, claim-by-claim reviews of AI and software tools — with the inputs, outputs and real costs on the page, so you can check our work before you pay.
Claim check · example
Test #[000]
[AI detector] says:
“Detects AI-written text with 99% accuracy.”
Flags AI text it’s shown
Verified
Leaves human writing alone
Doesn’t hold
“99% accuracy” overall
Overstated
Tested [DATE]
[N] samples
Paid plan
Hands-on tests
Real tasks on the plan buyers actually pick.
Claim by claim
One verdict per claim. No blended star scores.
Dated and re-tested
Every change recorded in a public log.
Verdicts not for sale
Vendors never see or edit a review.
Latest verdicts
Best AI Website Builder (2026): Speed Claims, Real Costs and Performance Data Compared
Wix, Hostinger, Squarespace, Durable, Framer and 10Web given the same business brief: time to a usable site, fixes needed, SEO basics and…
Cursor vs GitHub Copilot (2026): Do AI Coding Assistants Really Make You Faster?
Bug fixes, refactors and new features on a real open-source codebase: time to working code, how many suggestions needed fixing, and the…
Lovable vs Bolt vs v0 vs Base44 (2026): Pricing, Security and What the Evidence Shows
“Describe it and ship it” is the promise. Four AI app builders, one spec: time, credits burned, bugs, and whether the result…
The protocol
The same five steps,
every review. No exceptions.
01
Pull the claims
Quoted word for word from the product’s own pages.
02
Design a fair test
Real tasks, identical inputs, the plan buyers pick.
03
Record everything
Outputs, timings, errors and the real monthly cost.
04
Label each claim
One verdict per claim, from a fixed five-point scale.
05
Re-test on change
Every revision dated in a public change log.
A verdict scale you can read in a second.
Fixed definitions, applied the same way to every product — so “Verified” here means the same thing next year.
Verified
Tested; the claim holds as stated.
Holds, with caveats
Mostly true, with a limit you should know about.
Overstated
Something real, but the marketing promises more.
Doesn’t hold
Our tests contradict the claim.
Untested
Not yet testable — and we say why.