How We Check AI Tools

Every tool on this site has a profile made of facts we found in public sources. Each fact links to where we found it and shows when we captured it. We don't give tools a grade or a score, and we don't rank them by quality.

What we look at

• Company & Identity — Is there a registered company? Are the founders public? How old is the domain? Is there verifiable funding or press? Can you actually contact someone?
• Data & Privacy — What the vendor's own privacy policy says about deleting your data, GDPR/CCPA commitments and security certifications, plus any lawsuits or breaches on record. Whether a tool trains on your data is shown as a direct quote from its own policy, with no interpretation from us, because classifying dense legal wording fairly is too error-prone.
• Billing & Subscription — Published pricing, trial-to-paid behaviour, cancellation and refund terms, quoted from the vendor's own documents, and billing complaints we have verified.
• Product Reality — Whether the free tier is real per published terms, and what independent user reports say about output quality and support. Assessed from public signals, not hands-on use.
• Community Reputation — App Store ratings, verified reports from readers, and repeated complaint patterns. Some review platforms block automated collection, so we only include what we could actually verify.

The evidence rule

A fact does not appear on this site unless it is attached to at least one piece of documented evidence: a policy excerpt, a public source URL, a screenshot, or a record in our reports database. If we have no evidence, the section says "Not yet checked" — we never fill gaps with guesses.

How we collect evidence

Facts are collected automatically from public records — company registries and WHOIS, each vendor's own published privacy policy and pricing pages, App Store data and review platforms — and refreshed on a schedule. The date shown next to a fact is when we last captured its source.

Billing facts come from each vendor's own published pricing, refund and cancellation terms, read the same conservative way as privacy policies. Each one links to a verbatim quote from those documents; if the terms don't state something, we leave it blank rather than guess. This reflects what a company publishes, not a hands-on billing test.

Product reality is the hardest area to evidence from public records, so where we can't verify it from published terms or independent reports it stays "Not yet checked". We don't hands-on test products — we report only what the public record can verify.

When a profile is published

Behind the scenes we track which areas have sourced facts. We only let search engines index a profile once we have enough of them — at minimum the company and its privacy policy. Until then the page stays viewable but is marked noindex. This is an internal publishing rule, not a rating of the tool.

Facts we show separately

Technical signals (valid HTTPS, security headers, mail server, malware blocklists) and an official-app check that flags App Store apps using a tool's name but published by a different developer appear in their own sections. A new domain or a missing header doesn't make a tool unsafe, so we show these as plain facts and draw no conclusion from them.

What we exclude

We do not list AI companion/NSFW services, trading or investment bots, or tools that claim to diagnose or prescribe. These categories carry risks our checks are not designed to measure.

AI assistance disclosure

We use AI to assemble review text strictly from our own structured data — every generated sentence maps back to an evidence record, and the mapping is stored for audit. AI never invents claims, or evidence.