How We Review
We tell you which AI tools are safe, what they cost, and what they do with your data. That’s only useful if you can trust how we decided.
This page is the working rulebook: how facts get checked before they ship, how the privacy ratings are decided, and what happens when we get something wrong. Because we will, occasionally, get something wrong.
For the companion piece — how AI itself helps us build this platform, and what stays human every time — see How We Make AI Unspun.
What we check, every time
Four rules, applied to every article, tool review, and glossary entry on this site. None of them are complicated. All of them are non-negotiable.
Every claim gets a source
When we say a tool trains on your data, that comes from the tool's own policy documents — not from a headline about them. News posts link their primary source. Tool pages link the official policies we read.
Dates are visible
AI moves fast, and a fact about pricing, models, or privacy has a shelf life. Pages that carry fast-moving facts show when they were last checked, so you can judge the freshness yourself.
Fast-changing facts get re-checked
Model names, prices, free tiers, and privacy settings are treated as perishable. Our AI agents re-check them on a schedule — and when something has changed, a person verifies it and approves the update before the page changes.
A person signs off on anything new
AI helps us draft, check, and build — we're open about that. But nothing new is published, and no rating changes, without a person reading it and putting their judgement behind it.
How we rate AI privacy
The AI Privacy Checker answers one question per tool: is this safe to use? Here’s how those answers get made.
The official documents get read
The privacy policy, the data-usage FAQ, the settings pages. The dull stuff, in other words. Our AI agents do the scheduled reading; where those documents are linked on a tool's privacy page, that's the exact material the review is based on.
We answer the questions that actually matter
Does it train on what you type? Can you turn that off? What happens to your data? Plain answers, in plain English — not a summary of legal language, a translation of it.
We rate it, and date it
Each tool gets a safety rating and a privacy risk level, plus the date it was last reviewed. A rating never changes without a person approving it. A check that finds nothing changed still updates the date — that's the point of it.
What the ratings mean
The default settings are reasonable. Fine for everyday use with ordinary common sense.
Usable — but check your settings first, and think before sharing anything sensitive.
Read up before you trust it with anything that matters. There's something here worth understanding first.
Each tool also carries a privacy risk level — Low, Medium, or High — and the date it was last reviewed. Re-checks run on a schedule, and immediately if a tool changes its policy, changes owners, or has an incident worth knowing about. When a check finds a change, a person reviews and approves the fix before the page updates. When it finds nothing changed, the review date moves and the rating stands.
Corrections
“A site that reviews fast-moving tools and never issues a correction isn’t accurate. It’s just not looking.”
Spotted something wrong — an outdated price, a model name that’s moved on, a privacy setting that’s changed? Tell us. Here’s what happens next:
- We verify it against the original source.
- We fix the page, and update its "last reviewed" date.
- If the error changed the substance of a rating or recommendation, we say so on the page rather than fixing it quietly.
Who does the reviewing
A person, and a small team of AI agents — working the same way we teach. The agents do the scheduled checking and the first drafts. The founder QAs their work, makes the judgement calls, and approves anything of substance before it goes live.
The characters who present it — Kevin, The Skeptic, and the rest — are openly AI-generated, and the thinking behind them is not. You can read exactly where the line sits on How We Make AI Unspun, and get to know the cast on the cast page.
I’ve read more privacy policies than anyone should. You’re welcome.
What we won’t do
No ratings for sale. A tool cannot pay for a better privacy rating, and affiliate relationships never touch the ratings.
No unsupervised verdicts. Our AI agents check and draft — but a rating never changes without a person approving it.
No quiet edits on matters of fact. When we correct something that mattered, the fix is dated.
No pretending to certainty. When a policy is ambiguous, we say it's ambiguous.
Some links to tools on this site are affiliate links. They never influence a rating, and they’re disclosed — details on our affiliate disclosure page.