How we review and score AI tools

How scores work

Each tool receives a single editorial score from 1 to 5 — our editors' overall judgement, not an average of user reviews and not a vendor claim. A score is an opinion, published alongside the criteria below so you can weigh it yourself. We do not publish a review count: we have no mechanism for collecting user ratings, so any such figure would be invented. Listings we have verified but not yet reviewed carry no score and are labelled “Not yet rated”.

What the score covers

• Core capability — does the product deliver its headline features reliably?
• Usability — can a new user reach value quickly, without a specialist?
• Integrations — how well does it connect to the rest of a workflow?
• Value for money — is the pricing fair for what it actually delivers?
• Support and documentation — is help there when something breaks?

Our editorial process

1. 01Verify — we confirm the vendor’s official website, company details, pricing model and that the product is real and reachable.
2. 02Assess — an editor researches the product against its official documentation and pricing pages, then writes an opinionated summary: what it does well, where it falls short and who it suits. A score is our judgement, not a measurement.
3. 03Write in four languages — the overview, pros and cons, use cases and verdict are authored in English, Chinese, Japanese and Korean.
4. 04Mark what is outstanding — a listing we have verified but not yet assessed is labelled as such and carries no score, rather than being filled with placeholder text.
5. 05Re-review — listings are revisited when a product ships major changes, and the catalog is refreshed with every directory update.

Corrections policy

Spotted an error? Email [email protected] with the page URL and what looks wrong — we reply within a week.

Confirmed corrections are applied to every language version at once, so no locale ships a known-stale fact.

How we keep listings current

"Last verified" is the day an editor re-read that one entry against its vendor and re-checked what it claims. An entry that has never had that pass says "not recorded" instead of carrying a date: an empty field states a fact, and a filled-in one that no one did is the thing this directory is correcting.

Neither date is applied by automation. Stamping today's date across every entry would make the catalog look fresher than it is, and that is the failure mode this directory treats as worse than an obvious gap. Coverage is deliberately uneven: you can see which listings have been re-read and which have not.

Where a claim could not be confirmed — a site that errors out, pricing behind a login, a vendor figure nobody else can reproduce — the listing says so under "What we could not verify" rather than quietly asserting it anyway. A working page is not proof of a living product: parked domains and shutdown notices both answer politely.

Editorial team

Every reviewed listing is written by an editor and checked against the vendor’s official documentation. We mark the listings that are verified but not yet reviewed, and we never publish placeholder text as review content. Our scoring criteria, review process and corrections policy are published on this page — and we update them as the AI landscape evolves.

ct-jaryn

Founder & Lead Editor — oversees all tool reviews, rating criteria and editorial standards.

github.com/ct-jaryn

We publish no bylines for our reviewers and claim no credentials we cannot show. What each listing rests on is the vendor documentation it cites, and where that documentation could not be read, the listing says so.