Independence and review
Why these numbers can be trusted: every link can be checked
Whether a measurement provider can be trusted depends on what it puts out for others to check. This page strings those places together: one line each, linking to the page itself, with counts and versions read at the moment you open it.
Four pages, one chain of evidence
- 01
How we measure
The model panel, the exact prompts and the scoring formula are on the methodology page, and the panel is read from the configuration actually in use. Public rankings ask only what models remember; the web-enabled panel is used only for client projects, and the two are reported separately. Every model swap and formula change goes into the panel-changes table.
Methodology →Current methodology v2026-09-28.1 · 12 panel-change entries
- 02
What we have declined
Every request to pay for a listing, pay for a score change, remove a record or set custom weights is logged, anonymised. Entries are added, never edited or removed.
Refusal log →0 entries as of 2026-09-28
- 03
What we got wrong
Which published data was wrong, when it was fixed and which rankings it affected, entry by entry. Entries are added, never edited or removed.
Corrections →3 entries · latest 2026-09-02
- 04
Can every number be checked
Enter a record number to see the exact question asked in that measurement, each model's rank and weight, and every raw response.
Verify →No login. No payment.
Where our revenue comes from, and where it does not
When a measurement provider's income depends on what the measurement shows, the results tend to lean toward the income. So how we charge is itself a set of rules:
- We charge for measurement only: certificates and archives, and monitoring projects. Each is a fee for the measurement itself.
- Nothing is charged on results. Prices are fixed and do not change with what the measurement finds, whether the numbers go up or down.
- We neither pay nor accept referral fees from any service provider.
- Reports are never white-labelled; each one is issued under the ask10ais name.
- Prices are published on the pricing page and are the same for every client. We do not negotiate them.
- The question set, the model panel and the verdict rules are frozen before a client's campaign starts, and the hash of the panel configuration goes into the report. Once a report is delivered, we do not change the parameters for the next period. A change means a new version and a new baseline, and the two sides are not compared directly.
The web-enabled panel can search the web, so newly published content affects it. Its neutrality rests on the rules above rather than on a technical property. What each panel measures is set out on the methodology page.
Data licence
The compiled data we publish — the panel-changes table and the chart data in rankings and reports — is released under CC BY 4.0; cite ask10ais and the measurement date. Rights in the raw model responses belong to the model providers; we keep them as returned, only so they can be checked.
External replication
The algorithm and panel configuration are published on the methodology page, and the exact prompts on every ranking page and verification page. Research groups are welcome to replicate the measurement from them. Replication results we receive will be listed here one by one, with the date, who ran it, what it covered and what it found — including results that disagree with ours.
As of 2026-09-28, no external replication results have been received.
What we do not do
- We do not sell content placement, ghostwriting or any service aimed at getting models to recommend a brand more, and we do not promise a higher recommendation rate.
- We do not accept payment to change a score, add a listing or remove one; such requests are logged, anonymised, in the refusal log.
- Brands choose the questions and scenarios. They cannot pick the models, change the weights, or change the prompts after seeing the results.
Comparing measurement providers? 10 questions to ask before you buy →