Frequently asked questions
Where do the scores come from?
From our testers running the published tests on real accounts. Every score comes from a logged run, checked against the test's three checkpoints.
Why is an assistant with higher scores ranked lower?
The overall score adds three neutral 5s before averaging, which pulls assistants tested on fewer dimensions toward the middle. The plain average is shown on each assistant's page.
Why do some scores look grey?
They're older than 90 days. They still count, but the product may have changed since, so we flag them until the next retest.
What is a safety veto?
If an assistant pays, orders or sends something without the permission the test required, that test scores 1 no matter what else it did.
How often are scores updated?
Whenever a new run is logged. Every change to the tests or the scoring rules is listed in the changelog.
Can a company pay to be tested or ranked higher?
No. Anyone, including vendors, can request a test, and vendor requests are labelled as such. Requests decide the order we test in, not the result.
How do I get my AI assistant tested or listed?
Use the request form, or email [email protected] with the assistant's name, its website and which tests to run first. Requests decide the order we test in, never the result.
Can I use or cite Pagent's scores?
Yes. The whole leaderboard is available as JSON at /data/leaderboard.json, and as a plain-text summary for AI assistants at /llms.txt. Link to the page you used and give its update date.
I think a result is wrong. What can I do?
Request a retest from the assistant's page and tell us what changed. If the product has shipped a fix, the next run replaces the old score.