Every rating on this site comes from a test we ran ourselves. We do not copy scores from vendors or other sites, and we do not publish a score without a run behind it.
The process
- We write a test plan for each job with fictional sample data (no real people) and planted traps, such as a duplicate transaction or a skill the candidate does not have.
- We run the same tasks in each tool and save screenshots and the full output as evidence.
- We score the saved output against a 100-point rubric. Each job has its own criteria (for example accuracy, honesty, clarity, safety), and every rubric includes a friction score for sign-up walls, limits and upsells.
- We note the plan and model we tested, because results on paid and free plans can differ.
Why scores say "Provisional: 1 test run"
AI tools give different answers from run to run. Our plans call for at least two runs per tool. Until the second run is done, every score is marked provisional, and small gaps between tools should not be read as a firm ranking.
Independence
No vendor has paid for or reviewed a score, and we have no sponsorships. We may join affiliate programmes for some tools; that never changes a score, and tools without a programme are tested the same way. See our disclosure.