Methodology
Every tool indexed by AI Navigator is tested by humans, on the same prompt battery, in the same week. We publish methodology before scores. We accept zero sponsorship for rankings.
Raw Output
We run a fixed battery of 12 prompts per category — covering edge cases, prompt adherence, and consistency. Outputs are blind-rated by three reviewers against a human-produced reference.
Latency
Generation time is measured at peak (US business hours) and off-peak, averaged across ten runs. Queue waits count against the score.
Cost Scale
We compute cost per equivalent output across each tool's pricing tiers and compare to the prevailing API rate for that capability. Free-tier limits are documented in full.
Safety
Watermark resilience, copyright filters, and a standard red-team prompt battery. Tools that ship without basic safety lose half a point automatically.
Disclosure
AI Navigator buys every subscription we test from public funds. We do not accept review units, free credits, or affiliate revenue tied to ranking position. If a tool sponsors the site, that sponsorship is disclosed at the top of the relevant page and never affects placement.