AI Detector Review
Reviews are only worth something if you can check how they were made. This page is the contract between us and you: what our scores mean, where each number comes from, and what we can't yet do.
The evidence tiers
Every rating on this site carries exactly one of four labels:
| Label | Meaning | Who produced the evidence |
|---|---|---|
| Our test | We ran the tool ourselves on our documented test set | Us — procedure below |
| Public study | A named, dated, linked third-party study (university, journal, or independent researcher) | Third parties with no commercial tie to the tools |
| Vendor claim | The maker's own published figure | The vendor — shown for transparency, never treated as verification |
| Pending our tests | We have no evidence we trust yet | — (we show no number rather than a guess) |
We never silently convert one tier into another. A vendor's "99% accuracy" is always labeled a vendor claim even when it appears inside our comparisons.
Scoring dimensions
- Accuracy evidence — what independent testing (ours or public studies) shows about correct classification of AI text
- False-positive evidence — how often human writing gets flagged, the dimension that matters most in academic settings
- Feature fit — integrations, languages, team features, report export, API availability
- Pricing transparency — published prices and free tiers (dated snapshots; always verify on the vendor's site)
- Best-for fit — which user (teacher, publisher, individual writer) the tool genuinely serves
Scores are rendered from a versioned data file, not hand-typed into pages, so every number traces to an evidence record with a URL and date.
Our own testing (scope and limits)
Our hands-on program runs free-tier and trial-account spot checks on comparable text samples, logged with dates. Current scope and limits, stated plainly:
- Tools with institutional-only access (notably Turnitin) cannot be self-tested by us; their rows rely on public studies and vendor claims, labeled as such.
- Paid deep-benchmark passes are budget-gated; until they run, relevant dimensions stay on public-study evidence rather than pretending.
- As new AI models ship, we re-check previously tested tools and update pages with visible dates.
How we handle conflicts
When a vendor claim and a public study disagree, both are shown with their labels. We do not average them and we do not silently pick one. Where multiple public studies disagree, we cite the more recent or better-documented one and note the conflict.
Updates and corrections
Every page shows its last-updated date. Pricing is re-checked on vendor pages periodically. If you find an error or a stale number, email [email protected] — verified corrections are made and noted, not quietly patched.
Independence
We accept no payment for placement, ranking, or verdicts. Some outbound links are affiliate links (labeled at the link) and may earn us a commission; this has no effect on any score. Full details: Affiliate Disclosure.