AI Content Detector
Cautious triage signals on text origin — explicitly not proof of AI or human authorship.
Returns risk-band signals for whether text may be AI-generated, alongside source-similarity and batch analysis routes. It is built to abstain rather than guess: v2 does not publish an authorship-accuracy percentage, and every result is paired with mandatory not-proof language.
What it does
- POST /v2/analyze and /v2/authenticity/analyze return explainable, capped-confidence risk bands rather than a single definitive verdict.
- POST /v2/source-similarity and POST /detectBatch support comparing text against sources and analyzing multiple documents in one call.
- GET /v2/models lists the models the v2 pipeline currently exposes.
Endpoint contract
“Public” routes execute without credentials. “Auth-boundary” routes reject unauthenticated calls — that proves the route exists and is protected, not that it has been exercised end-to-end with a paid key.
| Method | Path | Kind |
|---|---|---|
| GET | /health | Public functional |
| GET | /health/ready | Public functional |
| POST | /detectBatch | Auth boundary |
| POST | /v2/authenticity/analyze | Auth boundary |
| POST | /v2/analyze | Auth boundary |
| POST | /v2/source-similarity | Auth boundary |
| GET | /v2/models | Auth boundary |
Authenticated demo
Use your own RapidAPI key when possible, or the tightly limited shared demo when enabled.
Demo unavailable: Sign-in is ready but disabled on this preview domain. A custom domain is required for Google authentication.
Product feedback
Authenticated feedback is private and cannot be read back through this site.
Feedback unavailable: Sign-in is ready but disabled on this preview domain. A custom domain is required for Google authentication.
Pricing
Pricing is not shown here yet.
Current paid-plan details require an authenticated plan review before publishing — check the live RapidAPI listing for today’s tiers.
Current limitations
- A small integrity probe found the legacy classifier agreed with its own live label mapping on only 5 of 12 samples (41.7%), while exceeding 0.99 confidence on 8 of 12 — this is not a production accuracy benchmark, and no accuracy percentage is published as a marketing claim.
- Because that calibration is not validated, v2 deliberately abstains from a definitive authorship verdict, caps confidence at medium, and requires "not proof" language in every surfaced result.
- Treat every output as a triage signal for further human review, never as proof that text is or is not AI-generated.