Text forensics · deterministic first

Is this text actually AI-generated?

Paste writing or upload a file. You get one word back — YES, NO, or review — the probability behind it, model-family attribution, provenance and watermark checks, and a numbered list of concrete fixes any editor or downstream service can apply. Every rule is auditable. Nothing is stored.

Your text

Upload file

0 words · 0 characters

Try:

Profile Genre

Waiting for text

The verdict lands here in under 100 ms. You'll see one big word, the AI-style probability, model-family attribution, provenance / watermark state, and every fix.

Deterministic forensics. Serialization tokens, hidden Unicode, dash punctuation, model self-reference, unresolved placeholders.
Editorial rules. Openers, closers, empty implication frames, generic transitions, buzzword density, contrast templates, AI phrases.
Model family attribution. Open-set — the answer can be unknown. No forced labels.
Provenance & watermark adapters. Claude, SynthID, OpenAI text provenance, KGW research watermark, external verifier sidecars.
Mixed authorship. Segment-level scoring on long inputs — flags where the tone shifts.

yes

AI-style score 0%

Model attribution

unknown

Provenance / watermark

not checked

Fixes

Raw JSON response

Same result. One `POST`.

Every verdict above is the exact response your service will get from /verdict. Wire it into moderation, editorial gates, or a downstream rewriter — no login flow, no proprietary schema, no watching a spinner.

curl -X POST https://amiai.theknowledgeformula.com/verdict \
  -H "Content-Type: application/json" \
  -d '{"text":"Certainly! Let me delve into..."}'

Is this text actually AI-generated?

Your text

Same result. One POST.

Same result. One `POST`.