Check if text or an image was made by AI
Free, ad-free AI detector. Everything runs on your device: your text and images never leave your browser.
- Runs on your device
- C2PA provenance
- Invisible-character scan
- Open source
Paste text to see the AI index, the evidence behind it and where it changes sentence by sentence.
Evidence
Why this score
Sentence by sentence
Detected AI phrases
Sentence analysis
Drop an image to read its Content Credentials, generator signatures and pixel statistics.
Evidence
Content Credentials (C2PA)
Provenance metadata
Surface statistics
Informational only. These do not decide the verdict.
Perceptual hashes
Counts
Time
Readability
Top keywords
Common phrases
Find the writing patterns AI models over-produce, with quotes and fixes. Based on the no-ai-slop and stop-slop rules.
Findings
Pattern catalogue
Phrase, character and emoji tells, scored by density and variety. Kept separate because curly quotes and non-breaking spaces are normal in many languages.
Cleaned text
Find near-duplicate wording between two passages, locally.
Shared phrases
Site statistics
Privacy-first: no cookies, no fingerprinting, no personal data stored.
Top pages
UI languages
Countries
Devices
Use it as an API
The same engine is available as a REST endpoint, an npm package and a CLI.
Endpoints
POST /api/detect { text, language?, sensitivity?, sentences? }
POST /api/count { text, language? }
POST /api/image { base64, mime? } or { data, width, height }
POST /api/provenance { base64, mime? } C2PA + SynthID + OpenAI
GET /api/health
GET /api/statsExample
curl -s https://is-it-ai-generated.org/api/detect \
-H 'content-type: application/json' \
-d '{"text":"Your text here"}' | jq .scorenpm package
npm install is-it-ai-generated
import { detectText, analyzeC2pa, scanUnicode } from 'is-it-ai-generated';Command line
echo "text" | isitai
isitai --json article.txtAbout and privacy
Privacy
Text and images are processed entirely in your browser.
Open source
The full source is on GitHub under the MIT licence.
Limits
This is a heuristic index, not proof.
What runs under the hood
- Unicode forensics — invisible and zero-width characters, Unicode tag smuggling, variation selectors, homoglyph and mixed-script substitution.
- Provenance — C2PA Content Credentials parsed from JPEG, PNG, WebP and ISOBMFF, with the claim, actions and signer. Optional OpenAI content provenance.
- Watermarks — SynthID-Text and green-list statistical detectors. They need the provider key, which is not public; they work with your own key.
- Stylometry — 16 weighted features including compressibility, burstiness, lexical diversity and structure.