AI Tells

Paste a passage. The analyzer counts the habits generative models lean on, from em dashes and "delve" to bold-led bullets and "I hope this helps", then compares the fingerprint with profiles built from thousands of labeled responses. Everything runs in your browser. Nothing is uploaded.

Text to analyze

Prose works best. Markdown is fine and counts toward the structure signals. Fifty words gives a rough read. Three hundred gives a steady one.

Try a sample:
Samples are verbatim replies from GPT-4o, Claude Opus 4 and Gemini 2.5 Pro in the LMArena dataset (CC-BY-4.0), and a Reddit answer from HC3.

AI-likeness

How far the fingerprint sits from the human baseline, as a probability.

Paste some text to begin.

Which family does it resemble?

Given that the text is AI-influenced, how the probability splits across labs. Click a family to see its versions.

Tells found

Every lexicon match in the text, and the families that lean on each one the most.

What moved the score

Features compared with the human baseline, weighted by how much each counts.

Full fingerprint

Every measured feature. Intensity is the 0 to 1 value the profiles are compared on. Copy the JSON to build your own profile from a known sample.

How this works, and how much to trust it