How ScanForAI works
The method behind the score, how to read a result honestly, the limits of AI detection, and the API — for both detecting AI writing and cleaning it up. ScanForAI returns an AI-writing score with a per-signal breakdown — a signal, not a verdict.
Methodology
The stylometric signals — rhythm, vocabulary, phrasing, predictability, structure — and how a trained classifier combines them.
Interpreting your score
What the bands mean (reads human → reads like AI), why a score is not proof, and how to act on a result responsibly.
API
Scan text programmatically — endpoints, request/response shape, limits, and authentication with API keys.
The short answers
What does ScanForAI actually detect?
Patterns that machine-written text tends to share — very even sentence rhythm, AI-favoured vocabulary ("delve", "a testament to"), few contractions, the "not just X, it's Y" construction, list/header scaffolding, and highly predictable word choice. It measures how densely those appear and combines them into one score, then shows which fired.
Is it accurate? Can it be wrong?
It's a strong signal, not proof. No detector — including ScanForAI — is 100% accurate. Formal, technical, or non-native human writing can score high; edited or newer AI can score low. Treat the score as one input, never the sole basis for a decision about a person, and never as grounds to accuse. Lightly editing AI text can lower the score, and the newest "human-tuned" models can evade detection almost entirely — a real, honest limit.
Does it store my text?
Anonymous scans are never stored — the text is processed only to produce the score, then discarded. Signed-in scans are saved only for your own history and can be deleted; see the Privacy Policy.
Does it work on any type of writing?
Best on 300+ words. It reads more reliably on writing than on short snippets, code, or lists. You can tell it the text type (e.g. an academic essay), and it relaxes the "formal register" signals so genuinely formal human writing isn't flagged just for being formal.
How should I use a result?
As a prompt for a human conversation, not a verdict. If a piece scores high, look at the highlighted sentences and the reasons, consider the context, and — for anything consequential — talk to the person rather than relying on a number.
How much does it cost?
ScanForAI is free while in beta — the scanner, accounts, history, Cleanup, and the API. Paid plans will come later; beta users will be told well in advance.
Can it help fix my writing too?
Cleanup strips the machine formatting AI tools leave behind (long dashes, curly quotes, invisible characters) and normalizes text to plain — free, and it never rewrites a word. A deeper detector-guided rewrite (Humanize) is coming.