AI checkers flag human writing, and some writers far more often than others. Read the plan.
Try an example:
Scoring starts as soon as you type.
Never sent to a detector

Switch to Check Risk for the score and the sentences behind it. Free, no sign-up, and the check runs in this tab.

Free AI checker · rehearse the check

Run the AI checker on your own draft first.

Whoever reads your work next may run it through a checker you cannot see the output of. This is the rehearsal: paste the draft, find out which sentences that kind of tool keys on, and decide what to do while you still have the choice. The checker runs in this tab, and your draft is never sent to a detector.

Three things people call the same thing

What an AI checker actually measures.

Most confusion about AI checkers comes from collapsing three separate tools into one word. They make different claims and only one of them can show you its evidence.

An AI checker

Estimates the probability that a language model produced the words. It compares your text against statistical habits, not against any archive. There is no source document it can point to, because there is no source document. The answer is always an inference.

A plagiarism checker

Matches your text against a corpus of existing documents and shows you the overlap. When it flags something it can name the source and quote the line. That is a different kind of claim, and a far more checkable one.

This tool

Neither. It reports which specific constructions in your draft are the ones AI checkers key on, and shows you the sentences that contain them. It is a rehearsal for the check you cannot run yourself.

41 surface patterns plus 3 whole-document cadence detectors, all executing in your browser. How the score is composed →

The workflow

Check it before someone else does.

1

Paste the draft

Straight into the box at the top of this page. No account, no upload, no word cap. Scoring starts once there are 25 words, because a rate computed over two sentences means nothing.

2

Read the band, not the number

Low, Moderate, Elevated or High. The number moves as you type and the exact value matters far less than which band you are sitting in and which paragraph put you there.

3

Read the highlighted sentences

Every finding underlines the characters that triggered it and explains itself in a sentence. The registry ships 44 of them, so the report usually tells you something specific about your habits.

4

Decide: rewrite, or document

If the draft is machine-flat, rewrite it. If you wrote it yourself and the reading is still high, do not sand your voice down — export a signed evidence report instead and keep the receipt.

False positives

The writers most likely to fail an AI checker.

Statistical checkers flag whatever sits furthest from the writing they were trained on. In practice that means they flag people, and not at random.

Checkers do not fail evenly. Peer-reviewed work found 61% of TOEFL essays by non-native English speakers falsely flagged across seven commercial detectors, while essays by native speakers came through nearly clean. Who gets flagged, and why it happens covers the study, the mechanism and the four cohorts in full.

The Detector Observatory

We are building a public audit of the checkers.

A planned audit of commercial detectors against cohort-labelled human writing. When it runs, the methodology and the corpora will be published and every run downloadable as CSV, across 4 cohorts. Cohort-level false-positive rates are the number this category does not report, and that is the number the audit is built to produce.

Figures in this build are illustrative. They model the shape of the published literature and are not measurements.
Widest cohort gap · June 2026 run
28.8pp

Detector H flags 6.5% of native-English writing and 35.3% of writing drafted with assistive tools. Same corpus, same run, same day.

If you have been accused

What a checker result can and cannot prove.

A percentage from any AI checker is a classifier's confidence, produced from writing style alone. It is not testimony, it carries no chain of custody, and the threshold that turned your draft red is not published anywhere.

What can be shown is the drafting itself. Write in the editor and it records your local edit history and cadence, then seals the record with a signing key your browser generates and keeps. The verifier checks that signature in the reader's own browser, no account required.

Cannot proveThat a machine wrote a document. No detector can, because text carries no record of its own origin.
Cannot proveThat a person wrote a document. A clean score is the absence of evidence, not evidence of absence.
Can showHow a draft evolved: revision history, timing, and the order the sentences arrived in.
Can showThat the record has not been altered since it was signed, checkable by anyone with a browser.
Check a signed report
Privacy

Check a draft without sending it anywhere.

The checker is 41 pattern rules and 3 document statistics running as JavaScript in this tab. No model, no request, no account. Your draft is never submitted to a detector.

No upload to check

The score is computed in this tab, over a string that stays in this tab. There is no analysis endpoint behind it.

Never sent to a detector

Your text is not passed to GPTZero, Turnitin, Copyleaks or any other classifier. That holds for every feature on this site.

Works offline

Load the page once and disconnect. The checker keeps scoring, because the rules are already here.

Next step

If your draft checks as AI.

Do not chase a number. Fix the writing that produced it. The humanizer takes the same highlighted clauses and rewrites them inside your own measured style, one at a time, with you approving each change.

You get the exact location of every stock opener, negation pivot and flat-rhythm run in the draft, so the sentences you keep are the ones you decided to keep. What a named detector will output on a given day is not something any tool can tell you.

What you get
A band and a score that update as you type, from 41 pattern rules and 3 document statistics.
The exact characters that triggered every finding, underlined in place, with a one-line explanation of each.
A rewrite of any flagged clause in your own measured style, one at a time, with you approving each change.
A signed evidence report of how the draft was written, checkable by anyone with a browser.

AI checker: common questions

Do I need an account, and is there a word limit?

No account and no limit. The checker runs in your browser, so there is no server cost per check and nothing to meter. Rewriting and evidence reports live in the full editor.

Is an AI checker the same as a plagiarism checker?

No. A plagiarism checker matches your text against documents that exist and can quote the source it matched. An AI checker estimates a probability from writing style and has no source to show you. The two get discussed as one thing and they are not remotely the same kind of claim.

Will my school's AI checker agree with this one?

Not necessarily. Turnitin, GPTZero, Copyleaks and the rest run their own classifiers on thresholds they do not publish. This tool shows you which features in your draft that class of system keys on, sentence by sentence, which is the part you can act on.

I used Grammarly or dictation. Does that count as AI?

Most institutional policies say no, but most AI checkers cannot tell. Grammar assistance and speech-to-text both flatten sentence rhythm, which is one of the properties detectors measure. That is a large part of why assistive-technology users show up in false-positive research.

Can I check work that someone else wrote?

Yes, for any text you have the right to read. The checker runs in your browser, so the text stays in the tab.

Is running my own draft through a checker against my institution's policy?

Checking your own draft is not usually restricted. Where generative AI is not permitted in submitted work, that rule governs regardless of which tool you use.

Keep reading

Check the draft.
Keep the draft.

Free AI checker. No account. Runs in your browser.

Open the Full ReportRead the Audit