A ChatGPT detector reads a piece of writing and scores how machine-made it sounds. It cannot trace text back to an account, and it cannot confirm which program produced a paragraph, because generated text is made of ordinary words and carries no hidden label. Paste 65 words or more into the box below, press check, and you get a score from 0 to 100 plus one of three readings: Reads as AI-generated, Borderline, or Reads as human-written. One run costs one credit at any length, and the free allowance is ten credits a day per network, with no account, no card and nothing to verify. Read the result as a description of the prose, because the prose is all any detector has to work with. That is the honest version of what these tools do, and it is also the useful one: if the text you pasted reads as AI-generated, the phrasing itself is telling you something, and you can see which sentences are doing it with the free registry of AI writing tells, which runs in your browser without touching your allowance. This page explains what a ChatGPT detector can actually see, why detection is a style question rather than an authorship question, what ChatGPT-flavored text looks like when you read it closely, and how to check a draft from either side of the desk.
A model emits ordinary words in ordinary characters, and nothing is stamped on them. Two identical paragraphs, one typed and one generated, are the same file byte for byte, and no tool can separate them, because there is nothing left to separate. That is why every detector on the internet works the same way at the level you can see: it scores how the writing reads, not where it came from.
The maker of ChatGPT has said the same thing in public. OpenAI's published guidance for educators states that AI writing detectors do not reliably distinguish machine text from human text, and it retired the classifier it once offered for its own output, citing a low rate of accuracy. We are not affiliated with OpenAI, so quote its current documentation rather than a summary. The point stands on its own: if the company that builds the models says the detection question is not settled, an outside website cannot settle it either.
What these classifiers actually respond to is how the sentences behave on the page: how predictable the phrasing runs, how even the rhythm stays, how often the same constructions recur. Text that holds to a narrow band of sentence shapes scores higher, whoever wrote it. So the honest reading of a high score is that the text reads the way generated text reads. Who produced it is a fact about an event outside the file, and the file did not record it.
That is why human writing gets flagged. Liang and colleagues at Stanford HAI reported in 2023 that 61% of TOEFL essays written by non-native English speakers were falsely flagged as AI across seven detectors. Every one had a human author. Plain, even, correctly punctuated prose organised to a template is what these scores respond to, and it is also the profile many people were taught to produce. A high score on a paragraph you wrote yourself is a description of register, not a small accusation.
Ask whether a paragraph came from ChatGPT rather than another assistant and no classifier can answer. Chat models are tuned toward the same even, helpful register, so their output converges on the page. A detector scores that shared register, so the same reading comes back for any of them. The honest answer to "which tool wrote this" is that no detector can say, and anyone who claims otherwise is overstating what a score is.
Mixed drafts are the harder case and the common one. A paragraph you wrote and a model tightened lands in the middle, and a sentence you drafted and a grammar checker smoothed reads differently from both. What a score supports is narrower than authorship and still useful: these sentences read this way. Here is what it does not carry:
Reading beats scoring when you have the time, because a sentence you can point at is worth more than a number you cannot explain. Each of these takes one edit, and the registry of 44 named AI writing tells highlights the exact phrase in your own text instead of scoring the document.
The list is not a test. Every pattern appears in human writing too, and a paragraph can hit three of them and still be yours. What the tells give you is names for the parts of the draft you were already unsure about, so the edit becomes a sentence you can point at rather than a feeling about the whole page.
Check the finished text rather than an outline. The floor is 65 words, a run costs one credit whatever the length, and a single run takes up to 12,000 characters, roughly 2,000 words, so most drafts fit in one pass. Then put the same text through the registry of named AI writing tells, free and unlimited in your browser. It names the exact phrase and the reason it fired.
Nothing in that registry is proof on its own, since every pattern appears in human writing too. What separates a generated paragraph from a written one is density: several tells in a short span, on a flat rhythm. Where the checker and the registry point at the same paragraph, you have found the work.
If the draft needs rewriting, the humanizer rewrites every sentence in one pass, up to 12,000 characters or roughly 2,000 words per run, from a 25-word minimum, at one credit per 50 words. Then check the rewritten version here so the reading you act on belongs to the text you are actually submitting.
A score is the start of a conversation, not a finding. Ask about process instead: what they read, what they cut, where the argument came from. A writer who did the work answers in specifics. Decide what your policy permits and what evidence you accept before the first case, not during it.
The score belongs in the file, not in the conversation. If you run a check, note what it said and what you did next, and keep the same standard for every piece of work, so students are not being judged by whichever detector happened to be installed this term. The page for teachers works through classroom process, and the false positives page covers who gets flagged and why.
Keep the record while you work. The signed drafting record captures your session as you write, including timestamps, how much you deleted and whether a large block arrived in one paste, then exports a file signed with ECDSA P-256 that anyone can verify without contacting us. It is free and unlimited, and it records forward, so start it on the piece you are writing next.
If an accusation has already been made, the falsely accused page has the questions to ask, written as sentences you can send: ask to see the report rather than a screenshot of a number, and ask whether the score is the only evidence.
On permission, briefly. Rewriting a draft you were allowed to generate is ordinary editing. Where AI is not permitted, that rule covers the draft however it is edited. Read the policy that applies to you.
No. Generated text is made of ordinary words and carries no hidden label, so there is nothing inside a paragraph to trace. Every detector, this one included, scores how the writing reads: whether the prose has the features these systems associate with machine writing.
No. OpenAI retired the classifier it once offered for its own output, citing a low rate of accuracy, and its guidance for educators says AI writing detectors are not reliable. We are not affiliated with OpenAI, so check its current documentation before citing it.
65 words minimum, and one credit per run whatever the length. The free tier is 10 credits a day per network, with no account, no login, no card and no email address. The tell registry, proofreader, word counter, readability checker, AI vocabulary checker, originality check and drafting record are free with no limit.
Run the tell registry over the same text and look at what fired. If the highlights cluster in one or two paragraphs, that is an ordinary editing job. If almost nothing fires and the score stays high, the cause is usually plain, orderly prose: how you write, not what you did. Start the signed drafting record on your next piece.
Structural edits move a reading much further than word swaps, because the score responds to sentence rhythm and predictability rather than a list of banned words. Vary sentence length, cut the openers that link nothing, and break the even paragraph shape.