A ChatGPT detector that reads every sentence.
Paste any text. Get a named verdict plus the sentences that still read like ChatGPT.
50 free scans a day · 2,500 words per scan
ChatGPT tells
What ChatGPT writing looks like
ChatGPT is the most widely used model in this category, which means its habits are the ones most people have learned to half-recognise without being able to name them. Its output tends to be relentlessly even: paragraphs of similar length, sentences of similar shape, and an unwillingness to leave any claim without a balancing clause. That evenness is the real signal. Human writing is lumpy — a two-word sentence next to a forty-word one, a paragraph that runs long because the writer got interested.
The balanced hedge
A claim, immediately followed by its own qualification. "While X is significant, it is important to note that Y." The structure repeats until nothing in the piece is ever quite asserted.
Rule of three
Lists arrive in threes with conspicuous regularity — three adjectives, three examples, three reasons — because the pattern is heavily represented in the text the model learned from.
Elevated filler vocabulary
Words that do decorative rather than semantic work: delve, tapestry, testament, realm, pivotal, underscore, navigate the complexities. None is wrong on its own; the density is the tell.
The symmetrical close
A final paragraph that restates the opening in slightly different words and adds nothing. Human conclusions usually end somewhere the introduction did not anticipate.
These are patterns a careful reader can notice. The detector itself does not work from a list of banned words — it scores lexical and structural patterns across the whole document and every sentence in it, which is why rephrasing around a word like delve does not clear a document.
The hard part
Why edited ChatGPT is the hard case
Almost nobody submits raw ChatGPT output any more. They generate a draft, then fix the worst phrasing, cut the throat-clearing, and add a personal line or two. That document is neither machine-written nor human-written, and a detector that returns a single percentage has to round it into one or the other — which is where both false accusations and missed drafts come from. It is the specific reason the AI-Assisted verdict exists here as its own category rather than as a band on a slider.
How a scan runs
What the ChatGPT detector actually returns
Paste the text — from about fifty words up to 2,500 in one free scan. The engine scores joint lexical and structural patterns across the whole document and every sentence in it, then reconciles those into one verdict: Human-Written, AI-Assisted, or AI-Generated. A mixed ChatGPT draft does not collapse into a single misleading average.
You get 50 free scans a day with no account and no card. Text is scored in memory and discarded when the response returns. Nothing is retained for training. The privacy policy in the footer is the longer version of that sentence.
The percentage backs the verdict beside it — how human the passage reads for a Human-Written verdict, how likely it is that a model wrote it for an AI one. The verdict is calibrated per length band, so the same raw score on a short snippet and a long essay is not the same finding. Read the sentence shading before you act on the number — the reasoning is the part you can actually check.
No detector is proof of authorship. False flags cluster on formal prose, technical writing, and fluent English written as a second language. Missed detections cluster on heavily edited machine drafts. Treat a report on OpenAI's ChatGPT as evidence to weigh, not as a finding about a person.
ChatGPT accounts for most of the AI writing people need to check, which is why this page exists as its own URL rather than a filter on the main detector. The scanner is the same engine. What is different is the explanation: the tells above are the ones a careful reader can name in GPT output, and they are not a banned-word list the model keys on.
Newer GPT versions write better and are correspondingly harder to call. Scoring is based on patterns in the writing rather than a fingerprint of a specific release, so the page is not pinned to GPT-4 or GPT-4o. Confidence is lower on a current frontier model than on an older one, and we would rather say that than imply a guarantee.
False flags cluster on formal prose, technical writing, and fluent English written as a second language — registers that overlap with GPT's house style. Read the sentence-level report before you act on the number. Short passages are genuinely noisy; the engine wants around fifty words before it will score anything.
Who this is for
When a ChatGPT detector is the right tool
Use this page when the question is specifically about GPT-family output — a suspected ChatGPT essay, a GPT-4o draft someone edited, a document that has the balanced hedge and the rule-of-three lists. If you do not know which model wrote it, the general detector is the better start.
A school or coursework draft
The most common paste. GPT's even paragraphs and restating close show up clearly on longer essays. Short answers are harder for every detector, this one included — treat a result on a few sentences as a hint, not a finding.
A draft someone 'cleaned up' in ChatGPT
A human outline run through GPT for fluency, or a GPT draft a person then rewrote. That mixed document is what the AI-Assisted verdict is for. Do not round it into an accusation or a clean bill of health.
Published copy that still reads assembled
Help-centre articles, blog posts, and outreach that hit the keywords and still sound like a template. Sentence shading shows which passages to rewrite rather than asking you to scrap the whole page.
Not: proof of who wrote it
A report is a probability estimate about a piece of text. It is not a finding about a person. Do not use it as the sole basis for a grade, a disciplinary outcome, or a hiring decision.
Limits
What a report cannot do
We do not publish a per-model accuracy figure for ChatGPT, or a head-to-head against GPTZero, Turnitin, Originality.ai, or anyone else. Vendor-measured numbers describe the vendor's own test set, not your document. The tells on this page are patterns a careful reader can notice; they are not how the detector works internally and not a banned-word list.
Do not use a report as the sole basis for a grade, a disciplinary outcome, or a hiring decision. Weigh it against drafts, version history, and a conversation with the writer. The FAQ at /faq covers this in more depth because it is the question that actually matters.
Short passages are genuinely hard. Below a few dozen words, any detector's output degrades toward noise. If you only have a sentence or two, the honest move is not to run a scan that will look more confident than it is.
FAQ
ChatGPT detection FAQs
What the detector can tell you about OpenAI's ChatGPT, what it can't, and how to read a result you're not sure about.
Can you detect ChatGPT text?
Yes — that is the primary case this detector is built for, since ChatGPT accounts for most of the AI writing people need to check. Paste the text and you get a verdict of Human-Written, AI-Assisted, or AI-Generated, plus per-sentence shading showing which passages drove the call.
Does it work on GPT-4, GPT-4o and newer versions?
The detector scores statistical patterns in the writing rather than matching against a fixed list of known outputs, so it is not tied to one model version. Newer models write better and are correspondingly harder to call, and the honest position is that any detector's confidence is lower on a current frontier model than on an older one.
Does it just look for words like "delve"?
No. The words listed above are illustrative tells a careful human reader can spot; a detector built on a banned-word list would be trivially defeated by find-and-replace and would flag any human who happens to write formally. The scoring works on joint lexical and structural patterns across the whole document and each sentence in it.
Will it catch ChatGPT text I edited myself?
Often, and that case has its own verdict rather than being forced into a yes or no. Heavy editing genuinely does move a document toward human, which is the correct outcome — at some point an edited draft really is substantially your writing. Read AI-Assisted as "a machine was involved here", not as an accusation.
Is the ChatGPT detector free?
Yes. No account and no card — paste text and scan. There are daily limits to keep the service up, shown on the scanner itself.