How to use this ChatGPT detector
The ChatGPT detector above is our free AI detector, set up for one question: did ChatGPT write this text? Paste 40 to 200 words and you get an answer in a few seconds. You can run 5 checks a day without an account.
- Paste as much of the text as you have, up to 200 words. A full essay or article gives the detector far more to work with than one paragraph.
- Leave out what isn't prose. Titles, reference lists, tables and code say little about who wrote a text.
- Read the score first, then the sentence highlights. Both are explained below.
How to detect ChatGPT writing by eye
ChatGPT has a recognizable style when nobody tells it to write differently. People who read a lot of it start to notice the habits in the table. Each one is a tendency, not a fingerprint: people write this way too, and one habit on its own tells you very little.
| Habit | What it looks like | Why it isn't proof |
|---|---|---|
| Even rhythm | Sentence after sentence of similar length and shape, paragraphs of equal size | Careful writers who follow a template do the same |
| Stock transitions | "Moreover," "Additionally" and "Overall" at the start of sentences | Schools teach these words, and formal writing is full of them |
| Lists of three | Three adjectives, three examples, three bullet points, again and again | It's an old rhetorical device that people use on purpose |
| Contrast formulas | "It's not just X, it's Y" and "not only... but also" | Speeches, ads and opinion pieces use them all the time |
| General claims | Smooth statements with no names, dates, numbers or first-hand detail | Summaries and short answers are general by design |
| Wrap-up endings | A final paragraph that repeats the points above | Many students are taught to end exactly this way |
ChatGPT also changes. Newer GPT models don't always write like older ones, and anyone can prompt them to drop these habits. That's why the detector doesn't hunt for single words. It looks at patterns across the whole text, such as predictable word choices and even sentence rhythm. How AI detectors work explains the method in plain terms.
How to read the % human score and the sentence highlights
The score is the share of your text that reads human. At 50% or more you see Looks human in green. Under 50% you see AI detected in red.
Each sentence is also marked green (reads human) or red (reads like AI). The pattern often tells you more than the number:
- Red almost everywhere: the text reads like an unedited AI draft. This is the case detectors handle best.
- Red in patches: often a person's text with some AI-written passages, or an AI draft that someone partly rewrote. Notice where the red sits, for example only in the introduction and conclusion.
- A score close to 50%: a weak signal either way. Don't read 48% and 52% as two different answers.
- Green almost everywhere: the writing reads human. That still doesn't prove who wrote it.
The score describes how the writing reads, not who wrote it. Use it as a signal, never as the only reason to accuse someone of using ChatGPT.
A GPT detector that also works on Claude and Gemini text
The detector checks the writing itself, not the app that produced it. The same check works on text from ChatGPT, Claude, Gemini, Copilot, DeepSeek and other AI writers, and you don't have to pick a model first.
The flip side is that it can't name the model. A red result means "reads like AI", not "written by ChatGPT". If you need to know which tool someone used, the text alone won't tell you.
When a ChatGPT check is less reliable
Detection works best on long, unedited AI drafts in English. Expect weaker results in these cases:
- Short text. Anything under 40 words isn't scored, and a text just above that line carries much less signal than a full page.
- Heavy editing. A ChatGPT draft that a person rewrote, or a person's draft that ChatGPT polished, mixes two styles, and the result is less certain.
- Non-English text. The detector works best on English. Scores for other languages are less reliable.
- Formulaic writing. Lab reports, legal text, cover letters and five-paragraph essays follow fixed patterns, and fixed patterns can read like AI.
Even OpenAI found this hard. It released its own AI text classifier in January 2023 and withdrew it that July because of its low rate of accuracy. For what is known about error rates, read how accurate AI detectors are.
If your own writing reads like ChatGPT
Plenty of people write the way ChatGPT does, especially in formal or school writing. If your text comes back red, rewrite the red sentences the way you'd say them out loud: shorter where they ramble, more specific where they're vague. Why your writing gets flagged as AI goes through the usual causes.
If you're checking a ChatGPT draft you plan to use, the LumenWrite AI humanizer rewrites it into natural prose and keeps every fact, number and citation. Students checking coursework will find essay-specific advice on the AI detector for essays page.