AI Content Detector
Analyze text to determine if it was written by AI or a human. Detects patterns from ChatGPT, GPT-4, Claude, and Gemini. Free, private, no signup required.
Text to Analyze
0 wordsBest results with 100+ words. This tool uses heuristic analysis, results are indicative, not definitive.
Analysis Results
Results will appear here
How to check text for signs of AI writing
Paste a decent length of text, run the checks, and read why it scored the way it did.
- 01
Paste at least 100 words
The tool won't score anything under 30 words, and results below about 100 are mostly noise. A full essay, article or cover letter gives the signals something to work with.
- 02
Paste one author's writing at a time
If a document mixes a human introduction with AI-drafted sections, check the parts separately. A blended text gets a blended score that doesn't describe either part.
- 03
Read the signals, not just the score
Each of the seven checks is listed with what it found. A high score driven only by "no contractions" means something very different from one driven by a dozen stock phrases.
- 04
Treat the verdict as a prompt to look closer
Use it to decide which drafts to read more carefully or to edit your own writing. Don't use it as proof of anything.
How the score is worked out
This isn't a trained AI model. It's a set of seven style checks that run in your browser, each worth a fixed number of points out of 100. The score is the share of those points the text earns on the AI side, clamped between 2% and 98% so it never claims certainty.
| Signal | Max points | What earns them |
|---|---|---|
| Stock AI phrases | 30 | 66 phrases such as "it is worth noting" and "delve into"; 5 or more found gets the full 30 |
| Heavy structure | 20 | Many numbered steps, headers or bold labels |
| Uniform sentence length | 12 | Standard deviation of sentence length under 3.5 words |
| Transition words | 12 | More than 0.4 per sentence from a list of 17, such as moreover and furthermore |
| No personal voice | 12 | Little or no use of I, me or my |
| No contractions | 8 | Zero contractions in a text over 80 words |
| No emotion or hedging | 6 | None of 31 words like honestly, unfortunately or I think |
The verdict follows the score: 65% and above reads as likely AI, 50 to 64% possibly AI, 38 to 49% uncertain, and below 38% likely human. Because stock phrases and structure carry half the weight between them, an ordinary formal report with numbered sections and no slang can score fairly high without a word of it coming from AI.
Why no AI detector is reliable enough to accuse someone
Detection is genuinely hard. OpenAI withdrew its own AI text classifier in July 2023, saying its accuracy was too low; it had correctly flagged only about a quarter of AI-written text in testing. A 2023 Stanford study published in Patterns found that several popular detectors labelled more than half of essays by non-native English speakers as AI-written.
The reason is that the signals overlap with plain, careful writing. Students taught to use "furthermore", writers who avoid contractions, and anyone following a template all look a bit like a language model. Edited or paraphrased AI text, meanwhile, slips through easily. That's why the results here should start a conversation, never end one.
Sensible ways to use it
Making your own drafts sound more human
If you've used AI to draft, the signal list shows exactly what to rewrite: the stock phrases, the uniform sentences, the missing "I".
Triage for an editor's inbox
A high score is a reason to read a submission more closely, check its sources and ask the writer about it, not a reason to reject it.
Teaching what generic writing looks like
Run a bland paragraph and a lively one side by side. The signals make a vague idea, "this sounds robotic", concrete.
Limits worth knowing
- It works on English only. The phrase and word lists are English, so other languages score close to neutral whatever their origin.
- It doesn't know which model wrote anything. It can't tell ChatGPT from Claude or Gemini, or any of them from a person who writes in the same style.
- It's easy to fool either way. Adding contractions and a few personal asides lowers the score; formal human writing raises it.
- Short text is unreliable. Below about 100 words, one stray phrase can swing the result by 10 points or more.
- Code, lists and tables confuse the sentence-length and structure checks, so check prose only.