AI Generated Text Detector
Check any piece of writing for patterns associated with AI-generated text, whatever its format.
0 words · Add at least 50 words for a more meaningful analysis.
AI detection is probabilistic and can be wrong. Text is sent to our detection provider for this check only and isn’t stored by us. How your text is handled
What you can check
The checker reads any English prose. People use it to look at a vendor’s email, a freelance article, a batch of product reviews, a report from a colleague or a viral social post. It is a general-purpose tool, so there is no setting to choose. Paste the text, or upload a document, and the same analysis runs.
Different formats behave differently. Marketing copy and templated business emails are often formulaic, which can make human writing look machine-like. Personal posts and opinionated writing tend to be easier to read. Keep the format in mind when you read the result.
What detection finds, and what it does not
Detection finds statistical patterns that are commonly associated with AI-generated writing. It does not find an author. Two people can write in a similar style, and a person can edit generated text until it reads as their own. The result describes the text in front of it, not the history behind it.
Under the hood, the Pangram AI detector labels each passage as AI-Generated, AI-Assisted or Human Written, with a confidence of High, Medium or Low. We turn those labels into a verdict and a “Flagged as AI-like” share, which is the proportion of passages labeled AI-like, not the chance that the entire text is AI. You can read more in how AI detection works.
What to do with an ambiguous result
Many results land in the middle. When they do, resist the urge to settle the question from the number alone.
- Read the highlights. A few scattered flagged sentences mean something very different from one long flagged stretch.
- Check the length. If the sample is short, find more writing from the same source and check that too.
- Look for specifics. Real names, dates, first-hand details and checkable sources are hard to fake without effort.
- Check the source. A sender, publisher or reviewer with a track record gives you more to go on than the text alone.
- Ask the writer. If it matters, ask how the text was made. A direct question usually settles more than a second tool.
If the result affects someone, such as a contractor’s payment or a person’s reputation, do not decide on a detector result alone. The accuracy page explains the limits plainly.
When results are less reliable
Short texts, heavily edited drafts, translated writing, very formal prose and text by non-native English speakers can all produce misleading results. Scanned PDFs cannot be read at all because they hold images, not text. Results can also differ between sections of the same piece, which is a sign to look harder rather than trust any single run.
If you are checking student work specifically, the teacher page covers a review workflow. For longer academic writing, see the essay checker. The methodology explains what each part of a result represents.
Detection, watermarks and provenance
A detector like this one reads the finished text and judges how much it resembles machine writing. Some AI vendors take a different route and build a signal into the text as it is generated. OpenAI, for example, says ChatGPT text in the EU carries an invisible watermark, but its detector is limited to approved researchers, so ordinary readers cannot check for it. A watermark is evidence of origin when it is found, and not finding one tells you nothing. For the details and sources, see how to tell if text is AI-generated.
The result you get here is a judgment about resemblance, never proof of who wrote something or how. How reliable that judgment is depends on the text; how accurate are AI detectors lays out what independent studies and vendors report, with the conditions attached.
Human, AI-assisted or AI-generated?
Most real writing sits between the extremes: a person drafts and a tool tidies, or a tool drafts and a person rewrites. The detector’s AI-Assisted label exists for that middle ground. What the difference means in practice, and how policies treat each case, is covered in AI-assisted vs AI-generated writing. If you want to improve a draft that reads flat, why AI writing sounds robotic explains the usual causes.
Frequently asked questions
What kinds of text can I check?
Emails, articles, product reviews, reports, social media posts, essays and similar prose. Paste it in or upload a TXT, DOCX or PDF file up to 10 MB. Each check needs 50 to 5,000 words.
Can it tell me who wrote the text?
No. The detector looks for statistical patterns in the writing. It cannot identify an author, a tool or a prompt, and it has no way to know how a piece was produced.
Why do short texts give weak results?
A short text offers few passages to judge, so a single sentence can swing the outcome. Below 50 words we report that there is not enough text. A longer sample gives more stable results.
What if the result says Mixed signals?
Some passages were labeled AI-like and others were not. Read the highlighted sentences, consider where the text came from, and look for other evidence before reaching a conclusion.
Is my text stored?
Our server does not store submitted text or uploaded files, and our logs contain no text. The text is sent to Pangram Labs for analysis, and Pangram’s own terms govern what happens to it there. See the privacy page for details.
Does it work on languages other than English?
It is most reliable for English. Other languages, translated text and non-native English writing can produce misleading results, so interpret them carefully.
Guides for reading your result
- How to Tell If Text Is AI-Generated
- AI-Assisted vs AI-Generated Writing
- What Does an AI Detection Percentage Mean?
- Best Free AI Detectors
- How Accurate Are AI Detectors?
- Why Does AI Writing Sound Robotic?
All guides on understanding ai text detection · Browse all guides