Free AI detector that shows which sentences read as AI

Yes. You get 3 free scans a day without signing up, up to 800 words each. A free account gives you 20,000 characters a month and 2,000 Writing Assistant words. Paid plans start at $7.49 a month billed yearly.

Yes. It's tested on text from ChatGPT, Claude, Gemini, Grok, Microsoft Copilot, Llama and Mistral, DeepSeek, Qwen, Kimi, GLM, Doubao, ERNIE, MiniMax and Hunyuan. New models are added once we've tested them.

Yes. TextSight is tested on DeepSeek, Qwen (Alibaba), Kimi (Moonshot AI), GLM (Zhipu AI), Doubao (ByteDance), ERNIE (Baidu), MiniMax and Hunyuan (Tencent). It works best when the text is in English and at least 300 words long.

It's built for English. Scores on other languages aren't reliable yet, so use them as a rough guide only. It works best on English text of 300 words or more.

No AI detector is 100% accurate, this one included. In our tests TextSight is about 88–92% accurate on English texts of 300 words or more, 70–78% on texts under 100 words, and it wrongly flags about 5.85% of human texts. Results also depend on the kind of writing, the model and how much it was edited. Treat a score as one signal, not proof. That's why TextSight shows the sentences behind the score. See the accuracy methodology.

Human writing can look like AI when it's very formal, follows a template, repeats the same sentence shapes, or has been heavily polished by a grammar tool. Short texts are harder to judge too. In our tests about 6 in 100 human texts were wrongly flagged (a 5.85% false-positive rate), so a flag is a signal, not proof. Check which sentences were flagged, and if you think it got it wrong, send the text to research@textsight.ai. How we test

They can be. Writers using English as a second language often pick simpler, more predictable words, and that can look like AI to a detector. Our test set includes second-language academic writing so we can measure this, but no detector is completely free of it. Never use a score on its own to accuse someone. How we test

A plagiarism checker compares your text with pages and papers that already exist and shows matches. An AI detector doesn't look for copies. It looks at how the text is written and estimates whether a model like ChatGPT produced it. AI text is usually new, so it can pass a plagiarism check and still read as AI. TextSight has both: this AI detector and a separate plagiarism checker.

Often, yes. Many schools use AI detectors such as Turnitin, GPTZero or Copyleaks, and teachers also compare work with a student's earlier writing, ask about drafts or check document version history. Detectors give a probability, not proof, so most schools treat a high score as a reason to talk, not an automatic penalty. If your course allows some AI use, follow its rules and say how you used it.

Unlimited scans, 25,000 characters per scan, PDF and DOCX upload, 50,000 Writing Assistant words a month, and API access. It's $14.99 a month billed yearly or $19.99 billed monthly.

Yes. Drop a PDF, DOCX or TXT file (up to 10 MB) into the box. Without an account you get 1 document a day, and the first 1,500 words (up to 5 pages) are scored. A free account gives you 5 documents a day, Pro gives you 100.

If you want writing help as well as detection, yes. Both highlight AI sentences, but TextSight also has a built-in Writing Assistant and lets you rescan in the same place.

No. Turnitin's AI detection is sold to schools, so students usually can't run it. TextSight is open to anyone, including writers, editors and teachers. Different detectors can score the same text differently.