AI Text Checker (carrierscan)

Check whether a text was written by AI (ChatGPT, Gemini, Claude). One read-only tool, check_ai_text: AI likelihood for the whole text and each part. English, Ukrainian, Kazakh, Uzbek, Estonian, Georgian, Lithuanian. Free, no auth.

View on carrierscan.app

Hosted MCP Server

npx add-mcp 'https://carrierscan.app/mcp'

Installs into Claude Code, Codex, Cursor and more

Documentation

Try: human example · AI example · Ukrainian AI example 0 words

Files:.txt,.md,.html,.rtf,.docx,.odt,.pdf. The file is read in your browser; only its text is sent for checking.

[02] How it worksAbout 250 words at a time

Paste it in.
Read every part.

  1. 01

    Paste or upload

    Paste 50 to 2,000 words or upload a document. Files are read in your browser.
  2. 02

    Each part is checked

    The text is split into sections of about 250 words, and each one is scored by a detector for its language.
  3. 03

    You get a likelihood

    You see how AI-like the whole text and each section look. It is a hint for a closer read, not a verdict.

[03] LanguagesAdded only after passing our test

More languages are in training. We add a language only after its detector passes our own test on texts it has never seen.

The most accurate AI detector for Ukrainian.

Tested on the same texts as GPTZero, Copyleaks, Plag and Navidminno.

0 / 20 human texts wrongly called AI by carrierscan

7 / 20 human texts wrongly called AI by Navidminno

40 Ukrainian texts checked on 16 October 2026: 20 written by people before ChatGPT existed, 20 by ChatGPT, Gemini and Claude. Free limits stopped three checkers early: GPTZero saw 17 AI texts, Plag 19, Copyleaks 15; on exactly those texts carrierscan caught 16 of 17, 18 of 19 and 14 of 15. GPTZero also flagged none of its 17 human texts.

[05] Before you judge anyoneRead this first

A hint,
not a verdict.

False positives happen, more often with short texts, formal or technical writing, and people writing in a second language. Lightly edited AI text can pass as human. Never use this alone to accuse a student, writer or employee.

We don't store your text: it is checked in memory and dropped when the answer is sent. Your text doesn't leave our server. Don't paste anything confidential.

If you sign in with Google, we keep a signed cookie in your browser for 30 days with your Google account ID, only to count your checks. We don't store your email or name.

[06] For developersMCP · Streamable HTTP · no auth

Use it from your AI assistant

The same check is available as an MCP server with one tool, check_ai_text. No sign-up; the same limits apply.

https://carrierscan.app/mcp

claude mcp add --transport http ai-text-checker https://carrierscan.app/mcp

Claude.ai, ChatGPT, Cursor and other clients: add a custom connector / remote MCP server with the URL above (Streamable HTTP, no auth).

[07] QuestionsShort answers

How do I check if a text was written by AI?

Paste the text into the box above or upload a file, then press Check text. In a few seconds you see how likely it is that AI wrote the whole text and each part of it.

Do you check for plagiarism too?

No. carrierscan checks whether AI (ChatGPT, Gemini, Claude) wrote a text, not whether it was copied. Plagiarism checkers compare a text with existing sources, and AI-written text is usually new, so it passes them. That gap is what an AI check is for.

Can I check an essay, coursework or a thesis?

Yes. Upload the file (.docx,.pdf,.odt and others) or paste the text. One check takes up to 2,000 words, so check a long thesis chapter by chapter. Each part gets its own score, so you see which sections look AI-written.

Can I check a Google Docs or Word document?

Yes. Upload a Word (.docx) file, or press From Google Drive and pick a Google Doc. We get access to that one file only.

Is it free?

Yes. Without an account you get 60 checks a day; signed in with Google, 200 a day. Each check takes up to 2,000 words.

Can it tell for sure that AI wrote a text?

No detector can. Ours gives a likelihood, and it can be wrong in both directions. Use it to decide what to read more closely, not to accuse anyone.

Which AI models does it recognise?

It is built to recognise text from current chat models such as ChatGPT, Gemini and Claude, including models it was not trained on. Heavily edited or rewritten text is harder to detect.

Do you keep my text?

No. The text is checked in memory and dropped as soon as the answer is sent. We keep only counts and timings, never the text.

Why a separate detector for some languages?

Popular detectors are trained mostly on English and miss much AI text in smaller languages. We train and test a detector on each of these languages before we turn it on.

[08] NextFree · no sign-up

Paste a text.
See who wrote it.