Reads flawless 띄어쓰기 as a signal, not a compliment.
Free Korean AI detector. Check Korean text for the patterns typical of AI writing: speech level, spacing, loanword spelling. Signals, never proof.
Korean AI detector
Paste Korean text and this Korean AI detector reads it for the patterns that show up in machine-written Korean, then tells you what it saw and why. You get a description of the signals, not a percentage and not a verdict on who wrote it.
The check runs on our server, so the text you paste is sent to us to be analysed. You get 5 free checks per rolling 24 hours and each check takes up to 12,000 characters. No account, no signup, no payment.
The problem, in plain terms
Most tools that advertise Korean detection are English tools with the word Korean printed on them. They talk about perplexity and writing patterns and never mention a single feature of the language. Korean has specific things worth looking at, and here is what they are.
None of these is proof on its own. A careful human writer produces some of them, and a lazy model avoids others. They are worth reading as a group, which is what the check does.
How to
The whole check happens in the box above.
The limit
The cap is 12,000 characters, the same number used across every server-backed tool here. In Korean that number goes considerably further than it would in English, because a Korean syllable block is one character and it usually carries what two or three English letters would.
In practice a 12,000 character limit covers most 리포트 and essay-length pieces whole. For a longer document, check it in sections, and remember each section spends one of your five daily runs.
Straight answer
No, and the tool is built so it cannot pretend otherwise. The instruction our server sends with every check tells the engine never to claim certainty about authorship or about which AI model wrote a text, and to describe observable signals and limitations instead.
Pages that promise to name ChatGPT or Claude or Gemini as the author are promising something that cannot be read off the finished text. What you can get is a description of what the writing looks like. That is what this gives you.
Why it misfires
Detectors reward irregularity, and formal Korean writing is trained out of it. A 보고서 written to a company template, a 자기소개서 rewritten twenty times, an academic abstract in 합쇼체: all three are supposed to be even, and evenness is the thing detectors read as machine-like.
This is not only a Korean problem, and it is documented. The paper "GPT detectors are biased against non-native English writers" by Liang, Yuksekgonul, Mao, Wu and Zou found that the detectors they tested consistently misclassified non-native English writing as AI-generated while classifying native writing correctly, and concluded that such detectors may penalise writers with constrained linguistic expression. That finding is about English written by non-native speakers, which is exactly the situation a Korean student submitting an English paper is in.
What to do
A flag is not a finding of fact and nobody should treat it as one. What helps is evidence of the writing happening, which a detector cannot produce and you can.
Who it's for
Teachers and 조교 reading submitted Korean work, editors checking copy from a freelancer, translators reviewing a Korean draft they did not write, and writers checking their own text before they hand it over. The last group is the biggest one.
Everyone gets 5 free checks per rolling 24 hours, counted by our server, with no account. When you run out the counter refills as the oldest run falls outside the window, so it is a rolling limit rather than a midnight reset.
FAQ
Everything below is answered in full, nothing to click open.
Paste it into the box above and run the check. You get a description of the signals found in the writing, not a score.
Yes. Five checks per rolling 24 hours, with no account and no payment.
Yes. This check runs on our server, so the text is sent to us to be analysed. It is not a browser-only tool.
Up to 12,000 characters per check, and at least one character. Longer documents need to be checked in sections.
More than it sounds. A Korean syllable block is one character, so the same cap holds considerably more writing than it would in English.
No. The engine is instructed never to claim which model produced a text, because that cannot be read reliably off the finished writing.
No, and we do not publish one. We have no measured accuracy figure for this route, so quoting a number would be inventing it.
No. It reports observable signals in the text. Authorship is not something any detector can establish.
Speech-level consistency, 띄어쓰기 accuracy, loanword spelling, particle variety and sentence-ending repetition, among others.
Because Korean spacing rules are hard and fluent native writers break them routinely in informal prose. Uniform correctness over a long passage is unusual.
Everyday Korean is full of non-standard forms like 메세지 and 초콜렛. Text that uses only the standard 메시지 and 초콜릿 reads as edited or generated.
Keep your drafts and version history. A document's edit history is stronger evidence of authorship than any detector output, in either direction.
Yes, because 보고서 and 자기소개서 writing is deliberately even and regular, and evenness is what detectors read as machine-like.
You can paste it, but the signals described here are Korean ones. A passage that is mostly English gives the check much less to work with.
No. This is a text tool. Paste the text itself, not a photograph or a PDF page.
No. This tool checks text, it does not rewrite it, and no tool here promises any particular result from someone else's detector.
This one reads the text and reports what it sees. The humanizer rewrites it. Different routes, different jobs.
Yes. Paste and run on mobile exactly as on a desktop.
No. It is one paste at a time, and there is no programmatic endpoint.
The check returns a message saying the free runs for the day are used. The allowance is rolling, so it refills as your oldest run ages past 24 hours.
No. It is independent of all of them and cannot predict what any of them would report.
No. It describes signals in writing. A decision about a person should rest on evidence, such as drafts and version history, not on a tool's description.
Five free checks every 24 hours, up to 12,000 characters each, with the signals explained rather than scored.
Run the Korean check