See every sentence
Each sentence is shaded by how it compares with the rest of your text, with an underline as well as a colour. A guide to where to look, not a list of AI sentences.
Private by design: runs entirely in your browser How to check
Is this AI written? The free AI detector with no limits.
Check text from ChatGPT, Claude, Gemini or DeepSeek for free. No sign-up, no word limit, and your text never leaves your browser: never uploaded, never stored, never used for training.
- No limit
- words or characters per check
Try an example:
Drop a .txt or .docx file
It's read in your browser. Nothing is uploaded.
Checks text from the models people use
- ChatGPT
- Claude
- Gemini
- DeepSeek
- Kimi
Inside every check
Everything in one check.
One paste gives you the whole report, free and on your device. Jump to any part.
- VerdictHuman, Mixed, AI or Inconclusive, with a confidence level and the reason.
- Sentence mapEach sentence shaded by how it compares with the rest of your text.
- AI-likenessHow the text compares with the human writing we tested.
- Style signalsStock phrasing, paired words and lists of three, shown as context.
- Invisible-character cleanerZero-width spaces and look-alike letters, found and stripped.
- Narrative fingerprintStorytelling features beside published human and AI base rates.
- Private by designRuns in your browser: no upload, no account, no word limit.
The report
Every part of the report. Nothing behind a paywall.
Check for AI opens the full workspace: your document on one side, the evidence on the other. Values marked Example are illustrations.
AI-likeness, in plain words
Every verdict comes with how the text compares with the human writing in our benchmark. No raw probabilities: when a text scores beyond every human text we tested, it says so.
Whole books, no limit
Paste or open a full manuscript. Every word is scanned, and chapters are found from their headings so you can jump straight to one.
Clean hidden characters
Zero-width spaces, look-alike letters and direction controls are made visible. Copy a cleaned version of the text in one click.
Nothing leaves your device
The model runs in your browser. Besides the page and the model files, the only request is a Google Analytics page view; no request carries your text, and you can check that in DevTools.
How to checkNarrative fingerprint for fiction
Storytelling features from your manuscript, beside human and AI base rates published in the StoryScope study. Context only; never used in the verdict.
How it works
Three steps, all inside your browser.
No upload, no queue, no account. The detector is a small machine-learning model that your browser downloads once and then runs locally.
- 01
Paste or open a file at any length
Paste text, or open a .txt or .docx file. A two-line note is too short to judge; an entire novel is fine.
- 02
Your browser runs the model downloaded once
The first full check downloads the detection model once and keeps it in your browser's cache. After that, checks run on your device, even offline. Your text is never sent anywhere.
- 03
Read the evidence not just a score
An outcome (Human, Mixed, AI or Inconclusive), a confidence level, the share of the text that reads as AI, the AI-likeness score and a sentence map.
Detection
From your text to a verdict, on your device.
A trained classifier reads the text in overlapping windows, a smoother joins the scores into passages, and calibrated thresholds decide. Every stage runs in your browser.
- 1Clean upInvisible and look-alike characters are found and set aside; sentences are split.
- 2WindowsThe text is cut into overlapping windows of about 256 tokens.
- 3ClassifierA small open model scores each window, on your device.
- 4PassagesScores are smoothed into human-like and AI-like passages.
- 5DecisionOne cut-off per text length, tuned to keep false flags on human writing rare.
- Calibrated
- Tuned so that false flags on human writing stay rare
- No word limit
- From a paragraph to a whole book, in one go
- Where it runs
- In this browser tab, in a background worker
Long documents
Novels, manuscripts, theses. Whole books, no limit.
Most free AI checkers stop after a few thousand words. This AI detector for long documents takes the whole manuscript in one go and shows where the AI-like passages sit, chapter by chapter.
No limit
No cap on words, characters or chapters. Only your device's memory sets a ceiling.
Every word
The whole text is scanned and shown, never a sample. Above 30,000 words it is split into pages of 5,000 words to keep your browser fast.
Per chapter
Chapters are found from their headings, so a flagged chapter stands out in a long book.
Narrative fingerprint
For fiction: dialogue share, smell imagery, narrator commentary and more, next to published base rates.
The narrative fingerprint follows StoryScope: Investigating idiosyncrasies in AI fiction (Russell, Rajendhran, Pham et al., arXiv:2604.03136, 2026), which found that AI and human fiction differ in narrative choices, not only in word choice. Our features are simple counts inspired by that work, shown for context and never used in the verdict.
Any AI model
Text from any chatbot. One free check.
The detector looks at how a text reads, not at which chatbot wrote it, so you use the same check whichever one you suspect. Each card shows the family's newest release, updated automatically.
- Made by OpenAIChatGPTNewest release
- Made by AnthropicClaudeNewest release
- Made by GoogleGeminiNewest release
- Made by DeepSeekDeepSeekNewest release
- Made by MoonshotKimiNewest release
- Made by xAIGrokNewest release
Reading a result
What the result tells you, at a glance.
You get an outcome with its confidence, an AI-likeness score from 0 to 100 and the share of the text that reads as AI, because those are different questions.
Human
The text reads like human writing: it stayed below the AI cut-off.
Mixed
At least one substantial passage, about 120 words or more, scored as AI and the rest didn't. The highlights show exactly where.
AI
Most of the text scored as AI and the overall score cleared a cut-off tuned to keep false flags on human writing rare.
Inconclusive
When there isn't enough to go on, for example a very short text, you get Inconclusive and the reason, never a guess.
AI fraction
The share of the words that sit in AI-like passages. It is not the chance that the whole text is AI: 40% means about 40% of the words, not a 40% probability. With a Human or Inconclusive result, or when the likely range is too wide (common in short texts), you get that range in words instead of a single number.
AI-likeness
Shown with every verdict as a score from 0 to 100: how the text compares with human writing we measured, as in “more AI-like than 80% of human writing we tested”. When a text scores above every human text we tested at that length, it says “more AI-like than all the human writing we tested”. A high value is a reason to look closer even when the verdict is Human.
Style signals
Stock phrases, AI-favoured words and em dashes are underlined right in the text, each with a plainer wording one click away. They explain the writing and don't decide the verdict.
Confidence
How firm the outcome is, given the length of the text and how far the scores are from the threshold. Shown as low, medium or high.
Hidden characters
Zero-width characters and look-alike letters are shown in the text and removed in one click.
Edit in place
Accept a suggestion or rewrite a passage right in the page, then re-scan to see the new result.
Who it's for
For anyone who needs to ask “is this AI?”
A free AI detector with no sign-up, used privately: no one sees what you check, including us.
Check your own essay before you hand it in
Paste a draft and see how it reads, privately. Nothing is uploaded or saved, so a check leaves no trace with us or anyone else.
Check an essaySee why it was flagged passage by passage
If formal writing gets flagged, the sentence map shows which passages score highest within your text, so you know where to look.
Check it before you submit privately
See what a detector sees in your essay before anyone else runs one. Nothing is uploaded, so your draft stays yours.
Check your essayStart a conversation not a verdict
Use the highlights as a reason to talk with the student, next to drafts, notes and version history.
Responsible useRead the evidence sentence by sentence
The verdict, its confidence, the AI-likeness score and where the scores run high. Free, with no class or word limits.
Student work stays local nothing is uploaded
Essays are checked in your browser. No account, no upload and no copy kept anywhere.
How to verifyWhole manuscripts no word limit
Open a .docx or paste the full novel; chapters are found from their headings and you can jump straight to any of them.
Unpublished stays private never uploaded
Your draft never leaves your device, so there is no server copy of it anywhere.
How to verifyNarrative fingerprint context for fiction
Dialogue share, smell imagery, narrator commentary and more, beside published human and AI base rates. Never used in the verdict.
Screen submissions then talk to the author
Check a submission of any length and see where the AI-like passages sit. The map shows where to look; it doesn't settle the question.
Hidden characters made visible
Zero-width spaces and look-alike letters, which some tools insert, are flagged and can be stripped in one click.
An open method every step published
How the detector is built, calibrated and tested is published in full, from the test set to the cut-offs.
How it worksReview freelance copy at any length
Check articles and landing pages without a credit meter, a word cap or an account.
Check copyKeep a record as JSON or PDF
Download each result as JSON, or print it to PDF, with the verdict, AI-likeness and sentence bands.
Check every draft no credits to ration
No credits and no daily cap: check every freelance draft, revision and pillar page as often as you like.
Check a draft
Compare
The free AI detector with no limits and no account.
There is no pricing page because there is nothing to sell. The model runs on your device, so a scan costs us nothing and there is no reason to meter it.
No word limit a page or a whole book
An AI detector with no word limit and an AI checker with no character limit: paste a page or a whole book.
No daily cap no credits to spend
No credits, no scans-per-day counter, no blurred results waiting behind an upgrade button.
No account nothing to install
A free AI detector with no sign-up: no email, no login, nothing to install.
How free tiers compare
What the free versions of popular detectors allow, from each vendor's own page.
| Detector | Free limit | Sign-up | Where the text is processed | Source |
|---|---|---|---|---|
| IsThisAiWritten This site | None | None | Your device (runs locally) | How to verify |
| Pangram | 20 credits a day, about 2,000 words | Required | Vendor servers | pangram.comchecked 16 Oct 2026 |
| GPTZero | 10,000 characters without an account | Needed above that | Vendor servers | gptzero.mechecked 16 Oct 2026 |
| QuillBot | 1,200 words per scan, 6 scans a day | Not needed | Vendor servers | quillbot.comchecked 16 Oct 2026 |
| Scribbr | 1,200 words per check | Not needed | Vendor servers | scribbr.comchecked 16 Oct 2026 |
| Originality.ai | 2,000 words per scan, 3 scans a day | Not needed | Vendor servers | originality.aichecked 16 Oct 2026 |
| ZeroGPT | 15,000 characters | Not needed | Vendor servers | zerogpt.comchecked 16 Oct 2026 |
| Copyleaks | 25,000 characters | Not needed | Vendor servers | copyleaks.comchecked 16 Oct 2026 |
Free tiers change often; follow the links for current terms. Pangram counts daily credits rather than checks; QuillBot's product page lists its free tier next to a “Create account” button, while its blog says scans under 1,200 words need no sign-up. "Vendor servers" means the text is sent to the service to be analysed, which is how hosted detectors work.
This table compares free-tier limits only. For a ranking by job, from free checks to classrooms and SEO teams, see the best AI detectors in 2026.
Privacy
A private AI detector. Don't take our word for it.
Most detectors send your text to a server. This one can't: there is no server to send it to. You can check that yourself in about a minute.
- Open your browser's developer tools (F12, or ⌥⌘I on a Mac) and switch to the Network tab.
- Paste some text and press Check for AI. The first time, you'll see the model files download from the hosts listed below.
- Search the requests for a word from your text. Nothing matches: no request carries it.
- Switch the Network tab to Offline and check again. Once the model is cached, it still works.
| Host | What | When |
|---|---|---|
| isthisaiwritten.com | This page, its styles, scripts, font and images | Every visit |
| cdn.jsdelivr.net | The open-source library that runs the model | First full check, then cached |
| models.isthisaiwritten.com | The detection model files, from our own copy | First full check, then cached |
| huggingface.co | The same model files, only if our copy is unavailable | Backup only |
| openrouter.ai | The public list of AI model names, to show the newest releases (no text, no cookies, no referrer) | At most once a day |
| google-analytics.com | Google Analytics page views (never your text or results; paused during a scan) | Each page view |
Responsible use
Get the most from a result.
The report is built to be read, not just glanced at. Here is how to use it well.
- Read the highlights.They show which passages and phrases drive the result, so you can see the evidence for yourself.
- Read the whole report.The verdict, its confidence and the AI-likeness score each answer a different question.
- Use the style notes.Stock phrases, AI-favoured words and em dashes are marked, with a plainer wording one click away. They explain the writing and don't decide the verdict.
- Clean hidden characters.Zero-width characters and look-alike letters are shown and can be removed in one click.
- Check long texts in one go.Paste the whole thesis or novel: every word is scanned and you can jump to any chapter.
- Look at the bigger picture.Drafts, notes and version history show how a text was written. Any detector's result works best next to them.
- Be fair to the writer.If you assess someone else's work, tell them you used a detector, show them the result, and let them respond.
FAQ
Questions, answered plainly.
Short answers here; the methodology page has the long ones.
01Is this AI detector really free?
Yes. There is no paid plan, no sign-up and no daily limit. The check runs on your own device, so there is nothing for us to meter.
02Is there a word limit?
No. Paste a paragraph or an entire book and every word is scanned, never a sample. No credits, no daily cap, no paywall.
03Is there a character limit?
No. Nothing is uploaded, so there is no server-side cap. Very large texts are limited only by your device's memory.
04Does my text get uploaded anywhere?
No. Your text stays in your browser tab. The page loads its own files, the detection model the first time you run a full check (from our own model server, with Hugging Face as a backup, and jsDelivr), and Google Analytics for page views. You can confirm it in your browser's developer tools: no request carries your text.
05Can it check a whole novel or manuscript?
Yes. Paste the manuscript or open it as a .txt or .docx file. You get a chapter-by-chapter breakdown, a sentence map and, for fiction, a narrative fingerprint based on the StoryScope research (arXiv 2604.03136), shown for context and not used in the verdict.
06How accurate is it?
It reads every word with a trained AI-text classifier, and its cut-offs are calibrated per text length on a fixed test set of human and AI writing, tuned to keep false flags on human writing rare. Instead of a bare percentage you get a verdict with its confidence, an AI-likeness score and highlights that show which passages drive the result. How it is built and tested is on the methodology page.
07Which AI models can it detect?
It isn't tied to one model: it looks at how the text reads, so you can paste text from ChatGPT, Claude, Gemini, DeepSeek, Grok or any other chatbot. You don't need to know which chatbot wrote it: the detector reads the text itself.
08Why might human writing be flagged as AI?
Formal, formulaic or heavily edited writing can look statistically similar to AI output to any detector. This one sets its cut-offs per text length to keep false flags on human writing rare, and shows you the exact passages behind a result so you can judge them yourself.
09Can I check .docx or .pdf files?
You can open .txt and .docx files; they are read in your browser and never uploaded. PDF is not supported yet: copy the text out of the PDF and paste it in.
10Do I need an account?
No. There is no login, no email and nothing to install.
11What does a Mixed result mean?
Mixed means part of the text scored as AI and part didn't. The highlights show you exactly which passages read as AI.
Is this AI written? Find out in your browser.
Free, no sign-up, no word limit. Your text stays on your device.
Check text free