A private AI detector you can check for yourself

IsThisAiWritten is a private AI detector: paste your text below and the analysis runs inside your browser tab, so the text you check is never uploaded. It is free and needs no account. You don't have to take our word for the privacy part. Further down, this page shows how to confirm it with your browser's developer tools in about a minute, and what other detectors do with the text you give them.

What happens to your text during a check

When you paste text and press Check for AI, the page first cleans it, setting aside invisible and look-alike characters, and splits it into sentences. It then hands the text to a detection model running in a background worker inside the same tab. The model scores overlapping windows of the text, and the page turns those scores into a result. All of this happens in your device's memory. Close the tab and the text is gone. We never receive it, so there is nothing for us to store or share.

What the page downloads, and from where

Traffic only flows one way, from these hosts to your browser:

The model is TMR, an open RoBERTa-based classifier published on Hugging Face, so anyone can see exactly what is being run. Once downloaded, it stays in your browser's cache, which is why later checks start faster. The full list is on the privacy page.

How to verify the private AI detector in DevTools

1. With this page open, open your browser's developer tools: F12 on Windows and Linux, or Option-Command-I on a Mac. Choose the Network tab.
2. Paste some text into the box above that includes a word nobody else would use, such as a made-up surname like “Quillfeather”. Press Check for AI.
3. Watch the requests appear. On a first check you'll see the model files arrive from huggingface.co and the runtime library from cdn.jsdelivr.net.
4. Use the search in the Network panel to look for your made-up word across all requests. Nothing matches, because no request carries your text.
5. Set the Network tab's throttling menu to Offline and check a different paragraph. With the model cached, it still works, which would be impossible if the analysis needed a server.

How other AI detectors handle your text

Most detectors run on the company's servers, so your text is sent there. What happens next differs a lot. We read each company's privacy policy and help pages in October 2026; this is what they say they do, not the result of an audit.

Product pages and policies don't always agree. Quetext's detector FAQ says scans are not stored or used for training, while its privacy policy, updated in October 2026, says user content may be included in AI training. When the two differ, the policy is the one to read.

The other way to keep text on your own machines is to host an open-source detector. Binoculars, from an ICML 2024 paper, is free to run, but it loads two Falcon-7B models, so it needs serious hardware, and its authors describe it as for academic purposes only.

When a private AI detector matters

Running locally removes the question of what a server does with your text, which matters in situations like these:

• An unpublished novel or a pitch you don't want sitting in someone else's database.
• A client report or contract draft covered by a confidentiality agreement.
• Student essays, where a school's rules may limit sending work to outside services.
• A message you received and want to check without forwarding it anywhere.

Because nothing is uploaded, there is also no sign-up and no stored history of what you checked. The publishers page covers handling submissions this way.

What privacy does not cover

Privacy and accuracy are separate questions. Like any detector, this one can be wrong in both directions, so treat a private result as a reason to look closer, not as proof. Our methodology page explains how it is tested.

There are also limits to what any web page can promise. A browser extension that has permission to read pages can see what you paste into any site, this one included. Anyone who can use your computer while you're signed in can see an open tab. The site's host, Cloudflare Pages, processes ordinary request data such as your IP address to deliver the page, though never the text you check. If you share a computer, close the tab after use, and delete the cached model if you'd rather not leave it on the machine.

Sources

1. TMR AI text detector (ONNX) on Hugging Face, checked October 2026
2. Pangram, Does Pangram train on my data?, checked October 2026
3. Pangram privacy policy, checked October 2026
4. Pangram, Can Pangram not store my data?, checked October 2026
5. ZeroGPT FAQ, checked October 2026
6. ZeroGPT privacy policy, checked October 2026
7. Winston AI privacy policy, checked October 2026
8. GPTZero, Are you storing data from API calls?, checked October 2026
9. GPTZero, Do I retain ownership of the work?, checked October 2026
10. GPTZero, Privacy and security, checked October 2026
11. Turnitin, Standard assignment settings, checked October 2026
12. Turnitin Trust Center, checked October 2026
13. Copyleaks privacy policy, checked October 2026
14. Originality.ai privacy policy, checked October 2026
15. QuillBot, How does QuillBot use my data?, checked October 2026
16. Grammarly, Product Improvement and Training control, checked October 2026
17. Grammarly privacy, checked October 2026
18. Quetext privacy policy, checked October 2026
19. Binoculars repository and licence, checked October 2026
20. Hans et al., Spotting LLMs With Binoculars, checked October 2026