AI Detector

Check whether a document or image was made by AI or a script, using the traces left inside the file. This does not analyze writing style. Your file never leaves your browser.

No upload · No sign-up

Drop a file to check

Looks at the traces inside the file, not the writing. No upload, no sign-up: it runs in this browser.

Files: documents DOCX, XLSX, PPTX, PDF, HWPX; images JPG, PNG, WebP, AVIF, HEIC; video and audio MP4, MOV, WAV, MP3

How to check a file

  1. 1

    Add a file

    Drop or choose the document or image to check. Word, Excel, PowerPoint, PDF and HWPX documents, images, and video or audio files all work.

  2. 2

    Read the result

    One line at the top sums it up: an AI generation label was found, the file was made by a script, a copied AI answer left traces, or no traces were found.

  3. 3

    See the evidence

    Each line shows which field said what. From there you can open every file detail or go and remove the traces.

What the AI Detector looks at

Documents: traces of the program that built the file

For documents it reads fields such as author, last saved by, created date and description. When ChatGPT writes a Word file with Python, the author says python-docx, the created date is 23 December 2013 (the fixed date in the python-docx template) and the description says generated by python-docx. These fields often stay as they are even after someone adds their own name and saves it again.

It does the same for libraries that build PowerPoint and Excel files (python-pptx, openpyxl, PptxGenJS) and PDF files (ReportLab and others). Each value was confirmed by building a file with that library ourselves.

Images and video: AI generation labels

Services such as ChatGPT, Gemini and Adobe Firefly add C2PA Content Credentials or an IPTC digital source type to the images they make. When a C2PA signature is present, the official C2PA library checks in your browser whether it is still intact. It also finds the generation data that Stable Diffusion tools (WebUI, ComfyUI, NovelAI and others) store in images.

These labels are often stripped by screenshots, re-saving and sending through messengers or social apps. So no traces does not mean a person made it.

How this differs from writing detectors

Writing detectors estimate how likely it is that AI wrote the text. This tool gives no probability and only reports what is written in the file. It cannot tell who wrote the text or whether an AI answer was pasted in. It also cannot read invisible watermarks such as Google SynthID; only the makers' own checkers can, and those send your file to that company.

Frequently asked questions

How is this different from GPTZero or Turnitin?+

Tools such as GPTZero and Turnitin analyze the writing and estimate how likely it is that AI produced it. This tool does not read the writing. It shows the traces written into the file, such as the program that built it or an AI generation label. They look at different things, so they work well together.

Will it catch an essay written with ChatGPT?+

If ChatGPT built the Word file itself, traces such as python-docx in the author field and the fixed 2013 date can remain. A document where only the text of a ChatGPT answer was pasted in leaves no trace in the file. If links copied from the answer (utm_source=chatgpt.com) or citation markers came along, it shows those.

Can it tell if an image was made by AI?+

Only when a C2PA or IPTC AI label is still in the file. The original file usually has one, but screenshots and messenger or social uploads often strip it. Invisible watermarks such as SynthID cannot be checked here.

If there are no traces, did a person make it?+

Not necessarily. Traces disappear easily when a file is re-saved or captured. This tool only reports what the file says and does not prove that a person made it.

Is my file sent to a server?+

No. The file is only read inside this browser. Checking a C2PA signature also happens here, with the verification engine downloaded to your browser.