Most files carry hidden data. Text can contain invisible characters and hidden messages. Photos usually store the GPS location, camera and time. AI images can hold Content Credentials, AI labels and the prompt used. Word and PDF files keep author names, edit history, comments and sometimes deleted text.
The short map
| File type | What it can hide | Risk | Check it |
|---|---|---|---|
| Copied text | Zero-width spaces, odd spaces, hidden tag messages | Broken formatting, tracking, prompt injection | Invisible character checker |
| Phone photos | GPS location, camera, serial number, time | Reveals where you live or were | EXIF viewer |
| AI images | Content Credentials, AI labels, prompts, SynthID | Shows how and with what it was made | AI image checker |
| Videos | Content Credentials, AI labels, GPS, phone model | Shows if AI-made, and where it was filmed | AI video checker |
| Word documents | Author, comments, tracked changes, hidden text | Leaks names, drafts and edits | Document viewer |
| PDFs | Author, software, hidden layers, text under redactions | Leaks names and supposedly deleted text | Document viewer |
Text
Text looks like the simplest thing to share, but copied text often carries characters you can't see. Zero-width spaces break searches and passwords. Non-breaking spaces trip up spreadsheets. And Unicode has a whole block of invisible tag characters that can spell out a hidden message inside a normal sentence.
None of these prove who wrote something. They're just characters. But they're worth removing before you paste text into a form, a website, code or an AI chatbot. Read more in why copied text breaks formatting and does ChatGPT watermark text?
Photos
Phone cameras save EXIF data in every photo: usually the exact GPS position, the phone model, the lens, and the date and time to the second. Share the original file and you share all of that. Messaging apps and big social networks often strip it; email, cloud links and many smaller sites don't. See which apps remove photo metadata.
AI-generated images
AI images can carry three kinds of label:
- Content Credentials (C2PA). A signed record saying which tool made the image.
- Metadata labels and prompts. A simple AI flag in the metadata, and sometimes the full prompt.
- Invisible watermarks. Marks like Google's SynthID, hidden in the pixels themselves.
The first two are in the file and easy to read. The third needs the maker's own detector. Our guide on how to tell if an image is AI-generated walks through all three.
Word and PDF documents
Office documents keep the author's name, the company, the template, comments, tracked changes and sometimes text formatted as hidden. PDFs keep author and software details, and badly redacted PDFs can still contain the text that was supposed to be blacked out. Both have built-in ways to clean them: see removing metadata from Word and from PDFs.
A simple habit
Before sharing a file outside your circle, take ten seconds to check it. For text and images you can do it right here, in your browser, without uploading anything.