Security & Threats
Does ChatGPT, Claude or Gemini Keep the Metadata in Files You Upload?
Uploading a photo or PDF to an AI assistant sends the whole file — EXIF, author fields and all.
· 8 min read
Short answer: When you upload a file to ChatGPT, Claude or Gemini, the whole file goes to the provider — metadata included. The model may not read EXIF unless it runs a tool that parses it, but the bytes are stored on the provider's servers under their retention policy. Strip metadata before you upload, not after.
What Actually Happens to an Uploaded File
- Your file is transmitted to the provider's servers and stored for the life of the conversation (and longer under abuse-monitoring retention).
- Text is extracted for the model to read. PDFs and Office documents are parsed; images are passed to a vision encoder.
- Any code-execution or analysis tool the assistant runs has access to the raw bytes, which means EXIF, XMP, PDF document info and Office docProps are all readable.
- Deleting the chat removes it from your view; provider retention schedules govern the underlying copies.
So the question "does the AI see my metadata?" has two separate answers: the model may not surface it unless asked, but the platform received it either way.
Can an AI Assistant Read EXIF From a Photo?
Yes, when it has a tool that can. Ask an assistant with code execution to run an EXIF parser on an uploaded photo and it will list GPS coordinates, camera serial number, and timestamps. Vision-only models describe the picture and typically do not report EXIF, but that is a limitation of the pipeline, not a privacy guarantee.
Try MetaClean — clean this kind of file in seconds.
Strip EXIF, GPS, author, and edit-history metadata from photos, PDFs, and Office documents right in your browser.
Clean a file now · See what gets removed · Step-by-step guides · Pricing
Provider Retention at a Glance
Policies change, so verify against the current documents before relying on them:
- OpenAI privacy policy and enterprise data controls
- Anthropic privacy policy
- Google Gemini privacy notice
Consumer tiers, business tiers and API access differ significantly on training use and retention. Business and API tiers generally exclude your content from training by default; consumer tiers may not.
What Leaks Most Often
| File you upload | Metadata that rides along |
|---|---|
| Phone photo | GPS coordinates, device serial, exact capture time |
| Scanned contract | Scanner model, network path, operator name |
| Work PDF | Author, company, producing software, revision times |
| DOCX draft | Author, "Last Modified By", track changes, template UNC path |
| Spreadsheet | Author, company, hidden defined names, print paths |
A resume uploaded for feedback commonly still carries the previous applicant's name from a reused template. A "quick question" screenshot of a dashboard commonly carries a logged-in username in the pixels.
The Safe Workflow
- Clean first. Run the file through MetaClean Pro before uploading anywhere.
- Redact visible identifiers in screenshots and scans.
- Use business or API tiers for work data, where training use is off by default.
- Prefer pasting text over uploading the original document when the model only needs the words.
- Keep regulated data out unless your organisation has a signed data-processing agreement covering the provider.
FAQ
Does ChatGPT store the photos I upload? The file is stored on OpenAI's servers under their retention policy. Deleting a conversation stops it appearing in your history; check the current policy for the underlying retention window.
Can Claude see the GPS location in my photo? Only if it parses the file with a tool. The bytes containing the coordinates are uploaded regardless, so strip them first.
Does Gemini train on my uploads? It depends on tier and settings. Workspace and API usage differ from the consumer app — check the linked notice for your account type.
Is it safer to paste text than upload a PDF? Yes. Pasting sends only the characters you paste; the PDF carries author, producer, timestamps and any hidden layers.
Does stripping metadata protect the file's contents? No. It removes hidden fields only. The visible content is still fully readable by the provider.
Bottom Line
Treat every AI upload as a transfer to a third party. Strip hidden fields first so the only thing you share is the content you meant to share. Clean a file before you upload it — and see the full metadata privacy checklist.
Try MetaClean Pro free — remove metadata from your files in seconds.