Uploading an ordinary photo to ChatGPT, like a plant you want identified or a menu you can’t read, is low risk. The risk rises with what’s in the picture and what’s attached to it. Photos can carry your location in their metadata, show other people’s faces, or capture documents, screens and addresses in the background. And once uploaded, the image is stored with your chat on the company’s servers, where it may be reviewed or used to improve the models unless you opt out.
Check a photo before you upload it, or use an AI that reads images on your phone without sending them anywhere.
What happens to a photo you upload to an AI chatbot? #
The image goes to the company’s servers along with your message and is saved in that conversation. After that, it’s covered by the same policies as your text:
- Training. OpenAI says it may use conversation content to improve its models unless you turn that off in Data Controls. Google uses Gemini activity when Keep Activity is on.
- Human review. Google says a subset of Gemini chats is reviewed by people, and reviewed chats are kept for up to three years. Its guidance is not to share anything confidential.
- Retention. Deleting a chat removes it from your history, but companies may keep copies for a period, and legal holds can override deletion.
For each service’s settings, see do humans read your AI chats.
What’s hiding in your photos? #
Before uploading, look at the photo the way a stranger would.
Location metadata #
Photos taken with your phone’s camera usually store GPS coordinates in the file. If you upload the original file, that location can go with it. It’s not always clear whether a given service strips it, so the safe move is to remove it yourself.
On iPhone: in the share sheet, tap Options at the top and turn off Location before sharing or saving the copy you’ll upload.
On Android: the option depends on your gallery app. Google Photos has location settings for shared photos, and some manufacturers’ galleries offer a “remove location data” toggle when sharing. Screenshots don’t contain GPS data, so a screenshot of the photo is a quick workaround.
Faces of other people #
A photo of your kid’s birthday party includes other people’s children. A picture of your desk might include a coworker. Crop to what you actually need the AI to see.
Documents and screens #
Mail on the counter shows your address. A laptop in the background shows your email. A prescription bottle shows your name and pharmacy. Crop or blur these before uploading.
IDs and financial documents #
Don’t upload photos of driver’s licenses, passports, credit cards, checks or Social Security cards to a cloud chatbot. If you need help reading a form, crop out the identifying parts. Our guide to sharing financial info with AI has more on this.
A quick checklist before you upload #
- Is the location data off?
- Is anyone in the frame who didn’t agree to it?
- Can you read any name, address, account number or barcode?
- Is there a screen, document or license plate in the background?
- Would you be fine with a human reviewer seeing this image?
If you answered “yes” to any of the first four or “no” to the last, crop, edit or skip the upload.
Can AI identify people in photos? #
Mainstream chatbots are designed not to identify real people from their faces, and the major providers have policies against it. What an AI can do is describe what’s visible: clothing, setting, signs, text. That’s still a lot of information about a person or place, which is why cropping matters.
How to ask AI about a photo without uploading it #
Some AI apps can understand images on the phone itself. A vision-capable language model reads the picture locally, so nothing is sent anywhere.
Personal LLM does this. Every model in its catalog can take an image after you add that model’s image-support file, a one-time download of 195 MB to 1 GB. After that you can point it at a menu, a sign in a language you don’t read, a whiteboard or a form and ask questions, even in airplane mode. The photos you attach stay on the phone, and the app has no server to send them to.
What local vision models are good at:
- Reading and translating text in signs, menus and labels
- Describing what’s in a picture
- Explaining a chart, diagram or screenshot
- Pulling key fields out of a form or receipt
What they’re weaker at: fine detail in busy images, small or blurry text, and identifying specific species or products with confidence. Our guide to asking AI about a photo offline covers what to expect.
Built-in phone features and photo privacy #
Your phone’s own AI has its own rules. Apple’s visual intelligence and Apple Intelligence handle many requests on the device, but some are passed to Apple’s Private Cloud Compute, and requests handed to ChatGPT or Google go to those companies. Apple says it asks before sending anything to ChatGPT. On Android, Gemini Nano features run on the device on supported phones, while the Gemini app sends images to Google.
We explain which requests go where in is Apple Intelligence private.
Frequently asked questions #
Does ChatGPT keep the photos I upload? #
Photos are saved as part of the conversation on OpenAI’s servers, just like your text. Deleting the chat removes it from your history, and OpenAI says it may use content for training unless you turn that off. Check Data Controls in the app for the current options.
Can ChatGPT see where a photo was taken? #
It can if the location is in the file’s metadata and that data reaches the model, or if the scene itself gives the place away. Turn off location when sharing, or upload a screenshot, which carries no GPS data.
Is it safe to upload a photo of my ID to an AI? #
No. IDs contain everything needed for identity theft. If you need help with an ID-related form, describe the question in text or crop the photo so only the relevant part shows.
Are photos of my kids safe to upload to AI? #
Be cautious. They become part of your chat history on the company’s servers. For a quick question, crop to the part that matters or use an on-device app so the photo never leaves your phone.
Can an offline AI app read photos? #
Yes, if it runs a vision-capable model. Apps like Personal LLM, Google AI Edge Gallery and Locally AI can answer questions about images entirely on the phone.