Before you paste a screenshot into ChatGPT

Last updated: August 6, 2026

A screenshot you upload to an AI chat carries everything visible in the frame, not just the part you meant to ask about — the notification banner at the top, the other browser tabs, the contact name on a message thread, the account number in a spreadsheet column you weren't asking about. Before uploading, redact anything in the image you wouldn't want stored or used outside the conversation, because once it's uploaded, you're relying on the AI provider's data policy, not your own control over the file.

What actually leaks when you paste a screenshot into an AI chat?

Usually it's not the thing you meant to ask about — it's everything else in the frame that you stopped noticing because you look at it every day. Common examples:

• A phone notification banner across the top of the screen, showing a message preview from someone else.
• Other open browser tabs or app icons visible in a window screenshot.
• A contact's name, phone number, or photo in a messaging screenshot you're asking about for an unrelated reason.
• An account number, balance, or transaction history in a spreadsheet or banking app screenshot, when you only meant to ask about one cell.
• An email signature with a direct phone number or address, included automatically at the bottom of a forwarded message.

None of this is intentional oversharing. It's the normal result of a screenshot capturing a whole rectangle of screen instead of the one fact you actually wanted to ask about.

Does the AI company keep the image after I upload it?

It depends on the product and your account settings, and it's worth checking rather than assuming. As one example, ChatGPT's consumer plans — including the paid Plus tier — use conversation content, including uploaded images, to help train future models by default; turning that off requires going into data controls and disabling it yourself. Business, Enterprise, and API-based products generally handle this differently. The specifics vary by provider and change over time, so the safe assumption is: unless you've explicitly confirmed otherwise for the product you're using right now, treat an upload as something that may be retained.

Does asking the AI to "ignore the personal info" help?

No. An instruction like that changes what the model chooses to talk about in its reply — it doesn't stop the image from being processed, and it doesn't affect whether the platform retains or trains on it. The image was already uploaded and read in full before the model ever saw your instruction.

What should I actually do before uploading a screenshot to an LLM?

1. Redact anything identifying that isn't relevant to your question — names, phone numbers, addresses, account numbers — with a permanent, non-reversible style. See why blurred screenshots can be unredacted before choosing blur over solid fill.
2. Crop out parts of the screen that aren't relevant, like notification banners or adjacent tabs.
3. Check the specific product's data-retention or training toggle if you're not sure what its default is.
4. When the content is simple, paste text instead of a screenshot — it's easier to control exactly what's included.

Is pasting text different from pasting a screenshot?

Yes, in one important way: text lets you choose exactly what to include. You can copy the three lines that matter and leave out the rest. A screenshot captures a rectangle of your screen indiscriminately, including whatever happened to be there. If the content you're asking about is plain text — an error message, a paragraph, a table — pasting the text is usually both easier to redact and easier to control than a screenshot.