Is it safe to send screenshots to an AI chatbot?

What a screenshot gives away

A screenshot is denser than it looks. Besides the thing you meant to ask about, a full-screen capture can show:

• Notifications that arrived as you pressed the key — names, message previews, one-time codes.
• Browser tabs and window titles, which often name customers, projects or documents.
• File paths and user names in title bars and terminals.
• Tokens and keys in terminals, configuration screens and URLs — reset links and sign-in codes travel in URLs.
• Other people's data: email addresses, phone numbers, faces, account numbers.

What happens after you upload it

Every major AI chat stores an uploaded image with the conversation it belongs to, under your account, until you delete it — and often for a period after that. Whether conversations also train future models depends on the service. As of October 2026: ChatGPT does by default on personal plans, until you turn off Improve the model for everyone in its data controls; Gemini does while Keep Activity is on, which it is by default for adults, and human reviewers may read some chats, keeping reviewed ones for up to three years; Claude asks each user to choose; and Microsoft says its current Copilot app does not train its foundation models on your prompts and files. Business and enterprise plans usually have stricter terms than free and personal ones.

These settings change. Before uploading anything sensitive, open the data or privacy controls of the service you use and read what they say today, rather than relying on what you remember.

A five-step checklist before you upload

1. Capture a region, not the whole screen. Most questions are about one dialog, one table or one paragraph.
2. Look for notifications, tabs, names and paths in the capture, and crop them out.
3. Redact keys, tokens, passwords, card numbers and other people's details — and redact properly, not with a light blur.
4. Check the service's training and history settings, or use a temporary chat if it has one.
5. If a credential was visible in anything you already uploaded, rotate it.

How ScreenFeel handles a screenshot

ScreenFeel was built so this checklist is mostly done for you:

• Local by default. Every capture, edit and answer is stored in a library on your PC. There is no server-side gallery.
• One image, only when you ask. Nothing watches your screen. Running a lens sends that one image once, with its window title and dimensions and nothing else; any web address in the title is stripped of the parts where reset tokens and sign-in codes live.
• A model that keeps nothing. Lenses run on private models hosted under zero data retention: the image is read for your answer and not stored, logged or used for training.
• A local tripwire. Before a lens runs, text ScreenFeel has already read is checked on your PC, with no network call, for key-, token- and card-shaped strings, and the Find secrets lens boxes everything else worth hiding.

When to keep it entirely offline

If a screenshot must not leave your computer at all — regulated data, unreleased financials, a client's systems — use tools that run locally: Snipping Tool's Text actions and PowerToys Text Extractor read text on your PC, and ScreenFeel's capture, editor, redaction and search work without sending anything. Ask an AI only about a redacted copy, or not at all.

Questions people also ask