What should I remove before sharing text, images, or documents with an AI tool?

Share the minimum information required for the task. Before uploading anything, ask whether the model needs the real identity, exact account number, full document, or complete conversation. Often it can do the work with names replaced, figures rounded, and irrelevant sections removed.

Remove direct identifiers such as full names, personal email addresses, phone numbers, home addresses, government identification numbers, passport details, signatures, bank information, passwords, access tokens, and authentication codes. Also look for indirect identifiers. A job title, rare medical condition, exact date, small location, or unique transaction can identify someone even when their name is absent.

Images need separate attention. Screenshots can expose browser tabs, notifications, profile photos, file paths, location details, customer records, or hidden content around the intended subject. Crop the image and inspect every visible area. Documents may contain comments, tracked changes, hidden sheets, revision history, embedded files, and metadata.

Do not paste secrets into a prompt. API keys, recovery codes, private keys, session cookies, and passwords should be treated as compromised if exposed. Revoke or rotate them rather than trusting deletion from a chat interface.

Check the service's privacy notice for retention, training, account access, subprocessors, deletion, and international transfers. For company or client information, follow the applicable contract and internal policy. Highly sensitive medical, legal, financial, employment, or children's data may require a specifically approved system rather than a consumer chat product.

After the task, delete unnecessary uploads and conversations where the service provides that control. Keep your own clean source document separate from the redacted copy. Privacy is easiest when minimization happens before upload, not after the information has already been transmitted.