Privacy Checklist: What to Check Before Uploading Files to AI

Uploading a contract, a spreadsheet of customers or an unpublished manuscript to an AI tool takes a second. Understanding where that file goes takes longer, but it is worth the effort. This guide explains the questions to ask before you upload anything sensitive, shows how several major tools describe their settings in their own documentation (as of October 2026), and ends with a checklist you can use every time.

This article is general information, not legal advice. If you handle regulated data such as health, financial or children's information, check your legal obligations and your organisation's policies first.

Why AI uploads deserve extra care

When you upload a file to an AI service, it is processed on the provider's servers. Depending on the product and your settings, your content may be stored in your chat history, retained for safety monitoring, reviewed by people in some circumstances, or used to improve future models. None of this is necessarily bad, but you should choose it knowingly.

The seven questions to ask

1. Will my content be used to train models?

This is the most common concern, and the answer varies by product and plan:

  • ChatGPT: OpenAI's help centre explains that you can turn off "Improve the model for everyone" under Settings, Data controls, so that new conversations are not used for training while remaining in your history.
  • Claude: Anthropic's pricing comparison lists model training as "opt-out" on Free, Pro and Max, and "none by default" on Team and Enterprise.
  • Gemini: Google's Gemini Apps Privacy Hub says that with "Keep Activity" turned off, future chats are not saved to your activity and are not used to train Google's models, unless you submit feedback.

Business and enterprise plans of most major tools default to not training on customer data. Consumer plans often require you to opt out.

2. How long is it kept, even if I delete it?

Deleting a chat from your history does not always mean it disappears immediately. Examples from official documentation:

  • OpenAI says Temporary Chats do not appear in history or train models, but may be retained for up to 30 days for safety and abuse monitoring.
  • Google says that even with Keep Activity off, chats are retained with your account for up to 72 hours to provide the service and protect users.

Look for a retention section in the privacy policy or help centre, and note any exceptions for legal or safety reasons.

3. Can humans read it?

Some services use human reviewers for quality or safety. Google's Gemini privacy hub, for example, states that human reviewers may review chats for safety purposes, and that chats which have been reviewed may be retained for up to three years even if you delete your activity. If a document must never be read by anyone outside your organisation, a consumer AI tool may not be the right place for it.

4. Is it public or shared by default?

Some creative tools publish outputs to a community gallery. Midjourney, for instance, lists Stealth Mode (which keeps images and videos private) only on its Pro and Mega plans. Meeting assistants can share notes with all attendees, and shared chat links can be viewed by anyone with the link. Check default visibility before you upload.

5. Who else processes it?

Many AI apps are built on models from another company. Your file may pass through the app's servers and a model provider's servers. Good vendors name their subprocessors and explain their terms; Notion, for example, states that its LLM providers use zero data retention for Enterprise plan workspaces. If a small app cannot tell you which model provider it uses, be cautious.

6. What does the business plan change?

For organisations, business tiers typically add contractual protections: no training by default, admin controls, single sign-on, audit logs, custom retention and sometimes data residency. Claude's Enterprise plan lists custom data retention controls and audit logs; Cursor's Teams plan lists a team-wide privacy mode; Descript says its Enterprise plan offers training opt-out and custom retention, and that it is SOC 2 Type II compliant. If you upload client or employee data, these features are often the real reason to pay.

7. Can I get my data out, and delete my account?

Check for export options and a clear account-deletion process. Descript's FAQ, for example, says you can delete your account and permanently wipe your data from its servers.

Special cases

Meeting recorders and transcription tools

AI note-takers raise a consent issue: you are recording other people. Otter's help centre explains how to stop its Notetaker from automatically joining calendar meetings, how to choose which meetings it records, and how to send pre-meeting recording notifications. Recording laws differ by country and, in some places, require everyone's consent. Always tell participants before recording.

Voice and face cloning

Uploading voice samples or photos to create a clone is uploading biometric-like data. Use only your own likeness or that of people who have given explicit, documented consent, and understand how the vendor stores the model.

Code

Source code can contain secrets such as API keys and database passwords. Remove them before sharing code with any assistant and check whether your employer allows the tool.

Browser extensions and agents

AI extensions that read web pages, or agents that act in your browser, can see whatever is on screen, including email and banking pages. Grant the minimum permissions and remove extensions you no longer use.

How to reduce risk without giving up AI

  1. Redact first. Replace names, account numbers and addresses with placeholders before uploading.
  2. Upload extracts, not whole files. Paste the relevant section rather than the full contract.
  3. Use temporary or incognito modes for one-off sensitive questions. Claude lists incognito chats on all plans; ChatGPT offers Temporary Chat.
  4. Separate personal and work accounts. Use the plan your organisation has approved for work data.
  5. Review settings after updates. Privacy options and defaults change; re-check them every few months.

Pre-upload checklist

  • Do I have the right to share this file (client confidentiality, NDA, employer policy)?
  • Does it contain personal data about others, and is uploading it lawful where I live?
  • Is model training turned off for this account, or does my plan exclude it by default?
  • Do I know the retention period and whether humans may review content?
  • Is the output private by default?
  • Have I removed passwords, keys and unnecessary personal details?
  • Would I be comfortable if this file were exposed in a data breach?

If you answer "no" or "I don't know" to any of these, pause and investigate before uploading.

Red flags in an AI tool's privacy terms

  • No privacy policy, or one that does not mention AI or uploaded content.
  • Claims of broad rights to use, sell or publish your content.
  • No way to delete data or close your account.
  • No information about who the company is or where it is based.
  • Vague answers to direct questions about training and retention.

Our guide on finding trustworthy AI tools covers other warning signs.

Bottom line

Before you upload, know whether your content trains models, how long it is kept, who can see it and how to delete it. Use business plans for work data, redact what you can, and get consent before recording or cloning anyone. Most tools offer reasonable controls; the risk comes from not using them. Explore privacy-conscious options in our AI chatbot, AI writing and AI coding categories.