Are the outputs of generative AI tools like ChatGPT and Copilot considered records that have to be retained?

Short answer: maybe. Whether the output of a generative AI tool is a record does not depend on the fact that AI created it. It depends on what the content is, what it documents, and how your organization uses it. The same test you apply to an email, a memo, or a spreadsheet applies here.

Focus on content and use, not the tool

A record is information, in any format, that documents an organization’s activities, decisions, or transactions and that has continuing value as evidence. Recordkeeping standards are deliberately technology-neutral: they care about the role a piece of information plays, not the software that produced it.

So an AI-generated draft, summary, analysis, or chat exchange becomes a record when it is used to conduct business. For example:

• A prompt-and-response exchange relied on to make a decision or give advice
• AI-generated text incorporated into a report, policy, contract, or filing
• An AI summary used to brief leadership or document a meeting
• Output that triggers or supports an official action

Conversely, a quick AI query you discard, or a throwaway draft that documents nothing, is generally a non-record or transitory information.

Practical implications

If the output meets your definition of a record, it must be captured, classified, and retained according to the applicable retention schedule, just like any other record of that type. The retention period follows the function the record serves, not its AI origin.

A few points deserve attention with generative AI:

• Capture. Outputs often live in chat interfaces or scratch documents. If they are records, they must be moved into your recordkeeping system, not left in the tool.
• The prompt may matter. Where the input shaped a decision, the prompt and the response together may need to be retained for context.
• FOIA, legal holds, and discovery. If AI output qualifies as a record or is otherwise responsive, it can be subject to access requests and litigation holds.
• Accuracy and provenance. Because generative AI can produce errors, documenting how and why output was used supports reliability and accountability.

The safest practice is to update your records policies and AI-use guidance together, so staff know when AI output must be kept and how to capture it.

For more foundational concepts, see the Fundamentals topic hub.