Redact the name before you paste an HR case into an AI tool

Before you paste a note about a real employee into Copilot, Claude, or any AI tool to help draft an investigation summary, a review, or a difficult-conversation script, swap the name for "the employee" and any identifying detail (team, exact role, unique project) for something generic. Do the same for anyone else mentioned - a manager, a witness, a complainant.

This isn't about distrusting the tool. It's that you don't control, and usually haven't checked, what a given AI provider's terms say about retaining or training on what you send it, and an HR case is about as sensitive as personal data gets - it can touch performance, conduct, health, or a grievance. Treating it the same way you'd treat any other confidential HR document, rather than assuming a chat window is private the way a locked filing cabinet is, costs you nothing.

The practical habit: draft with placeholders, get the structure and wording right, then do the final find-and-replace back to the real name yourself, in your own document, after the AI tool is out of the loop. If you need the model to reason about something that depends on identity - seniority, tenure, a protected characteristic that's genuinely relevant to the analysis - describe it generically ("a senior employee with eight years' service") rather than naming the person it applies to.

The one place this habit pays for itself twice over: if the case ever does become a formal grievance, tribunal claim, or subject access request, you want to be certain you know everywhere that person's data went. A rough note you can't account for is a worse problem than the extra ten seconds redaction takes.