The tension most people feel
You know AI can help you research faster. You've seen it summarise complex information, explain industry trends, or surface relevant data in seconds. But when you sit down to use it for real work, you hesitate. The research involves a client. Or a pending deal. Or financial information. Or something that just feels like it shouldn't leave the building.
That hesitation is reasonable. A 2026 survey found that 43% of IT leaders cite data exposure as their top concern around workplace AI use. But the hesitation often stops people from using AI for research that would be entirely fine — because they've never worked out where the actual line is.
Where the actual line is
The rule is simpler than most people think: AI is fine for researching things that are public, general, or don't require you to share confidential information to get a useful answer.
Research AI handles well includes: understanding how an industry or regulation works, finding typical benchmarks for a task or metric, understanding what questions to ask before a meeting, summarising publicly available information about a market or competitor, explaining concepts or terminology before you use them in a presentation.
Research that carries real risk: anything where getting a useful answer requires you to paste in client names, deal values, HR data, financial figures, legal agreements, personal details, or information marked confidential in your organisation.
The distinction matters because most research tasks don't actually require you to share the sensitive details to get a useful answer. You just need the framework, the context, or the background — and that can be described without naming anyone.
The anonymisation technique
When you need AI to help with something that involves specific confidential details, replace the specifics with placeholders before pasting anything in. This gives AI the structure it needs to help without exposing the actual data.
For example: instead of pasting "Our client Thornton Manufacturing signed a three-year contract worth $4.2 million for equipment maintenance," write: "A client in the manufacturing sector signed a multi-year maintenance contract. I need to understand what renewal terms are standard in this type of agreement."
You get the same quality of research help. The client name, the dollar figure, and the specific deal are nowhere in the conversation.
Types of research that are always safe
Industry knowledge and market context: how a particular sector works, what regulations apply, what the standard terms mean. This information is public and asking AI to explain it carries no privacy risk.
How to do something: how to structure a proposal, how to run a particular type of analysis, how to approach a negotiation. You're asking for a method, not sharing data.
Checking your own thinking: "I'm planning to approach this problem this way — what am I missing?" You can describe your approach without naming the client or sharing the specifics.
Think of the last piece of research you did manually that took longer than it should have. Could you ask an AI the same question without sharing any confidential details? If yes, try it now. If you'd need to share confidential details to get a useful answer, use the anonymisation technique above — replace names and figures with placeholders and ask for the general framework instead of the specific answer.
When in doubt, check your organisation's policy
Many larger organisations now have an approved AI tool — often Microsoft Copilot within the M365 environment — where data stays inside the company's compliance boundary. If your organisation has one, use it for anything involving confidential information. Save personal AI tools for the general research, writing, and thinking tasks where no sensitive data is involved.
If you're unsure whether your organisation has a policy, ask IT or HR directly. It's a five-minute conversation that clarifies the rules and takes the uncertainty away.