A note before the method. This works identically in ChatGPT, Claude and Gemini, because it changes what you give the tool rather than which tool you use. Company policy still governs whether and where you use AI at all, and nothing here overrides it.
There is a moment that stops thousands of professionals a day. The text is copied, the cursor is blinking in the chat box, and the hand hesitates, because the email contains a client's name, or the situation involves a colleague, or the numbers are the company's, and some well-calibrated instinct says this should not go in.
The instinct is right. The everyday versions of these tools can learn from what you type and keep it for a while, which is exactly why careful firms restrict them. What almost nobody teaches is the move that comes after the hesitation, the one that gets you the full value of the tool while giving it nothing worth protecting.
The move has a name here, because it deserves one. It is the hypothetical method, it takes about thirty seconds, and it is the single skill that separates people who use AI safely at careful companies from people who either take risks or give up.
The short answer: instead of pasting the real situation, you rebuild it as a hypothetical that keeps the structure and drops the identity. Strip who, keep what, restore the specifics yourself afterwards. The tool never needed the names, the company or the real numbers to help with the wording, the options or the logic, which is nearly always what you actually wanted.
The method
Step one, strip who. Remove every detail that could identify a person, a company or a deal: names, employers, locations, amounts, dates that narrow it down. You are not deleting the situation, only its address.
Step two, keep what. Preserve the shape that makes the problem itself: the relationship, the tension, the constraint, what you need to happen. This is the part the tool actually works with, and it survives anonymisation completely.
Step three, restore the specifics yourself. Take what comes back, the draft, the structure, the options, and put the real names, numbers and details in on your side of the screen, where they never left. The pre-send check covers the working version of that final pass.
A worked example
The real situation: your client Meridian Logistics is six weeks late paying a large invoice, their finance director has stopped replying, the relationship is worth keeping, and you need a firmer email than the last two chases without burning anything.
The hypothetical version: "A small services firm has a client six weeks overdue on a significant invoice. The client's finance contact has gone quiet after two polite chases. The relationship matters and the firm wants paying without souring it. Draft a firm but warm third email, under 150 words, that creates a reason to reply this week."
What comes back is a complete, usable draft, because every ingredient that shapes good wording survived the translation: the history, the tension, the goal, the length. You swap in the name, the amount and the date in twenty seconds, and the sensitive version of the situation has never existed anywhere but your screen.
That is the whole method, and it generalises to an employee situation, a deal, a dispute, a restructure, anything whose difficulty lives in the dynamics rather than the identities.
What survives the translation, and what does not
| Works fully as a hypothetical | Needs the real thing, so needs a different route |
|---|---|
| Wording and tone for any difficult message | Summarising a specific real document, which requires pasting it, so use a sanctioned tool or do not |
| Options, structure and plans for a situation | Checking real numbers, which the tool cannot do reliably anyway |
| Rehearsing a conversation before having it | Anything where the identifying detail is the substance, like a named legal or tax question, which belongs with a professional |
| Understanding a concept, rule or pattern in general terms | Current, situation-specific facts, which need a live source rather than a model |
The left column is, for most professionals, the large majority of what they wanted AI for in the first place, which is the quiet good news of the whole subject. The right column is where a company-sanctioned tool with proper data protections earns its existence, and the jobs that should never go near a consumer tool have their own full treatment. Where no sanctioned tool exists, the honest answer is that the task stays off AI entirely.
Why this beats the alternatives
The other two responses to the hesitation both cost more than they look. Pasting anyway trades a thirty-second rewrite for a risk that cannot be recalled, since what enters a consumer tool is out of your hands from that moment. Giving up trades away the tool's entire value over a problem the method solves in half a minute.
The hypothetical method is not a compromise between safety and usefulness. For the wording-and-thinking work that fills professional life, it is both at once. Once it becomes reflex, the hesitation at the chat box disappears, replaced by a translation your hands do automatically.
If your workplace blocks the tools outright, the block governs, but the pattern is usually about work data rather than about you, and professionals in tightly governed fields have been using exactly this translation habit for as long as the tools have existed.
Not sure where to start?
The method needs a tool. The finder picks it.
The free finder tells you which AI to open for the task in front of you, in about 60 seconds, and the full "AI, sorted." reference lands in your inbox.
Try the free finder →One question in, one answer out.
Clair helps non-technical professionals know when to trust their AI, when to check it, and when to skip it.