Choose what you share
This step happens before you ask, in the few seconds between knowing your question and typing it. It is where you decide what the tool needs to know to answer well, what it does not need, and what you will never give it. Done once, properly, it becomes a reflex.
Why it matters
Anything you type into an AI assistant leaves your hands. Depending on the tool and its settings, it may be stored, reviewed by people, used to train future versions, or kept in a history that anyone with your login can read. The terms change, and most people have not read them. The safe assumption is that whatever you paste can come back somewhere you did not intend.
Money questions pull toward oversharing. The tool answers better with detail, so the instinct is to paste the statement, the letter, the whole picture. But the detail that makes an answer better is almost never the detail that identifies you. A loan question needs the rate, the balance and the term; it does not need the account number, your name, or the lender’s reference. Separating the two is this step’s whole job.
What actually happens
Decide what the question actually needs
Write the question, then list the facts the tool would need to answer it well. For most money questions that list is short: a few amounts, a rate, a date, a type of product, a country or state for anything rule-bound. Everything else is context you can describe in general terms. “A credit card with a balance of about four thousand” tells the tool what it needs; the card’s last four digits add nothing to the answer.
Strip out what identifies you
Go through what you were about to share and anonymise it. Replace names with roles, account numbers with nothing, exact figures with round ones if the exact figure is not what the question turns on. If you are pasting a document, paste the paragraph you have a question about, not the page with your address on it. The tool answers a round, anonymous version of your question just as well as the exact, identified one.
Keep a never-share list
Some things do not go into an AI assistant under any framing: account numbers and sort codes, passwords and one-time codes, identity numbers such as a tax or social security number, dates of birth, full addresses, scans of identity documents, and anything that would let a stranger reset a login. If a question seems to need one of these, the question belongs with the institution or a person, not the tool.
Watch what the tool reflects back
A reply can repeat what you gave it, sometimes in a form that is easier to lift than the original. If you shared personal data by mistake, delete the conversation where the tool allows it, change any credential that was exposed, and treat the exchange as if it had been sent to a stranger. The fix is the same as for any other leak; the mistake is only worse if it is ignored.
Know the tool’s own settings
Most assistants have settings that govern whether conversations are kept and whether they are used for training. Find them once, set them the way you want, and check them again when the tool updates. The settings do not make sharing identifying data safe; they lower the exposure for everything else.
Questions to ask yourself
- Which of the facts I am about to share actually change the answer?
- Could a stranger reading this reconstruct who I am or get into an account?
- Would the question work just as well with round numbers and no names?
- Is there anything here from my never-share list?
- Do I know what this tool does with what I type?
Common mistakes
- Pasting a whole statement or letter when one paragraph holds the question.
- Assuming a conversation is private because it feels like one.
- Sharing a credential “just this once” to get a faster answer.
- Treating the tool’s privacy settings as a licence to share anything.
Which knowledge areas apply
- Privacy and your data: what happens to what you type, in more depth, and how to read a tool’s data terms.
- Scams and impersonation: what the data you leak is worth to someone who wants to impersonate you.
- Information versus advice: the questions that need an institution or a person precisely because they need your real details.
For practitioners
A client who is doing this step on their own needs one rule more than a lecture: the tool gets the shape of the problem, never the identifiers. Give them the never-share list in those words and tell them to keep it where they type. If they have already pasted account details or documents into a tool, treat it as a disclosure: walk them through changing credentials and deleting the conversation, and note the date and what was exposed in their file. That note matters if a later fraud has to be traced.
Next step
Ask it well: how to frame the question so the answer is usable and its limits are visible.