Decide what to use it for
This step is where you decide, before you type anything, which of your money questions an AI assistant is a reasonable place to take and which ones it is not. The sorting takes a few minutes and it shapes everything that follows: what you share, how you ask, and how hard you check.
Why it matters
An AI assistant will answer almost any money question you give it, in confident prose, whether or not it is the right tool for that question. Nothing in the reply tells you which kind of question you asked. That judgment has to be made by you, in advance.
The questions where it helps most are the ones where you need an explanation, a list of things to consider, a first draft, or a way to make sense of a document in front of you. The questions where it helps least are the ones that turn on your exact numbers, a current rule, or a decision you cannot undo. Mixing the two is how people end up acting on a fluent answer to a question the tool was never fit to settle.
Deciding first also protects the rest of the loop. If you know a question is one you will have to verify in full, you can ask it in a way that makes the checking easier. If you know it is one you should not be asking at all, you have saved yourself the whole exercise.
What actually happens
Write down the question you actually have
Start with the real question, in your own words, before you think about the tool. “Should I pay off the car loan or put the money in savings” is a different question from “how do people usually think about paying off a loan versus saving”, and the difference decides whether the tool belongs in the picture.
Sort each question into one of three groups
The first group is questions about how something works: what a term means, how a product is structured, what a letter is asking you to do, what options people in a situation like yours usually consider. An AI assistant is good at these, and the cost of a wrong answer is low because you will check it before acting anyway.
The second group is questions that turn on a current fact: a rate, a limit, a deadline, a rule that changed last year. The tool can help you find the words to search for and can explain what the fact means once you have it, but it cannot be your source for the fact itself. Those come from a primary source, and the tool’s version of them is a lead, never an answer.
The third group is decisions: whether to buy, sell, borrow, switch, sign, or stop. The tool can lay out the trade-offs and list the questions a careful person would ask. It cannot weigh them for you, because it does not know your situation, your tolerance for things going wrong, or what you have not told it. Decisions stay with you, and the larger or less reversible they are, the more likely they need a person as well.
Set the stakes before you start
For each question, note what happens if the answer is wrong and you act on it. A wrong explanation of a term costs you a few minutes. A wrong figure in a tax calculation can cost you money and a penalty. A wrong steer on a decision you cannot reverse can cost far more. The stakes decide how much checking the answer will need in the fifth step, and whether the question belongs in the loop at all.
Keep a short list of never-ask questions
Some questions should not go to an AI assistant at all, no matter how they are asked: anything that would need your account numbers, passwords, or identity documents to answer; anything time-critical where a wrong answer cannot be caught before it matters; anything where the honest answer is “ask a licensed person” and you already know it.
Questions to ask yourself
- Is this a question about how something works, a question about a current fact, or a decision?
- If the answer is wrong and I act on it, what does that cost me, and can I undo it?
- Do I already know this needs a person, and am I asking the tool to avoid that?
- What would I need to see, and from where, before I would trust the answer?
- Is there anything the tool would need from me that I should not be giving it?
Common mistakes
- Treating the tool as a source for a current rule or figure: it has no way to know whether what it tells you is still true.
- Asking the decision question when you meant to ask the how-it-works question, then acting on the decision.
- Letting the quality of the prose stand in for the quality of the answer; the two are unrelated.
- Skipping the sorting because the question feels small; small questions are where the habit forms.
Which knowledge areas apply
- How language models answer: why the tool answers every question with the same confidence, whichever group it belongs to.
- Information versus advice: where the line falls between an explanation you can use and a decision that needs a person.
- Primary sources: what counts as a source for a current fact, so you know what the tool cannot be.
For practitioners
A client who is doing this step on their own benefits most from the three-group sort, so give it to them in those words: how it works, current fact, decision. Tell them the tool is fine for the first group, a lead for the second, and a sounding board at most for the third. Ask them to bring you the questions they put in the third group rather than the answers they got. In their file, note that you had this conversation and which questions they said they were taking to a tool, so a later decision can be traced back to where it started.
Next step
Understand how it answers: what the tool is actually doing when it replies, and why that explains both where it helps and where it fails.