A client sits down for an annual review and opens with a list. Not their list. "I asked an AI what I should ask my financial advisor before this meeting."
The first wave of this was clients arriving with AI-generated answers: a savings rate, an allocation, a retirement date. That pattern had a useful response, covered in [an earlier article in this series](/blog/clients-arrive-with-ai-answers/): before evaluating the answer, ask to see the prompt, because the conclusion depends on what the system was told.
The new pattern is different in one important way. The client is not bringing an answer to check. They are bringing questions to ask you. The AI has moved from the client's analyst to the client's diligence partner.
The same response still applies, for a different reason.
The lists are usually reasonable
The question lists clients bring are generally sensible: how are you compensated, how do you decide when to rebalance, how do my accounts work together for taxes, what happens to my relationship with the firm if you retire.
These are questions an advisor should welcome. Most of them are questions a well-prepared client might have asked twenty years ago after reading a magazine column. The technology has not invented new diligence. It has lowered the cost of preparing for it.
Treat the list itself as good news. A client who prepares for a review meeting is engaged in the relationship, and every question on the list is one the advisor can answer better in conversation than a general-purpose system can answer in the abstract.
Answer the questions plainly. Do not compete with the list, critique it, or treat it as a challenge to the relationship. The list is not the interesting part of the meeting.
The prompt is the interesting part
A question list is an output. Like any AI output, it reflects the request that produced it. The 2026 working paper by Choukhmane, de Silva, Lin, and Akuzawa discussed earlier in this series found that AI-generated financial guidance varied with the information and language in the prompts that produced it. The same dependence holds here, with a twist that favors the advisor: the request is where the client put their actual concern.
Consider the difference between these prompts, each of which produces a similar-looking list:
- "What should I ask my financial advisor at our annual review?"
- "What should I ask my financial advisor to decide whether I should stay with them?"
- "What should I ask my financial advisor about protecting my children's inheritance?"
The first is preparation. The second is a client who is quietly evaluating the relationship and has not said so. The third is a client trying to start an estate conversation they have not found a way to open, possibly for years.
The advisor who only answers the questions has a fine meeting. The advisor who understands which prompt produced them is having the meeting the client actually came for.
So, at the natural moment, ask some version of the question this series keeps returning to: "That's a good list. What did you ask it?"
Ask it with curiosity rather than suspicion. If the client is comfortable sharing the exchange, the phrasing, follow-ups, and context they provided will often say more about their current concerns than the list does. If they would rather describe it than show it, that works too. The goal is not to inspect the client's use of AI. It is to recover the concern behind the preparation.
What to do with what you learn
If the prompt reveals something the list did not, that becomes the agenda.
A client who asked an evaluation-flavored question deserves an advisor who addresses the relationship directly rather than performing well on the list and hoping. A client whose prompt circled an inheritance concern deserves an advisor who opens the estate conversation rather than waiting for the client to find the words a general-purpose model could not supply either.
Afterward, preserve the substance in the firm's approved recordkeeping process: what the client raised, what concern sat behind it, what was discussed, and what was agreed as the next step. The point is continuity. If the concern surfaced obliquely once, it will surface again, and the record should let the next conversation [start from understanding rather than reconstruction](/blog/your-firm-has-more-client-information-than-client-context/).
Keep the review inside firm-approved workflows
A client's AI conversation can contain a dense mix of financial, family, and personal detail, some of which the client may not realize they included. If a client shares a prompt or a transcript, review and store it only in systems and workflows the firm has approved for client information.
Removing a name does not automatically make the surrounding information appropriate for an unapproved tool. Questions about records, supervision, or the use of any particular system are firm- and deployment-specific, and advisors should follow their firm's policies and involve the appropriate compliance, privacy, or security professionals when needed.
The pattern worth noticing
Across this series, the constant is not the technology. It is where the useful information sits.
When a client arrives with an AI answer, the answer is less informative than the inputs behind it. When a client arrives with an AI question list, the list is less informative than the request that produced it. In both cases the model supplies the generic artifact, and the client supplies the specific detail that matters, usually without noticing they have done so.
The answers live in the model. The worry lives in the prompt. The advisor who asks to see it is the one having the real meeting.