The slow part of a client review meeting is not the conversation. It is the prep, assembling the agenda, summarizing where the account stands, and drafting talking points that explain what changed since you last met. That assembly is the kind of work AI is good at. The trap, and it is a serious one for a fiduciary, is that the same tool will happily add a recommendation no one asked for, project a return, or reach for performance and guarantee language your compliance team would reject. Coming from an advisor, that is not a clerical slip, it is a regulatory problem. So the rule here is strict: AI assembles from your own figures and your firm's approved language, and the recommendation stays with the fiduciary.
One ground rule before any of it: client financial data is sensitive. Keep client identifiers out of a public chatbot, test with de-identified figures, and for real client work use a private tool that keeps the data inside your own systems. More on that at the end.
The honest picture
AI is genuinely useful for assembling review prep from what you already know about the account:
- What AI does well today: take the figures you provide, balances, contributions, allocation, and what changed since the last review, and lay them into a clean meeting agenda, a plain account-status summary, and a draft set of talking points. It turns your scattered notes into something organized you can walk into the meeting with. It compresses the prep, not the judgment.
- What AI does not do: advise the client. It cannot recommend a buy, sell, or allocation change, cannot project or promise a return, cannot make a suitability or fiduciary call, and cannot decide what is right for this client. The fiduciary owns every recommendation and the suitability call. AI assembles from your own data and your firm's approved language. It must never make a recommendation or add a projection that is not yours.
The right way to think about it: AI is a fast clerk organizing what you already know, not an advisor. The recommendation is yours. The assembly is what you hand off.
The line: it will make a recommendation you have to stand behind
The specific failure to watch for is helpfulness, and for a fiduciary it is the dangerous one:
- Recommending. Ask AI to "prep talking points for the review" and it will often suggest rebalancing, trimming a position, or shifting an allocation. That is a recommendation, and it is yours to make, not the model's. Never let it propose a buy, sell, or allocation change.
- Projecting. It will reach for "this should grow to" or "you can expect" language. A projected or promised return on a client document is exactly what compliance exists to stop. It must project no return.
- Suitability. It cannot judge what is appropriate for this client's goals, risk tolerance, or situation. That is the fiduciary call, and it stays with you.
- Compliance language. It will use performance, comparison, or guarantee phrasing that your firm's approved language never would. Anything it writes is a draft until you and compliance clear it.
The fix is one discipline applied without exception: AI uses only the figures you paste and the firm-approved language you give it, makes no recommendation, projects no return, and flags anything that needs your judgment or compliance review rather than deciding it.
The setup that keeps the recommendation with the fiduciary
Three habits make AI safe for review prep, and the prompts below build them in:
- Give it only your figures and approved language. Hand it the account figures and what changed, plus your firm's approved phrasing for anything sensitive. Tell it to use only these and to add no number or claim that is not in what you gave it.
- Make it flag, not decide. If a talking point would require a recommendation, a projection, or a suitability call, it lists that under "needs your judgment or compliance review" and you decide. It never makes the call.
- Keep the data yours. De-identified figures for testing in a public tool; real client data only in a tool that stays inside your own systems.
How to test it on your own work
Do not trust a polished demo, including this one. Pull two or three of your own recent reviews, de-identified, with the source figures and your firm's approved language, a timer, and the prompts below. Rate each output 1 to 5 on usefulness and accuracy, and compare the time against how you prep today. Keep what wins.
Paste-ready prompts
Copy these as written. Bracketed text is what you swap per client.
Test 1: Build the agenda and account-status summary (text model)
I am giving you figures for a client review meeting: current balances,
contributions since the last review, current allocation, and what changed.
Build a meeting agenda and a plain account-status summary from these figures.
Rules:
- Use only the figures and the firm-approved language I paste. Make no
recommendation, do not suggest a buy, sell, or allocation change.
- Project no return and add no number or claim that is not in what I gave you.
- Flag anything that needs my judgment or compliance review under a heading
"Needs my judgment or compliance review" rather than deciding it.
Figures: [paste]
Firm-approved language: [paste]
Watch for: did it slip in a recommendation or a projection? Anything it suggested that I did not provide is mine to decide, not the model's. Move it to the judgment list or cut it.
Test 2: Plain-language talking points for what changed (text model)
Using only the figures and firm-approved phrasing I provide, draft plain-language
talking points explaining what changed in this account since the last review.
Rules:
- Describe only what the figures show. Make no recommendation and project no
return.
- Use only my firm's approved phrasing for anything sensitive. Do not introduce
performance, comparison, or guarantee language.
- Flag anything that needs my judgment or compliance review rather than writing
around it.
Figures and approved phrasing: [paste]
Watch for: did it explain what changed without telling the client what to do about it, and did it stay inside the approved phrasing?
Test 3: Turn my notes into a client-facing recap (text model)
Here are my own notes from the review. Turn them into a clean client-facing recap.
Rules:
- Use only what is in my notes and the firm-approved language I paste.
- Make no recommendation, project no return, give no guarantee, and add no
advice the notes do not contain.
- If a note implies a recommendation or projection, do not write it as one. Flag
it under "Needs my judgment or compliance review" instead.
My notes: [paste]
Firm-approved language: [paste]
Watch for: does the recap stay a record of what was discussed, or does it harden a stray note into a recommendation or a forward-looking promise?
Test 4: Compliance and recommendation audit (text model)
Compare this draft review document against the figures and firm-approved language
I provided. For each problem, quote the exact line.
1. Any recommendation: a suggested buy, sell, allocation change, or course of
action.
2. Any projected, promised, or expected return.
3. Any guarantee, performance, or comparison language not in the approved
language I gave you.
4. Any number or claim that is not in the source figures.
Do not fix anything. Only flag it for my compliance review. Treat any added
recommendation, projection, or claim as a defect.
Source figures and approved language: [paste]
Draft document: [paste]
Watch for: does it catch the line that quietly recommended a change or projected a return? Run it on a recap you already sent.
What success looks like, and where it could go
If your own testing shows real time savings, the next step is a small pilot: run a week of reviews through the prompts and measure the prep time. If that holds up, the natural next step is a simple agent, running on your firm's own cloud, that you use in plain language. The most useful version pulls only the advisor's own figures and the firm-approved language, drafts the agenda, the status summary, and the talking points, cites the source figure or approved line behind each statement, and leaves anything that needs a judgment or compliance call visibly flagged, with compliance review still in the loop. You make every recommendation and sign off. Because it runs in your own systems, client data never leaves for a public chatbot.
The principle holds the whole way through: AI assembles what your figures and your approved language already say. It does not make the recommendation, project the return, or judge suitability. The fiduciary owns every recommendation and the compliance call. Used this way it speeds the prep without ever putting advice in front of a client that you did not make.
This is general information about workflow tools, not investment advice. Nothing here is a recommendation to buy, sell, or hold any security.
Want a straight answer for your advisory practice?
I build practical AI and custom software for businesses, on Google Cloud. If you want a second set of eyes on how AI could fit your client-review workflow, or on a tool you are considering buying, tell me what you are working with. No pitch, just a straight answer.