Wealth management teams write market overviews and client recaps again and again, and AI tools can speed up the first draft. They can also produce confident, persuasive sentences that a compliance reviewer would never approve. This guide shows how to write prompts that push a model toward careful, sourced and balanced drafts, and where human review has to take over.
About this guide. PromptByJob publishes prompt templates and writing guides. The author is not a financial adviser, lawyer or compliance professional. The prompts below are starting points, and they are not presented as tested, validated or approved by any compliance department or regulator. Nothing here is legal, regulatory or investment advice. Any AI assisted client material should go through your firm’s own supervision and approval process before it is used.
What a Prompt Can and Cannot Do
A well built prompt can reduce certain wording problems in a first draft: predictions, guarantees, unsourced numbers and opinions that read like facts. It cannot make a document compliant. Whether a communication meets your firm’s obligations is a decision for the people responsible for compliance, based on the rules that apply to your firm type and how the document will be used. Think of the prompt as a way to hand your reviewers a cleaner draft, not as a substitute for them.
Why Generic Prompts Create Problems
AI models are built to produce fluent, confident text. Wealth management writing needs the opposite default: balanced, sourced and careful. If you ask a generic prompt to “write a market outlook for clients,” you may get sentences like these:
- “This sector will likely outperform.”
- “Clients should expect strong returns next quarter.”
These two lines are illustrative examples of wording to watch for, not output from a recorded test. Both are forward looking and read as promises. Statements like them are the kind that compliance teams routinely question under standards for fair and balanced communication.
Regulators have also said that using AI does not change the underlying rules. FINRA’s Regulatory Notice 24-09 reminds member firms that its rules and the securities laws apply to generative AI just as they apply to other technology, including the content standards for communications with the public. For SEC registered investment advisers, the SEC’s marketing rule guidance and the text of Rule 206(4)-1 are the starting points. Which rules cover a particular market overview or recap depends on your firm and on how the document is used, so confirm that with your compliance team.
Three Design Principles for the Prompt
The two prompts below are built on three ideas. Each one targets a specific weakness in how general purpose models write about finance.
1. Define the Role and the Boundaries
Open the prompt by saying what the model may and may not do. A boundary paragraph might read like this:
You are drafting a market overview for review by an investment adviser’s compliance team. Do not predict future performance, guarantee returns, or imply certainty about market direction. Do not state a statistic unless it appears in the source material, and give its source and date. If a figure is missing, insert a bracketed placeholder such as [S&P 500 quarterly return, source and date].
Plain, specific limits work better than general instructions such as “be compliant,” because the model has no reliable way to know what your firm considers compliant.
2. Force Source Discipline
Models often state numbers with no source, such as “the S&P 500 gained 4 percent this quarter.” That looks harmless until a reviewer asks where the number came from. Tell the model to use only the data you supply and to use a bracketed placeholder for anything else. The person publishing the document then verifies every figure against the original source before it goes out.
3. Separate Fact From Opinion
Models blur reporting and opinion unless told not to. Require two labeled sections in every draft:
- A factual recap that reports only what happened, in past tense, with sourced figures.
- A house view that is clearly labeled as the firm’s interpretation, using softer language such as “we believe” or “our view suggests” instead of definitive claims.
This structure helps reviewers because they know where to look for opinion language, and it helps readers because they can tell reported facts from the adviser’s judgment.
Prompt 1: Two Step Market Overview (Compliance Lock and Recap Fill)
Most drafting prompts ask the model to write the document and police itself in the same pass. This prompt separates the two jobs. In Step 1 the model only flags risky phrases in your source material and proposes safer wording. It writes no client facing text until you reply “confirmed.” In Step 2 it drafts using only the approved language.
How to use it. Paste your notes or data into the source section and fill in the brackets. Read the Step 1 table and fix anything the model missed, then reply “confirmed.” Check every figure in the Step 2 draft against its original source before the draft goes to compliance. Do not paste client names, account numbers or other personal details into any AI tool your firm has not approved for that purpose.
How to use it. Paste your notes or data into the source section and fill in the brackets. Read the Step 1 table and fix anything the model missed, then reply “confirmed.” Check every figure in the Step 2 draft against its original source before the draft goes to compliance. Do not paste client names, account numbers or other personal details into any AI tool your firm has not approved for that purpose.
Prompt 2: Client Recap From Meeting Notes
A client recap carries a different risk from a market overview. It summarizes a real conversation with a real person, so the most important rule is that the recap must not invent or inflate what was said. A model that is asked simply to “summarize this meeting” may turn a general discussion into something that sounds like advice already given. This prompt keeps options and decisions apart and routes unclear notes to the adviser instead of letting the model guess.
A Hypothetical Example
The example below is constructed to show the pattern. It is not output from a recorded test.
Suppose the adviser’s notes say: “Client asked about bonds. Talked about holding individual bonds versus a bond fund. No decision. Client will think it over.”
A loose prompt such as “summarize the client meeting” could plausibly produce a line like this: “We encourage you to increase your bond exposure.” The notes never say that. The line turns a general discussion into a recommendation.
With Prompt 2, the draft should stay closer to the notes: “We discussed two ways to hold bonds, individual bonds and a bond fund. No decision was made, and you plan to think it over.” It should also list under “Items to confirm” that the notes do not say which maturities or amounts were discussed.
The point is not that the model will always behave this way. The point is that the prompt gives the adviser and the reviewer something specific to check.
A Review Process You Can Put Around the Prompts
A prompt is only the first layer. Here is one review flow a firm could adopt, to be adjusted to its own policies:
- Draft from an approved prompt template.
- The adviser or content owner checks every figure against its original source.
- Optionally run a keyword scan for risky words such as “guarantee,” “will outperform” and “promise.” A scan only catches the words on its list, so it supports human review and does not replace it.
- Compliance reviews the document, with particular attention to the opinion section, since that is where interpretation lives.
- Final approval happens before anything reaches a client.
- Keep the records your firm’s policy requires. Ask your compliance team whether that includes the prompt, the source material, the AI output and the final version.
Common Questions About Using These Prompts
Can AI draft in a specific adviser’s voice?
Yes, with an extra prompt layer. You can supply two or three of the adviser’s own writing samples and add an instruction such as “match the tone and sentence rhythm of the sample, without copying phrases directly.” Keep two cautions in mind. Do not use samples that contain client information, and still run the result through the same review, because a model that imitates a voice can also imitate that voice’s overconfidence.
What if the AI still produces a risky phrase?
It can happen, even with a careful prompt. That is why the prompt is a filter and not a guarantee. A human reviewer is the safeguard that matters, and the prompt’s job is to make that review faster and more focused.
How often should the prompts be updated?
There is no universal schedule. A reasonable habit is to review your templates on a regular calendar and whenever your firm’s compliance policy changes, since guidance such as FINRA’s AI resources can be updated. Treat the prompt library as a living document and not a one time setup.
Before You Start: A Short Checklist
- Write down the rules your firm already applies when a person drafts these documents. Turn them into hard constraints in the prompt, not suggestions.
- Confirm with compliance which AI tools may be used and what data may be entered into them.
- Replace every bracket in the templates with your own firm type, document type and approved disclosure text.
- Verify every figure against its original source.
- Decide who reviews, who approves and what gets saved.
AI is not a shortcut around compliance in wealth management. Used carefully, it can give a team a more consistent first draft, with the judgment and the sign off still resting with the people responsible.
FAQ’s
1. Has the SEC taken action against firms over misleading AI claims?
Yes. On March 18, 2024, the SEC announced settled charges against two investment advisers, Delphia (USA) Inc. and Global Predictions Inc., for false and misleading statements about their use of artificial intelligence. The firms agreed to pay civil penalties of $225,000 and $175,000 without admitting or denying the findings. This practice is often called “AI washing.” See the SEC press release. Enforcement priorities change over time, so check current SEC and FINRA materials for newer developments.
2. Do AI assisted market overviews and client recaps need to be kept on file?
It depends on the firm and the document. Registered investment advisers are subject to books and records requirements under Rule 204-2 of the Advisers Act, and FINRA members have their own recordkeeping and supervision obligations. Whether a draft, a prompt or an AI output counts as a record that must be kept is a question for your compliance team. A sensible default is to ask before you start, not after a document has gone out.
3. Why can a firm have a written AI policy and still have a problem?
Because a policy only helps if daily practice matches it. If the policy says every AI draft is reviewed but advisers are pasting drafts straight into emails, the policy is not doing its job. Test your actual workflow against your written procedures.
4. What is the biggest AI risk when drafting market commentary or client recaps?
There is no single answer, but common risk areas include unbalanced or promissory language, unsourced numbers, recaps that overstate what was discussed, client privacy, recordkeeping and weak supervision. FINRA’s Regulatory Notice 24-09 notes that its rules are technology neutral and points to communications and supervision requirements as relevant when firms use generative AI. In practice, human review of everything that reaches a client is the common thread.
5. Is overclaiming AI capability a risk as well as undisclosed AI use?
Overclaiming has led to enforcement, as the Delphia and Global Predictions cases show. Whether and how a firm should disclose its internal use of AI is a decision for its compliance team. A practical rule is to describe what your tools actually do, accurately and without exaggeration, in any public or client facing statement.
Rehan Malik is a blogger and the creator of PromptByJob. He publishes articles and prompt examples about using AI tools for different tasks. He uses AI-assisted drafting in his writing process. He is not a healthcare professional, and his nursing-related content should not be treated as medical advice or evidence of clinical testing.
