Can financial advisors use AI agents? What the rules actually allow
Yes, if the agent drafts and a human approves. The SEC Marketing Rule makes you responsible for advertisements whoever wrote them, FINRA 2210 requires principal review, and books-and-records rules mean an AI draft is still a record. Here is how advisors use agents safely.
WorkAgent console
Delegate a task · watch it run · get the result
Pick a task to delegate
It works the task end to end and hands back .
| Company | Contact | Fit | |
|---|---|---|---|
Subject:
Research, lead gen, outreach, inbox and reporting, delegated and done.
Like what it did? Put WorkAgent on your real work.
The short answer: yes, US advisors and RIAs can use AI agents, as long as the agent drafts and researches while a human reviews and approves. The rules that matter are not about whether you may use software. They are about responsibility and records: under the SEC Marketing Rule an adviser is responsible for its advertisements no matter who or what wrote them, FINRA Rule 2210 requires principal review and approval of retail communications for broker-dealer representatives, and the Advisers Act books-and-records rule means those communications have to be retained. An AI draft is still your communication and still your record.
What an AI agent can genuinely do in an advisory practice
The useful work sits in research, drafting and administration, which happens to be where most advisor hours quietly disappear. Prospect and centre-of-influence research is the strongest fit. Before a meeting with a CPA or an estate attorney you would like to build a referral relationship with, someone should know their firm size, practice focus, recent moves and mutual connections. That research is public, structured and repeatable, and an agent can produce a sourced brief in minutes rather than the thirty you will not spend.
Meeting preparation is the same shape. Pull the client's last review notes, what changed since, what was promised and not delivered, and what is scheduled to happen in the next quarter, into a one-page brief you read in the car. Nothing in that requires judgment about investments. It requires assembly, and assembly is the thing an agent does without getting tired.
Then there is drafting: outreach emails, follow-ups, newsletter copy, seminar invitations. The agent writes the first version, you or your CCO edit it, and it goes out through your normal approval and archiving path. Nothing changes about the review workflow, you just start from a draft instead of a blank page. We break the practice workflows down further on the AI agents for financial advisors page.
Finally, the CRM. Redtail, Wealthbox and Salesforce are only as useful as what gets logged, and advisors are famously bad at logging. An agent that updates records after each meeting, files the notes, sets the follow-up tasks and flags stale contact data solves a real and expensive problem. Our CRM updates use case covers how that loop runs.
Does the SEC Marketing Rule allow AI-written marketing?
The rule does not ban any particular drafting tool. Rule 206(4)-1 governs the content and substantiation of an adviser's advertisements, and the adviser remains responsible for every claim in them regardless of who or what produced the text. So an AI draft is permitted, and it is also entirely your problem if it contains an unsubstantiated performance claim or a misleading statement.
That is why the practical control is a review gate rather than a tool ban. The agent produces a draft, a person who understands the rule reads it, and only approved text goes out. In practice the risk with AI drafting is not exotic. It is the ordinary temptation of confident, promotional language: implied guarantees, cherry-picked results, testimonials used without the required disclosures. A reviewer catches those the same way they would catch them in a junior associate's copy.
Principal review, and why nothing goes out unreviewed
For registered representatives of broker-dealers, FINRA Rule 2210 requires that retail communications be approved by an appropriately qualified principal before use, with categories and timing set out in the rule. That obligation exists whether the first draft came from a person, a template or a model, and no vendor can take it off your desk.
The design that works is boring and reliable: the agent has no send authority. It writes into a review queue. A human reads, edits, approves, and the approved version goes out through your compliant channel where it gets archived. If you set the agent up so that it can send email directly to clients or prospects, you have built a supervision problem for yourself that no amount of prompt instructions will fix. Treat send permission as the single control that matters most.
The records point advisors underestimate
Advisers Act Rule 204-2 requires retention of advertisements and client communications, which means AI-drafted material that gets used is a record you have to keep, in the form you actually used it. The mistake is to treat the AI workspace as a scratchpad outside the record set. If a draft becomes a client communication, it belongs in your archive with everything else.
There is a subtler version of the same issue. If the agent is doing research that informs an advertisement, keep the substantiation. A newsletter statement about, say, contribution limits or a tax change needs a source you can point to, and the moment to capture that source is when the draft is created, not eighteen months later during an exam. Ask the agent for citations as part of the output and file them with the draft.
Firms that scale this well usually keep their obligations and controls mapped in one place rather than spread across a spreadsheet, a shared drive and one person's memory, because the review step only holds up if everyone can see what the rule actually requires. None of this is legal or compliance advice, and your own counsel or CCO should sign off on how you apply it.
What an AI agent must never do in an advisory practice
It never gives investment advice and never makes recommendations, to a client or to a prospect. Not in an email, not in a chat widget, not in a "here is what we would suggest" paragraph in a draft that might get skimmed and sent. The moment text moves from information to recommendation, it belongs to a licensed human who owes a fiduciary duty and can be held to it.
It also should not answer inbound client questions unsupervised, particularly the ones that sound administrative but are not. "Should I take the distribution this year or next" arrives looking like a service request and is a tax and planning question. Route those to a person, always. And it does not touch anything involving custody, money movement, or authorizations. Those need controls that have nothing to do with AI and everything to do with fraud.
Paperwork chasing and the quarterly report
Two administrative jobs justify the subscription on their own. The first is chasing paperwork: unsigned account transfer forms, missing beneficiary designations, outstanding advisory agreements, annual questionnaires nobody returned. These are polite, repetitive, scheduled reminders, and they are the last thing an advisor gets to on a Thursday. An agent runs the sequence and reports what is still outstanding.
The second is the practice report. Every quarter someone assembles new accounts, assets gathered, referral sources, meetings held, pipeline by stage, and client anniversaries coming up. It is data collection and formatting, done identically every time, which makes it ideal to hand over. The agent builds the draft from your CRM and you edit the interpretation.
How does this compare to hiring an assistant?
Flat AI agent pricing starts at $149 a month with no credit meter. A US-based virtual assistant generally runs $3,000 to $6,000 a month, and the BLS median annual salary for executive assistants was $76,590 as of May 2025. For a solo advisor or a small RIA, that gap is usually the difference between having administrative help and doing it yourself at night.
The comparison to be careful about is the one people make in the other direction. An agent does not replace a good client service associate who knows your clients, handles the awkward phone call and catches the thing that looks slightly wrong on a form. It absorbs the mechanical share of that role so the person you do have spends their time on the part that needs a person. The AI executive assistant page covers that split in more detail.
How to start without creating a compliance headache
Begin with something that never touches a client: prospect research briefs, meeting prep, or CRM cleanup. Run it for a month and see whether the quality holds. Then move to drafting, with an explicit written rule that the agent cannot send and that every draft goes through your existing review and archiving path. Tell your CCO what you are doing before you start, not after, and write down which categories of communication the agent may draft.
If you want to judge the output rather than the pitch, hand the agent a real task at the top of this page, something like "research these six CPAs in my county and write a one-page brief on each," and see what comes back. The full picture for advisory practices, including the review workflow, is on our AI agent for advisory practices page.