Reg BI & the Marketing Rule Online: A 2026 Advisor Playbook

The regulators settled the "should we use AI" debate by moving past it. In March 2024 the SEC brought its first enforcement actions for what it calls "AI washing", charging two advisers for overstating their use of artificial intelligence and fining them under the Marketing Rule. The lesson was not that AI is forbidden. It was that everything an adviser says about AI, and everything an AI says on an adviser’s behalf, is already inside the rules that govern advice and advertising.

This guide is the companion to the threat side of that story. The threat piece covers what goes wrong when an unguarded chatbot sits on an advisory site. This one is the standard: what a compliant deployment looks like for a financial advisor in 2026, and the specific lines the system has to hold.

The Obligation Attaches to the Advice, Not the Author

Giving investment advice is what pulls a firm under the rules in the first place. An adviser owes a fiduciary duty under the Investment Advisers Act: act in the client’s best interest, surface conflicts, recommend only what suits the specific person. A broker-dealer owes the parallel best-interest standard under Regulation Best Interest. Neither obligation cares whether the recommendation came from a person or a paragraph the website generated.

That is why the dangerous answers look like ordinary helpfulness. A visitor asks how to invest a windfall and a friendly tool hands back an allocation. It knows nothing about the person’s debt, timeline, risk tolerance, or taxes, so the recommendation cannot possibly be suitable, and an unsuitable recommendation made under your firm’s name is your problem. General education is safe. The trouble starts the instant the answer applies to this person’s money.

The 2026 Compliance Standard, Line by Line

A compliant advisory assistant is defined by what it is built to refuse. Treat the list below as the floor.

• No personalized recommendations. An asset allocation, a "you should roll over your 401(k)," a "max the Roth first" all require suitability analysis the website cannot do, so they route to a licensed advisor.
• It does not opine on specific securities. "What do you think about this stock?" is a recommendation; the assistant declines and hands it to a human.
• Forward-looking returns are off the table. A number stated as what to expect reads as a promise, and that is exactly the misleading-communication problem the rules police.
• Marketing claims stay honest and bounded. The Advisers Act marketing rule and, for the brokerage side, FINRA Rule 2210 govern how the firm describes itself, including any claims about the AI itself, which is the trap the SEC’s AI-washing cases sprang.
• Sensitive financial data is handled with care. A prospect’s income, balances, and goals do not get piped into a vendor system the firm never vetted or that falls outside its Regulation S-P safeguards.
• And every conversation is logged and reviewable, because a governed system produces the record an examiner expects to see.

The pattern is the same one underneath all of it. The assistant answers what needs no license, your services, your fee model, your minimums, how to book a call, and routes every recommendation, projection, and security-specific opinion to a person. That division is the whole standard.

Honesty About the Tool Is Now Its Own Rule

The AI-washing cases added a wrinkle worth stating plainly. It is not enough to govern what the assistant says to prospects; you also have to be truthful about what the assistant is. Calling a basic chatbot an "AI-powered advisor" or implying it does analysis it cannot do is the precise misstatement the SEC penalized under the Marketing Rule. A compliant deployment describes the tool accurately and modestly, and it can back the description up.

The reach of this is wider than the website. The SEC found problems across Form ADV filings, a press release, and social posts, which means the claim you make about your assistant has to be consistent everywhere a prospect might read it. A landing page that calls the tool a "robo-advisor powered by machine learning" while the assistant is really a scripted FAQ is the kind of mismatch examiners now look for. The simplest way to stay clean is to keep the description plain and let the tool’s actual behavior match the words.

Why a Prompt Cannot Meet the Standard

The common shortcut is to write the rules into the assistant’s instructions. Tell it to give only general education, never a specific recommendation, and consider the boundary set.

It is not set, because of how the model reads a request. It follows an instruction when the question resembles the wording it was warned about, and slips the moment the phrasing changes. You tell it never to recommend. The prospect never says "recommend something." They ask, "what’s a good allocation for someone my age?" The model hears an educational question and returns a specific allocation. The instruction was loaded the whole time. It just did not recognize the sentence that crossed the line.

That is the gap between an instruction and a standard. An instruction asks the model to behave; it does not stop the model from speaking. A real boundary is built into the system and decides what the assistant is allowed to say before it answers, so a personalized recommendation or a return projection never reaches the prospect no matter how the question is framed. "Will not" is a suggestion. "Cannot" is an architecture.

What a Compliant Deployment Looks Like

Meeting the 2026 standard does not mean switching the assistant off. It means deploying one built to hold the line your registration depends on, and one that produces the record your examiner will ask for.

Fred is built that way. It answers from your firm’s own content, captures and scores the lead, books the intro call, and routes anything that touches a recommendation, a projection, a security, or sensitive financial data to a licensed advisor. It runs more than 50 industry guardrail packs, and the financial pack is built around suitability, performance claims, and the recommendations a registration reserves. Fred does not allocate a portfolio or call a stock. It cannot. It answers what it should, logs every exchange, and books the licensed work for the people licensed to do it.

That is the difference between hoping the assistant stays educational and being able to show a regulator why it cannot do otherwise.

Frequently asked questions