When Your Website Makes a Recommendation You Can’t Defend

A prospect on your advisory firm’s website tells the chat widget she is 32, has about $40,000 saved, and wants to know what to do with it. "How should I invest this?"

The assistant answers like it has done this a thousand times. Put roughly 60 percent in low-cost index funds, 30 percent in bonds, keep 10 percent in cash, and consider maxing the Roth first. Clean, confident, specific.

It is also a securities-law problem with your firm’s name on it. An AI chatbot on a financial advisor’s site just gave personalized investment advice to a member of the public, with no idea of her debt, her timeline, her risk tolerance, her tax situation, or whether she can stomach a 20 percent drawdown. No suitability analysis happened. No fiduciary judgment was applied. A paragraph generated from three facts told someone where to put her money.

Giving Advice Is What Pulls You Into the Rules

The threshold question in this industry is whether you are giving investment advice, because that is what triggers the obligations. Under the Investment Advisers Act, advising others about securities for compensation is what makes you an adviser in the first place, and an adviser owes a fiduciary duty: act in the client’s best interest, surface conflicts, recommend only what suits this specific person.

A tool that hands a stranger an allocation has given advice without doing any of that. There is no best-interest analysis when the system knows nothing about the person. If your firm operates on the brokerage side, Regulation Best Interest sets the parallel bar: a recommendation to a retail customer has to be in that customer’s best interest, judged against their full profile. Either way, the recommendation came first and the profile never existed.

A Suitability Answer With No One to Be Suitable For

Suitability is not a formality you can paraphrase. It is the requirement that a recommendation fit the actual investor in front of you, and it depends on facts a website visitor has not given and a chatbot has not gathered.

So watch how ordinary the failure looks. "Is now a good time to buy stocks?" pulls a market-timing opinion out of the tool. "Should I roll my old 401(k) into an IRA?" gets a recommendation on one of the most consequential, conflict-laden decisions a person makes, with none of the analysis a rollover recommendation legally requires. "What do you think about Tesla stock?" produces a view on a specific security. Each answer feels like service. Each is a recommendation your firm did not vet, made to a person your firm has not assessed.

Two Sentences That Become Compliance Events

Performance and guarantees are their own category of trouble. A visitor asks what kind of return to expect, and a helpful tool offers a number, "historically the market returns about 7 to 10 percent a year," stated as a forward expectation. The moment that reads as a promise about future results, it is the kind of claim securities regulators treat as misleading.

Then there is the privacy angle people forget. A prospect who types in her income, her account balance, and her goals has just sent nonpublic personal financial information through whatever third-party AI service powers the widget. Regulation S-P expects firms to safeguard that information, and a vendor pipeline the firm never vetted is not safeguarding anything.

The Marketing Rule Reads the Chat Box Too

Whatever your website says about your services is a communication regulated like the rest of your advertising. The Investment Advisers Act marketing rule governs how advisers may talk about themselves and bars misleading claims, including the careless implied testimonial or the cherry-picked performance number. On the brokerage side, FINRA Rule 2210 sets content standards for communications with the public and expects them to be fair and not misleading.

A chatbot improvising claims about your track record or your clients’ outcomes is generating regulated communications at volume, with no principal reviewing them before they reach the public. That is the exact thing the rules are built to prevent.

The Disclaimer Does Not Cover the Advice

"This chat is for informational purposes only and is not investment advice." Then the tool hands out an allocation, recommends a rollover, and floats an expected return. Regulators weigh what the tool did. The fine print does not convert a personalized recommendation back into general education.

A retail investor who received a specific allocation from your website relied on your website. If the market falls and she lost money in the mix your site suggested, the arbitration claim points at the conversation, not at the disclaimer beneath it.

Why the Better Prompt Fails

The vendor’s reassurance never changes: instruct the model to give only general education and to avoid specific recommendations.

The instruction holds until a question stops looking like a request for advice, which is most of them. "What’s a good asset allocation for someone my age?" sounds educational, so the model returns a specific allocation. It thought it was teaching. It made a recommendation. The guidance covers the questions that announce their intent and misses the ones phrased as casual curiosity, which is nearly all of them.

And that is the structural point. An instruction is a preference the model can override; it does not constrain what the model is capable of saying. A real boundary lives in the system around the model and decides what is allowed out before a word is generated, so a personalized recommendation never reaches the visitor no matter how the question is dressed up. "Will not" is a suggestion. "Cannot" is an architecture.

What AI Chatbot Liability Costs a Financial Advisor

The SEC and state securities regulators bring enforcement for unsuitable recommendations, fiduciary breaches, and misleading communications, and the remedies stack: fines, disgorgement, censure, and at the far end a bar from the industry. FINRA fines and arbitration awards run parallel for the brokerage side. A single client who acted on a bad website-generated recommendation and lost money is a claim, and the defense costs real money before anyone reaches the merits.

The reputational hit can be the worst of it. This business runs on trust and on the discipline to do real analysis before giving advice. A public tool that fires off allocations to anyone who asks tells prospects, regulators, and referral partners the opposite, and that story is expensive to unwind.

What a Financial Advisor Actually Needs

Chat belongs on an advisory site. Prospects want to understand your services, learn your minimums and your fee model, book an intro call, and ask plain questions about how you work. None of that requires a tool that gives investment advice.

So the real question is not whether the assistant can be coached to stay educational. It is whether the assistant is structurally incapable of recommending an allocation, opining on a security, projecting a return, or taking in nonpublic financial data, regardless of how the visitor phrases the question. If the protection depends on prompt wording, it breaks the first time someone asks in an unexpected way.

Fred is built around that limit. It answers from your firm’s own content, books the intro call, captures the lead, and routes anything that asks for a recommendation, a projection, or a security-specific opinion to a licensed advisor. It runs more than 50 industry guardrail packs, and the financial pack is built around suitability, performance claims, and recommendations. Fred does not allocate your prospect’s portfolio or call a stock. It cannot. It explains how your firm works and lets a licensed human give the advice.

Frequently asked questions