Investment Advisor Marketing Rule (Rule 206(4)-1) and AI-Generated Content

The Marketing Rule Meets AI-Generated Content: A Supervision Problem Nobody Fully Scoped

The SEC's revised Marketing Rule, Rule 206(4)-1 under the Investment Advisers Act, went into effect on November 4, 2022, after a compliance date that gave advisers about 18 months to prepare. The rule consolidated the old advertising rule and the cash solicitation rule into a single framework. It was, by most accounts, a reasonable modernization. The prior rule dated back to 1961 and was struggling to accommodate anything beyond print ads and mailers.

What the SEC did not fully anticipate, and what compliance teams are now grappling with, is the question of what happens when AI systems generate the client-facing content that falls squarely within this rule's scope.

A Quick Refresher on What the Rule Actually Requires

Rule 206(4)-1 defines "advertisement" broadly. It covers any direct or indirect communication to more than one person, or to one or more persons if the communication offers or promotes advisory services. It also covers endorsements and testimonials, which now have their own set of disclosure requirements.

The rule imposes seven general prohibitions. An advertisement cannot include untrue statements of material fact, unsubstantiated material claims, untrue or misleading implications, or statements about potential benefits without fair and balanced treatment of associated risks. It cannot cherry-pick time periods for performance. It cannot be otherwise materially misleading.

There is also the performance advertising framework, which requires net performance alongside gross, specific time period requirements (1, 5, and 10 year returns or since inception), and prohibitions on hypothetical performance unless you can demonstrate it is relevant to the recipient's financial situation and objectives.

And then there is the books and records requirement under Rule 204-2(a)(11), which requires advisers to keep all advertisements and related documentation, including the substantiation underlying any material claims.

Where AI Content Generation Creates Real Exposure

Consider a straightforward scenario. Your firm uses a large language model to draft client newsletters, LinkedIn posts for advisors, or responses to prospect inquiries. These outputs are advertisements under the rule if they promote your advisory services to one or more persons. The moment that content goes out, your firm is on the hook for every factual claim, every implication about performance, every risk disclosure that should have been included but was not.

Here is where it gets interesting. LLMs hallucinate. They generate plausible-sounding but fabricated statistics, invent case studies, and produce performance figures that have no basis in your firm's actual track record. A model might generate a sentence like "our strategies have consistently outperformed the S&P 500 over the past decade" because that is the kind of language it has been trained on. If your firm has not, in fact, outperformed the S&P 500 over the past decade, you have just violated the prohibition on untrue statements of material fact under 206(4)-1(a)(1).

The SEC has been clear that the general prohibitions are principles-based and apply regardless of the medium. In the adopting release (IA-5653, December 2020), the Commission specifically noted that the rule was designed to be "flexible enough to remain relevant as technology and industry practices evolve." They were thinking about social media and digital marketing at the time, but the language is broad enough to cover AI-generated content without any interpretive stretch.

The Supervision Gap

Section 203(e)(6) of the Investment Advisers Act and Rule 206(4)-7 (the compliance rule) require advisers to adopt and implement written policies and procedures reasonably designed to prevent violations. The SEC's Division of Examinations has consistently flagged marketing compliance as an exam priority; it appeared again in the 2024 examination priorities published in October 2023.

The supervision question with AI-generated content is not whether you need to review it before publication. Obviously you do. The question is whether your current review process is designed to catch the specific failure modes of generative AI.

Traditional compliance review assumes a human author who might exaggerate or omit disclosures. AI introduces different risks: fabricated data points, subtly inconsistent performance claims across different pieces of content, and confident assertions that lack any substantiation in your records. A compliance reviewer scanning for tone and obvious misstatements might not catch a hallucinated statistic presented with perfect confidence.

The SEC's enforcement action against Titan Global Capital Management in March 2024, which resulted in a $192,454 penalty, involved misleading hypothetical performance in social media ads. The content was human-generated, but the violation pattern, making claims that could not be substantiated with actual records, is exactly the pattern AI content generation tends to produce at scale.

Substantiation and Books and Records

Rule 206(4)-1(b) requires that advisers have a "reasonable basis" for believing they can substantiate material statements of fact. When a human writes marketing copy, the substantiation trail is relatively straightforward: the person either pulled the data from your performance records or they did not.

When an AI generates a material claim, the substantiation question becomes more complex. Where did the claim originate? Can you trace it to your firm's actual data? If the model was fine-tuned on your materials, did those materials contain outdated performance figures? If the model was prompted with context, was that context accurate and current?

Under Rule 204-2(a)(11), you need to retain advertisements and the records supporting them. For AI-generated content, a reasonable interpretation would extend this to prompts, model outputs, any retrieval-augmented generation (RAG) sources the model drew from, and the review trail showing who approved the final version. The SEC has not issued specific guidance on this yet, but the logic follows directly from existing requirements.

Practical Steps for Compliance Programs

• Classify AI outputs before they are generated. If the intended use falls within the rule's definition of advertisement, route it through your marketing review process from the start, not after someone has already sent it to a prospect.
• Build substantiation checks into the workflow. Every factual claim in AI-generated content should be traceable to a specific source in your records. If the model asserts a performance figure, your reviewer needs to verify it against actual composites or account data.
• Retain the full generation chain. Prompts, model version, RAG sources, raw output, edits, and final approval. Treat this as part of your books and records obligation.
• Train reviewers on AI-specific failure modes. Hallucinated statistics, fabricated testimonials, and inconsistent performance claims across content pieces are different from the errors human copywriters typically make.
• Restrict hypothetical performance generation entirely. Given the rule's strict requirements around hypothetical performance (relevance to recipient, policies and procedures to ensure compliance), letting an AI generate hypothetical scenarios is high-risk with limited upside.

Where This Is Heading

The SEC's approach to AI in the advisory space has been evolving. The proposed predictive data analytics rule from July 2023 (though its future is uncertain given the current regulatory climate) signaled that the Commission is thinking about AI's role in client interactions. Even without new rulemaking, the existing Marketing Rule framework gives examiners plenty to work with when they encounter AI-generated content that violates the general prohibitions.

Firms that are using generative AI for marketing without updating their compliance infrastructure are accumulating risk that compounds with every piece of content published. The violations are not hypothetical; they are the natural output of systems that optimize for plausibility rather than accuracy.

How FirmAdapt Addresses This

FirmAdapt's architecture treats compliance constraints as inputs to the generation process rather than filters applied after the fact. For investment advisers subject to Rule 206(4)-1, this means AI-generated content is checked against firm-specific performance data, required disclosures, and the rule's general prohibitions before it reaches a human reviewer. The platform maintains the full generation chain, including prompts, sources, model outputs, and approval records, in a format designed to satisfy books and records requirements under Rule 204-2.

The practical effect is that compliance reviewers spend their time on judgment calls, such as whether a particular framing is materially misleading, rather than hunting for hallucinated statistics. FirmAdapt does not eliminate the need for human review, but it substantially reduces the surface area of risk that review needs to cover.