AI Trading Coach Legal Risk: Education, Advice, and Consumer Protection

I am Sergei Tokmakov, a California attorney (CA Bar #279869). If you are building a trading product with an AI "coach" feature, the coaching layer is where a lot of the legal risk concentrates. The same feature can pull in two opposite directions. AI coaching can strengthen an educational characterization of your product, because it can teach process, discipline, and self-review without ever recommending a security. Or it can quietly create investment-advice risk, because the moment the AI starts telling a specific user what to buy or sell, you may have moved from teaching to advising. Which way it cuts depends almost entirely on what the AI actually says to users, not on how you label the product. This guide is legal information, not legal advice, and it does not reach a conclusion about whether any particular product is or is not an investment adviser. That question is fact-specific and needs its own analysis.

1. The Education-Versus-Advice Line

The federal definition that matters most here is the definition of "investment adviser" in the Investment Advisers Act of 1940. It appears at section 202(a)(11) of the Act, codified at 15 U.S.C. section 80b-2(a)(11). Read the operative text rather than a summary, because the structure of the definition is what drives the analysis.

The statutory test

Section 202(a)(11) defines an investment adviser, in relevant part, as any person who, for compensation, engages in the business of either of the following:

• Advising others, either directly or through publications or writings, as to the value of securities or as to the advisability of investing in, purchasing, or selling securities; or
• Issuing or promulgating analyses or reports concerning securities, for compensation and as part of a regular business.

Courts and the SEC commonly describe this as a three-part inquiry: (1) does the person provide advice or issue reports about securities, (2) for compensation, (3) as part of a business. All three elements generally need to be present. The "for compensation" element is read broadly and does not require a separate or itemized fee for the advice; an economic benefit can be enough. The "securities" element is important because advice purely about non-security instruments may fall outside the Advisers Act, although other regimes can apply. A trading coach that talks about securities, for an economic benefit, as part of a business, is squarely inside the elements that the definition cares about, which is why the content of its output matters so much.

Statutory exclusions, including the publisher's exclusion

The definition is followed by a list of statutory exclusions. One that frequently comes up for content and coaching products is the publisher's exclusion, which excludes the publisher of any bona fide newspaper, news magazine, or business or financial publication of general and regular circulation. The Supreme Court interpreted that exclusion in Lowe v. SEC, 472 U.S. 181 (1985). In Lowe, the Court treated bona fide publications that offer impersonal, generally circulated commentary differently from person-to-person advice tailored to a specific client's situation, and it held that the newsletters at issue fell within the publisher's exclusion. The practical line that emerges from Lowe and later guidance is the distinction between general, impersonal market commentary on one hand and personalized advice directed at an individual's circumstances on the other.

That distinction maps directly onto AI coaching. Impersonal, educational commentary that any user could read is closer to the publishing side. Output that is generated for a specific user, reacts to that user's positions, and tells that user what to do with their own portfolio looks more like personalized advice. I want to be careful here: the publisher's exclusion is narrow and fact-specific, an AI tool that personalizes output per user is not obviously a "publication of general and regular circulation," and reliance on the exclusion should never be assumed without analysis. The point for design purposes is the underlying theme. The more personalized and prescriptive the output, the more it looks like the kind of advice the Act was built to reach.

2. Safer Educational Output vs Riskier Personalized Output

Because substance controls, the most useful thing I can give a product team is a concrete sense of what tends to read as educational versus what tends to read as advice. The line is not mechanical, and context can move any single example, but the pattern is reliable: descriptive feedback about the user's own past or simulated behavior is safer; specific, forward-looking instructions about named securities are riskier.

Notice what separates the two columns. The safer items look backward at behavior the user already engaged in, usually in a simulated or journaling context, and they coach process: risk discipline, consistency, adherence to the user's own plan. The riskier items look forward, name specific securities, and tell a specific user to take a specific action, sometimes with an implied or express prediction of profit. The forward-looking, named-security, you-should-do-this output is the output most likely to be characterized as a recommendation about the advisability of investing in a security.

Design implications

If you want to keep the coach on the educational side, the constraints follow from the columns above. Constrain the model so that it describes and explains rather than directs. Favor feedback on the user's own logged or simulated behavior over forward-looking calls. Avoid naming specific securities in a buy or sell instruction directed at the individual user. Keep outcome and probability claims out of trade-level output. None of this guarantees a particular classification, but it pushes the substance toward education and away from personalized advice, which is the lever you actually control. Even 'model portfolio,' 'trade setup,' or 'signal score' language can become advice-like when it is tied to named securities and sold behind paid access.

3. Broker-Dealer and Affiliate Compensation

The adviser definition is not the only registration regime in play. How the product connects users to actual trading, and how it gets paid, can implicate broker-dealer rules and can sharpen the adviser question.

• Routing users to brokers. If the product hands users off to a broker, or executes or arranges trades, that activity can raise broker-dealer questions depending on the role you play and the compensation involved. "We just send users to a partner broker" is not automatically clean; the structure and the payment matter.
• Recommending trades. Recommending specific securities transactions to specific users is the kind of activity that, combined with compensation, can implicate both adviser and broker-dealer frameworks. A recommendation feature is materially different from an education feature.
• Affiliate or transaction-based compensation. Taking affiliate compensation tied to users' trading activity, payment that scales with how much or what users trade, is a classic flag. Transaction-based compensation for steering or recommending trading is exactly the fact pattern that draws broker-dealer scrutiny. If your revenue goes up when users trade more, assume that fact will be examined closely.

4. FTC Marketing and Substantiation

Separate from the securities-registration question, how you market the coach is governed by consumer-protection law. Section 5 of the FTC Act, codified at 15 U.S.C. section 45, prohibits unfair or deceptive acts or practices in or affecting commerce. Earnings claims about money-making and trading products are a long-standing FTC enforcement priority, and the agency treats unsubstantiated claims as deceptive.

Earnings and outcome claims

Marketing language such as "make money," "trade profitably," "financial freedom," or specific or implied income figures is exactly the category the FTC scrutinizes for money-making opportunities. The core requirement is substantiation: a person making an objective claim about likely results needs competent and reliable evidence to support that claim at the time it is made. Express claims and reasonably implied claims both count, so an image of a luxury lifestyle next to a trading product can communicate an earnings message even without an explicit number. The conservative practice is to avoid earnings and profitability promises about what users will achieve, and to make sure any results-related statement is backed by real, representative evidence rather than aspiration.

Testimonials and influencer posts

Testimonials, endorsements, and influencer or affiliate posts carry their own disclosure obligations. Material connections between you and an endorser, for example payment, free access, or an affiliate relationship, need to be clearly and conspicuously disclosed. Endorsements should reflect honest, typical experiences rather than cherry-picked outliers, and a testimonial does not relieve you of the duty to substantiate the underlying claim. If your growth plan relies on influencers posting trading results, the disclosure and substantiation issues scale with it. I would put written terms in place with every endorser and review what they actually post.

5. Privacy and Behavioral Profiling

A behavioral coach is, by design, a data product. To score a user's discipline, consistency, or risk behavior, you generally have to collect and analyze personal information: trading habits, preferences, account or device data, sometimes geolocation, and then you generate user scores or profiles from it. That collection-plus-scoring pattern is what brings privacy law into the picture.

CCPA and CPRA

For a business that meets the statutory thresholds, collecting personal information from California consumers can trigger a California Consumer Privacy Act analysis, as amended by the California Privacy Rights Act (Cal. Civ. Code section 1798.100 and following). At a high level that can include notice at or before the point of collection, purpose limitation and data-minimization expectations, honoring consumer rights such as access, deletion, correction, and opt-out of sale or sharing, and special attention if any data qualifies as sensitive personal information. Whether and how these obligations apply depends on the business's size and revenue, the volume and nature of the data, and the users involved, so the thresholds and specifics need to be checked against the actual facts.

Automated decision-making and profiling

Generating a behavioral score is profiling, and using that score to make or inform decisions about the user implicates the developing rules on automated decision-making technology and profiling under the CPRA framework, alongside transparency obligations about the logic and use of such scoring. This is an active and evolving area, and other states' privacy laws plus the FTC's unfairness and deception authority can apply to data practices as well. The practical takeaways are concrete: know exactly what you collect, say so clearly, limit use to disclosed purposes, give users the rights the law requires, and be ready to explain how scoring works and what it drives.

6. Practical Guardrails

Pulling the threads together, here is the conservative posture I would generally start from for an AI trading-coach product. None of this is a substitute for a fact-specific analysis of your build, and none of it guarantees a particular regulatory classification. It is a way to keep the substance on the educational side while the specific questions get worked out.

Frequently Asked Questions

Is an AI trading coach an investment adviser?

It depends on what the AI actually outputs and on the surrounding facts. The Investment Advisers Act test at section 202(a)(11) (15 U.S.C. section 80b-2(a)(11)) turns on whether a person, for compensation and as part of a business, advises others about the value of securities or the advisability of investing in securities, or issues reports or analyses about securities. There are statutory exclusions, including a publisher's exclusion for bona fide publications interpreted in Lowe v. SEC. General, impersonal educational commentary is treated differently from personalized, prescriptive advice, but whether a particular product crosses the line is a fact-specific question that needs its own registration analysis. I cannot, and this page does not, conclude that any specific product is or is not an adviser.

What AI coaching outputs are safer from an investment-advice standpoint?

Descriptive, educational, and process-focused feedback about a user's own past or simulated behavior is generally safer than specific, personalized trade instructions. Safer examples include "you concentrated too much simulated risk in one position," "your position sizing was inconsistent," and "you exited earlier than your stated plan." Riskier outputs are specific recommendations about named securities such as "buy this named stock tomorrow," "short this named asset at this level," or "this trade has a high probability of profit." This is a framework for reducing risk, not a guarantee that any particular output stays on the educational side of the line.

Can I market an AI coach as helping users "make money" or "trade profitably"?

Earnings and profitability claims are a known enforcement concern. FTC Act section 5 (15 U.S.C. section 45) prohibits unfair or deceptive acts or practices, and the FTC treats unsubstantiated "make money" or "financial freedom" claims about trading and money-making products as potentially deceptive. Claims about likely results need competent and reliable substantiation in hand at the time they are made, and testimonials or influencer posts need clear disclosure of material connections and should reflect typical, not cherry-picked, results. The conservative practice is to avoid promising what users will earn.

Does behavioral scoring of users trigger privacy law?

Behavioral scoring usually means collecting personal information such as trading habits, preferences, device data, and geolocation, and generating user scores or profiles. For a business that meets the thresholds, that activity can trigger a CCPA and CPRA analysis, including notice at collection, purpose limitation, consumer rights, and the developing rules on automated decision-making and profiling. Other states' privacy laws and the FTC's unfairness authority can also apply. The specifics depend on the data, the users, and where the business operates, so the thresholds need to be checked against your facts.

Do I need disclaimers or registration?

Disclaimers help characterize a product as educational, but they do not by themselves convert advice into education; substance controls over labels. Whether registration as an investment adviser or broker-dealer is required is a separate, fact-specific question that depends on what the AI outputs, how the product is compensated, and whether it routes users to brokers or recommends trades. The conservative approach is to keep coaching educational and non-personalized unless a registration analysis supports a more advisory model. I would run that analysis before adding any feature that recommends specific securities or ties revenue to user trading.