How bar advertising rules apply to AI visibility work

Which bar rules actually touch AI visibility work?

Four rules carry almost all of it. ABA Model Rule 7.1 bars false or misleading communications about a lawyer or the lawyer’s services. Model Rule 7.2 governs advertising and what a firm may pay for. Model Rule 7.3 governs solicitation. Model Rules 5.1 and 5.3 make a lawyer responsible for the vendors working under the firm, which is where an outside answer-engine program lands.

The useful move is to ask which rule each task engages. Most engage none: fixing an address string across five directory profiles touches no rule at all. A handful engage Rule 7.1 directly, because they put a new sentence in front of a prospective client. One engages Rule 7.2 in a way that surprises firms, because paying a directory for a listing and paying for a recommendation are not the same act.

Does Rule 7.1 apply to a sentence that an answer engine wrote?

Model Rule 7.1 reaches communications made by or on behalf of the lawyer. A paraphrase that ChatGPT or Google’s AI Overview generates on its own is not a communication the firm made. What the firm supplied to the engine is. That distinction is where the exposure sits, because an answer engine strips context by design: it lifts one sentence away from the page that qualified it and presents the sentence alone.

The Harvard Journal of Law and Technology digest published in January 2026 made the structural point directly. Attorney advertising rules were written for communications a lawyer controls end to end, not for algorithmic intermediation, where a third party restates the firm’s content without the firm in the loop. The rules have not been rewritten. The mitigations the JOLT commentary describes are the ones a firm can run: precise, jurisdiction-qualified statements, and monitoring what the engines say about the firm.

For an AI visibility program this changes the drafting standard rather than the legal test. A sentence that is accurate in context but misleading when quoted alone is a Rule 7.1 problem waiting for a machine to create it. Firm AEO writes to the quoted-alone standard for that reason, and that is also the standard that gets content cited.

Do the state filing regimes cover a website or a directory profile?

Several states still require lawyer advertisements to be filed with the bar, and the regimes differ enough that a national program cannot assume one answer. Florida requires a copy of each advertisement to be filed at least 20 days before first dissemination, at $250 for a timely filing and $750 for a late one, according to The Florida Bar’s advertising filing requirements. Texas runs the opposite way round.

Texas requires filing with the Advertising Review Committee of the State Bar of Texas no later than 10 days after the date of dissemination, and offers optional pre-approval for a lawyer who submits not fewer than 30 days before first dissemination, per the Texas Center for Legal Ethics. New York imposes no pre-filing but does impose labelling and retention: under 22 NYCRR 1200.7.1 an advertisement must be labelled Attorney Advertising on the home page of a website, computer-accessed advertising must be retained for at least one year, and the contents of a covered website must be preserved on initial publication, on any major redesign or meaningful content change, and in no event less often than every 90 days.

Most of these regimes exempt material that is not a paid advertisement, and whether a firm’s own website content counts varies by state. That classification question is the one a marketing vendor must not answer. What a vendor can do is keep the record: a dated copy of every page version published.

Is an AI intake chat a solicitation under Rule 7.3?

Model Rule 7.3 defines solicitation as a communication initiated by or on behalf of a lawyer and directed to a specific person known to need legal services. An inbound chat widget a visitor opens on the firm’s own site is not lawyer-initiated, so the live person-to-person prohibition in Rule 7.3(b) is generally not the concern. The concern is Rule 7.1 and a relationship nobody intended to form.

Florida Bar Advisory Opinion 24-1, issued in January 2024, set out four caveats for generative AI in intake and advertising, and they are the cleanest published checklist on this. The chatbot must disclose that the user is talking to a machine rather than a lawyer. It must not give legal advice. It must screen for persons already represented. And it must not create an accidental attorney-client relationship.

A firm running intake automation because answer engines send people to the site at odd hours should treat those four items as build requirements, not review notes. The person an engine sends has already described their situation to a chatbot in another window, and arrives primed to keep describing it.

• •State plainly, in the first message, that the assistant is automated and is not a lawyer.
• •Constrain the assistant to scheduling, eligibility screening and routing, with no answer that reads as advice on the person’s own matter.
• •Ask whether the person is currently represented, and stop the intake if they are.
• •Include the no-relationship-formed language where the person will actually read it, not only in a policy page footer link.

Do I have to disclose that marketing content was AI-generated?

In most US jurisdictions, no. The published state guidance is about supervision and accuracy rather than labelling. ABA Formal Opinion 512, issued July 29, 2024, applies Model Rules 1.1, 1.6, 1.4, 5.1, 5.3, 3.3 and 1.5 to generative AI and treats competent supervision as the control. The New York State Bar Association AI task force report is the outlier that raises disclosure of AI use directly, and it is a report rather than a rule.

Florida Opinion 24-1 is a real disclosure requirement but a narrow one. It requires disclosing that a chatbot is AI, a statement about the interaction rather than a label on ad copy. No US jurisdiction reviewed here requires a law firm to mark a blog post or a practice-area page as AI-assisted. Platform rules are separate and do bite: YouTube, Meta and TikTok each require disclosure of realistic synthetic media, which reaches a firm running AI-generated video or voice in paid social.

The reason bars are attentive is adjacent rather than advertising-related. The 5WPR and Haute Lawyer Legal AI Visibility Index, published April 2026, counted 487 documented AI-hallucination cases in US filings in 2025, roughly ten times the 2024 figure, with 37.8% involving licensed attorneys. That record shapes how a disciplinary body reads any AI-assisted work from a law firm, marketing included.

Can a firm say it is the best when the prompt literally asks who is best?

No, and this is the sharpest collision between how people query answer engines and what Model Rule 7.1 permits. Rule 7.1 bars unverifiable superlatives and comparisons that cannot be factually substantiated. Consumers ask engines for the best lawyer in a city constantly, and the temptation is to answer that prompt in its own words. A firm that calls itself the best personal injury lawyer in Phoenix cannot substantiate the claim.

The route that works is corroboration rather than assertion. Engines assemble a best-of answer from third-party sources, not from the firm’s self-description. Citorian’s five-engine study of 359 answers in June 2026 found engines named at least one firm 62% of the time, and 94% on vetting and direct prompts where the user asks who to hire, built from directory and review sources. A verifiable award, a bar certification in a specialisation, a Chambers or Best Lawyers listing and a dated record of recognitions each survive Rule 7.1, because each is attributable to a named third party.

Two further constraints are worth naming. Texas Opinion 705 carries an express list of prohibited terms, so safe phrasing is jurisdiction-dependent. In Canada the position is stricter: the Law Society of Ontario treats best and super style claims as impermissible absent objective verification. Firm AEO publishes a separate Canadian guide and does not carry Canadian rules into a US compliance map.

Does an AI Overview carry your case-result disclaimer?

It does not, and no firm should plan as if it might. Google’s AI Overview, ChatGPT and Perplexity each restate source material in their own words and cite the page. None is obliged to carry a disclaimer sitting in a separate paragraph, a footer or a hover state. A results page that satisfies Rule 7.1 through a block of qualifying text at the bottom can be extracted in a form that satisfies nothing.

The mitigation is structural. Every number describing an outcome belongs in a sentence carrying its own qualifier, so the smallest extractable unit is still compliant. The settlement figure, the matter type, the year and the statement that results depend on the facts of each case belong in one sentence rather than four. Firm AEO treats the sentence, not the page, as the unit of compliance for anything with a number in it.

This post states the principle. The sentence-level rewrite mechanics, with before-and-after examples, are covered separately in the Firm AEO post on writing the qualifier into the extractable sentence.

What does a compliant AI visibility program look like in practice?

It looks like a normal marketing program with three added controls: a named lawyer who reviews every externally published claim, a retained record of what was published and when, and a written division of responsibility with the vendor. Model Rule 5.3 makes the firm responsible for the vendor’s conduct, so the engagement itself is a compliance artifact. A vendor that cannot say who at the firm approved a page should not be publishing it.

The Federal Trade Commission rule on fake and incentivised reviews, effective October 21, 2024, with penalties reported up to $51,744 per violation, sits alongside the bar rules rather than inside them and applies regardless of what any state bar says. Review generation is where a well-meaning process crosses a federal line fastest.

• •One named reviewing lawyer per content batch, recorded by name and date before anything publishes.
• •A dated archive of every published page version, which New York requires and which every other jurisdiction makes useful.
• •A written statement of which party drafts, which approves and which files in a filing state.
• •A standing instruction that no outcome number publishes without its qualifier in the same sentence.
• •A quarterly read of what the engines say about the firm, which is the only practical answer to a misstatement the firm did not write.
• •Ethics counsel, not the marketing vendor, resolving every question about how a rule applies to the firm.