AI Search for Professional Services: How Law Firms, Consulting Firms, and High-Ticket Service Businesses Get Discovered

Publish more content. Target more queries. Add schema. That’s the usual advice. It fits a software company with a catalog more easily than a firm selling judgment.

A law firm has no SKU, no stock level, no price to compare and no product feed. Its buyer arrives with a problem, not a shopping list. The question on the table is rarely “which employment law firm is best” - it’s closer to “can I still classify this contractor as a contractor, and what do I do on Monday”. Those are different retrievals, and they resolve against different sources.

One thing up front: this is a marketing and search-visibility guide. It isn’t legal advice, and nothing in it replaces review by the rules-of-professional-conduct regulator in your own jurisdiction. Attorney advertising is governed separately from search and advertising law, and those rules vary.

Expertise has no product feed

Google’s own generative AI guide points product-minded businesses at two specific plumbing systems:

“Using products like Merchant Center (such as Merchant Center feeds) and Google Business Profiles can help your products and services to be visible in both AI responses and other Google Search results.”

A firm selling expertise has no Merchant Center feed. There’s no catalog to submit, no GTIN to declare, no variant structure, no shipping policy. The second half - Business Profile - is real and relevant, and I come back to it below, because it carries rules that surprise most professional-services marketers.

The practical consequence is that a professional-services firm competes almost entirely on the content side of Google’s stack: crawlable, indexable, snippet-eligible pages that say something an answer engine can actually use. Google’s guide is blunt about what that means:

“Commodity content (for example, something like ‘7 Tips for First-Time Homebuyers’) is often based on common knowledge, which could originate from anyone, and typically adds little unique insight for readers. In contrast, non-commodity content (such as ‘Why We Waived the Inspection & Saved Money: A Look Inside the Sewer Line’) provides unique expert or experienced takes that go beyond common knowledge and the ordinary.”

Notice that Google’s own examples are both from home buying. The commodity one is generic advice anyone could publish. The non-commodity one is a specific engagement, with an outcome, a reason and a process attached. Translating that to a professional-services firm: “7 Tips for Hiring a Lawyer” is commodity. “We reviewed 40 California wage-and-hour class actions last year; here is what the top three findings were and how the settlement math actually worked” is non-commodity. That isn’t a content-volume strategy. It’s an evidence strategy.

Query fan-out works against thin practice-area pages

Google describes query fan-out this way:

“if the original user’s query is ‘how to fix a lawn that’s full of weeds’, fanout queries might include ‘best herbicides for lawns’, ‘remove weeds without chemicals’, and ‘how to prevent weeds in lawn’.”

For a firm, the fan-out of one question decomposes into a comparison prompt, a procedural prompt and a prevention prompt - three retrievals with three different source preferences. A single practice-area landing page written as a positioning brochure is structurally positioned to lose all three.

And the obvious workaround is explicitly named and prohibited:

“while it might be tempting to create separate content for every possible variation of how people might search … doing so primarily to manipulate rankings or generative AI responses in Google Search violates Google’s scaled content abuse spam policy.”

Google’s separate page on AI-generated content repeats the risk in a different register:

“using generative AI tools or other similar tools to generate many pages without adding value for users may violate Google’s spam policy on scaled content abuse.”

If your plan for AI discovery was “spin up a page per fan-out query”, you were going to end up at scaled content abuse anyway. It’s worth knowing that the ceiling you were aiming for is also the penalty line.

Google’s local machinery assumes you have a shopfront

Professional-services firms are, structurally, awkward for Google’s local stack. The schema is built for places you walk into.

Google’s LocalBusiness structured data documentation (last updated 2026-09-08) lists exactly two required properties: name and address. The address property is described as “The physical location of the business.”

Meanwhile, the Business Profile guidelines tell service-area businesses to do the opposite:

“If you’re a service-area business, you should hide your business address from customers.”

“If your business rents a physical mailing address but doesn’t operate out of that location, also known as a virtual office, that location isn’t eligible for a Business Profile.”

“Businesses can’t list an office at a co-working space unless that office maintains clear signage, receives customers at the location during business hours, and is staffed during business hours by your business staff.”

The markup spec requires a physical address. The profile policy says you can’t pretend to have one. That’s awkward if you’re a fully remote consultancy or your only qualifying location is a virtual office. Use the address you actually operate from and accept that you aren’t optimizing for map-pack proximity. A weaker local result is better than a suspension.

The good news: individual practitioners are explicitly covered

Not every professional-services firm is disqualified from local search. Google’s guidelines have a dedicated section, and it names your sector:

“An individual practitioner is a public-facing professional, typically with their own customer base. Doctors, dentists, lawyers, financial planners, and insurance or real estate agents are all individual practitioners.”

The section sets two conditions: the practitioner must “operate in a public-facing role” and must “be contacted directly at the verified location during stated hours.” Support staff and lead-generation agents are explicitly excluded - which is worth internalizing, because “create a profile for every person on the team” is a common and non-compliant suggestion.

The bad news: your profile name is not a keyword slot

The same page enumerates what a business name may not contain. Not permitted: marketing taglines, store codes, trademark symbols, fully capitalized words, business hours, phone numbers or URLs, special characters and legal suffixes, service or product information, location information, containment information. Google’s worked example of a rejected name is instructive:

Not acceptable: “Midas Auto Service Experts” Acceptable: “Midas”

You can add the special characters or legal terms back, but only by “providing real-world proof (e.g. signage, business cards, invoices) that consistently displays those elements as part of your business name.”

This is why the “Best Employment Lawyers in California - Smith & Co” profile naming tactic fails. It isn’t a gray area. It’s listed, with an example of the exact shape most firms try to deploy.

Local Services Ads is the licensed-services surface, and lawyers are served through it

If your business is eligible and you can pass screening, Local Services Ads is a different product from Business Profile, and it’s the one built around licensed services. The screening is described as covering license, insurance and background checks:

“Google verifies that advertisers hold applicable state, provincial and/or country-level licenses for businesses and owners or managers, to the extent practical, and confirms that all workers are appropriately licensed.”

“Business owners, and in some cases, service professionals, may undergo background checks by third-party risk management companies with which Google has partnered.”

Google doesn’t advertise every provider directly. The document names Lawyer.com as one of four partner affiliate networks that screen Local Services providers, alongside Homesnap, Networx and Top Rated Local. That points to legal services using the partner route. Check what’s supported in your category and region. If someone promises placement, ask which screening route they’ll use.

Two details matter for your reputation. The screening isn’t one-way - Google “continues to collect ratings and reviews from people who hired Local Services providers,” and “serious or repeatedly negative customer feedback may result in lower rankings (including not showing at all).” And the Business Profile must be connected to the ad: one of the verification checks is “making sure that Business Profiles on Google are properly connected to matching Local Services ads.”

The two robots.txt mistakes that cost firms their place in AI answers

This is where most of the category’s published advice is simply wrong, and you can check it yourself in about five minutes against primary documentation.

The recurring instruction is “make sure GPTBot, ClaudeBot, PerplexityBot and Google-Extended are allowed.” It appears in agency research reports because it sounds obviously safe - “if in doubt, allow everything”. It isn’t what the platform documentation says.

OpenAI’s crawler overview defines four distinct agents. For search visibility, the one that matters isn’t GPTBot:

“OAI-SearchBot is for search. OAI-SearchBot is used to surface websites in search results in ChatGPT’s search features. Sites that are opted out of OAI-SearchBot will not be shown in ChatGPT search answers, though can still appear as navigational links.”

GPTBot is the training crawler. Anthropic’s documentation makes the same split with three robots: “ClaudeBot helps enhance the utility and safety of our generative AI models by collecting web content that could potentially contribute to their training,” while “Disabling Claude-SearchBot on your site prevents our system from indexing your content for search optimization, which may reduce your site’s visibility and accuracy in user search results.” Perplexity separates PerplexityBot (search) from Perplexity-User (user-initiated fetches).

Allowing a training crawler doesn’t decide whether you appear in AI search results. If you allow GPTBot and block OAI-SearchBot, you’re allowing training access while staying out of ChatGPT’s search answers. Set each switch deliberately. Allow the search agents you want. Restrict training agents according to your licensing decision.

On the Google side, the AI features documentation (last updated 2025-12-10 UTC) is explicit about what controls Search itself:

“robots.txt directives for Googlebot is the control for site owners to manage access to how their sites are crawled for Search. To limit the information shown from your pages in Search, use nosnippet, data-nosnippet, max-snippet, or noindex controls.”

The same page then says “to limit AI training and grounding in some of Google’s other systems, read more about Google-Extended” - that’s, it points the Google-Extended control somewhere other than Search. As this page describes it, Search crawling is governed by Googlebot and the preview controls, and Google-Extended is scoped to systems outside Search.

Two timing details from the same primary docs: OpenAI notes it “can take ~24 hours from a site’s robots.txt update for our systems to adjust,” and Perplexity says changes “may take up to 24 hours for our systems to reflect.” A robots.txt change isn’t something you verify in five minutes.

Reconciling two Google pages that disagree

Google currently publishes two statements about AI-feature eligibility, and they don’t say the same thing. This is worth surfacing rather than resolving quietly, because you’ll meet both if you read carefully.

The 2025-12-10 page says:

“To be eligible to be shown as a supporting link in AI Overviews or AI Mode, a page must be indexed and eligible to be shown in Google Search with a snippet, fulfilling the Search technical requirements. There are no additional technical requirements.”

The 2026-07-10 generative AI optimization guide adds a condition:

“In addition to the technical requirements for Search, a site must be included in Search generative AI features in Search Console to be eligible for display in generative AI features on Google Search.”

What you can act on is the direction of travel - the newer document adds a gate the older one didn’t mention, and it also adds a measurement surface the older one lacks (see the measuring section below). Check the Search Console setting exists for your property; that costs nothing.

The same pair of documents also disagrees about where AI traffic is reported. The older page says AI features traffic “are included in the overall search traffic in Search Console. In particular, they’re reported on in the Performance report, within the ‘Web’ search type.” The newer guide points at a dedicated report. Treat the older page’s statement as superseded on this specific point, and check which surfaces your property actually has.

What Google’s own documentation says about credentials, authorship and claims

The single most load-bearing page for this audience is Creating Helpful, Reliable, People-First Content, last updated 2026-10-05.

It frames evaluation as “Who, How, and Why”. For professional services, “Who” is the operative one:

“We strongly encourage adding accurate authorship information, such as bylines to content where readers might expect it.”

“Do bylines lead to further information about the author or authors involved, giving background about them and the areas they write about?”

A named lawyer on an employment law post, linking to a page with bar admissions, matters handled and practice areas, is the concrete form of that. A byline reading “The Smith Law Firm” is the opposite.

The same passage carries the warning that matters most for anyone planning to use AI to manufacture authority:

“However, avoid using deceptive authorship information. Fabricating creator profiles (such as by using AI-generated headshots, made-up names, or false credentials to make content appear as if it was written by human experts) is a form of deception. Any form of deception makes a page untrustworthy to both users and our automated quality systems, and is a signal of a low-quality page.”

Note the scope: this isn’t limited to consumer-facing review content, and it isn’t limited to AI Overviews. It’s Google’s general people-first content standard.

Legal work sits squarely in the category Google treats as highest-stakes:

“our systems give even more weight to content that aligns with strong E-E-A-T for topics that could significantly impact the health, financial stability, or safety of people, or the welfare or well-being of society. We call these ‘Your Money or Your Life’ topics, or YMYL for short.”

and

“For topics that could significantly impact people’s lives or well-being (YMYL topics), the content must be highly accurate and consistent with established expert consensus.”

Consulting and advisory work sits at the edge of this, since corporate advisory affects clients’ financial and legal exposure even where the end consumer isn’t the patient.

And the expertise test is first-hand, not credentialed-on-paper:

“Does your content clearly demonstrate first-hand expertise and a depth of knowledge (for example, expertise that comes from having actually used a product or service, or visiting a place)?”

The Expertise-to-Citation model

The premise is a single observation: generative answers assemble claims rather than rank pages. A page contributes to an answer insofar as it contains statements that are (a) attributable, (b) extractable without the surrounding rhetoric, and (c) corroborated somewhere other than the page itself. Everything else on the page is context that will be dropped.

Under that premise, I score a firm on five attributes, each 0–3.

1. Named authorship. 0: no bylines anywhere. 1: bylines to the firm or “the team”. 2: named individuals without linked credentials. 3: named individuals, each linking to a profile carrying verifiable credentials - jurisdictions admitted, matters, dates.

2. Claim extractability. This is the attribute most firms over-score, because good marketing prose feels strong. Score the sentence, not the page. 0: “Jane is an accomplished attorney dedicated to serving her clients.” 1: claims present but hedged. 2: most substantive claims are factual and specific. 3: statements are individually quotable without the surrounding paragraph - a number, a deadline, a named statute, a named outcome.

3. Credential specificity. 0: no jurisdiction mentioned. 1: practice areas listed without jurisdictions. 2: jurisdictions stated. 3: jurisdiction and specificity that would be wrong somewhere else - “California wage-and-hour class actions”, not “employment law”.

4. External corroboration. 0: nothing outside your domain asserts anything about you. 1: a directory profile exists but is incomplete or unclaimed. 2: independent third parties assert consistent facts. 3: independent third parties assert consistent facts and your own properties point at them, so the corroboration is machine-discoverable in both directions.

5. Structural machine-readability. 0: client-rendered content behind a bot wall. 1: indexable but no entity markup. 2: Organization markup with sameAs links to authoritative profiles. 3: server-rendered content, stable URLs, entity markup that matches the visible text.

Total, 0–15. The bands exist to direct effort, not to pass a threshold:

• 0–5 - rhetoric only. Nothing here is extractable. You aren’t competing; you’re invisible.
• 6–9 - partial. Claims exist but aren’t corroborated, or credentials aren’t specific enough to distinguish you from another firm in the same city.
• 10–15 - extractable. Claims are attributable, specific, corroborated and machine-discoverable.

The diagnostic value is in the shape, not the total. The near-universal pattern for established professional-services firms is high on attribute 2 and low on attributes 3, 4 and 5. Excellent, warm, un-specific marketing copy; no jurisdictions; a directory profile nobody has claimed; no entity markup. That’s precisely the condition in which an answer engine reaches for a directory instead of the firm, because the firm’s own pages contain nothing that survives being extracted.

So attribute 2 isn’t the lever. Most firms already have it. Attributes 3, 4 and 5 are where the movement is, and they’re also the cheapest - they’re usually a documentation exercise rather than a writing exercise.

On attribute 4, the mechanism is documented. Google’s Organization markup guidance defines sameAs as “The URL of a page on another website with additional information about your organization” - your professional-body registration, your industry directory profile, your regulatory filing record. The same page notes that properties like iso6523Code and naics are used “behind the scenes to disambiguate your organization from other organizations”, and recommends legalName where your registered name differs from your trading name. For a firm where “Smith & Co” is genuinely ambiguous, that disambiguation work is worth doing.

Note what Google’s guide says about structured data and AI: “Structured data isn’t required for generative AI search, and there’s no special schema.org markup you need to add.” Its value is that it makes your entity legible to the retrieval layer, not that it unlocks a feature. Also worth knowing: schema.org’s ProfessionalService sits under LocalBusiness and inherits hasCredential, knowsAbout, areaServed, award and ethicsPolicy - properties that cover much of the ground a professional-services firm’s entity description needs, and which Google’s LocalBusiness page doesn’t list among the properties it supports. So don’t expect a rich result from them.

Building the practice-area prompt map

The second piece of original work here is a method for deciding what to publish when you can’t publish everything. The instinct to cover every phrasing is exactly the pattern Google’s scaled content abuse policy names. The method below is designed to cut scope deliberately.

It’s built on the fan-out mechanism described earlier. If one user question fans out into a comparison, a procedural and a prevention retrieval, then a firm that has genuinely answered all three has covered the fan-out. A firm that has answered only one of them has covered a third of it.

Step 1 - decompose the practice area into buyer situations, not topics. “Employment law” is a category label. The retrieval happens at the level of a specific situation someone is in. For employment: classifying a worker; the first hire and handbook; a contractor misclassification claim; a wage-and-hour exposure; an executive departure; a diligence question during an acquisition. You’re looking for the nouns and verbs of a bad week, not for sub-practice-areas.

Step 2 - attach the jurisdiction modifier to each. This is the most useful step and the most commonly skipped. A correct answer to a California question is frequently a wrong answer to a New York one - and sometimes a wrong answer to a different California question, because the governing rule differs by city. If your content doesn’t name its jurisdiction, it’s competing in a broader and weaker space than the one you actually occupy.

Step 3 - tag each situation with a decision stage. Three, in my view, are enough: orienting (I need to understand this), comparing (I am choosing between options), deciding (I am about to act or sign). Deciding prompts carry the most value and are the most under-served. Almost every professional-services firm publishes for the orienting stage and nothing else.

Step 4 - predict the answer format for that stage. Orienting produces an explanation. Comparing produces a list with criteria. Deciding produces a recommendation with conditions and trade-offs. A page that will answer a deciding prompt therefore has to carry conditions and trade-offs; under this model, a page that doesn’t is a poor candidate for that retrieval however well it’s written.

Step 5 - mark the corroboration gap. For each prompt, ask whether a third party already publishes a clear answer. If a directory or an industry body already holds the slot, competing with them on your own domain alone is expensive - they have the domain authority and the aggregation. You either need a genuinely better answer than they publish, or you need to be present on their domain and be verifiable there.

Step 6 - cut. Score each situation on stage weight (deciding highest) times jurisdiction specificity, and publish the top tier properly rather than the whole list thinly. The cutting is the point. A six-page map where all six are excellent beats a forty-page map where none is.

A worked fragment, using employment law for startups, constructed by me for illustration - these are plausible prompt shapes, not measured volumes:

Situation	Jurisdiction	Stage	Expected answer shape	Corroboration gap
Contractor reclassification risk	California	Deciding	Recommendation with cost estimates and a decision date	Low - well covered by directories
Equity grant exercise and 83(b) timing	US federal + state	Orienting	Explanation with hard deadlines	Medium
Severance terms at a US acquisition	Delaware + buyer state	Deciding	Trade-off list against buyer protections	Low - trade press covers it
Handbook adequacy for a 12-person firm	State-specific	Orienting	Checklist	High - undifferentiated
Post-termination non-compete enforceability	State-specific	Deciding	Yes/no per state with exceptions	Medium

The pattern that jumps out: the deciding rows have the lowest corroboration gaps and the highest value, and they’re the rows a generic practice-area page is worst at answering. The orienting rows are the ones most firms already have, and they’re the least differentiating.

Pull your own from Search Console, or from whatever keyword tooling you use, and let volume adjust the ordering. The method gives you the structure; volume gives you the weights.

The directory wall: what the evidence supports

For high-intent legal prompts, the empirical picture is that directories, not firms, are what gets cited. I want to be precise about the status of that claim, because it’s exactly the kind of number that circulates without scrutiny.

One vendor study is InterCore’s State of AI Search Visibility for Law Firms 2026, published 2026-07-20. Its method is disclosed unusually well. It ran 540 high-intent legal queries - 6 practice areas across 30 US metros, in 3 phrasings - three times each, against Perplexity Sonar only, yielding 1,620 answers and 18,900 source citations. It reports that a legal directory was the first-cited source in 77.8% of answers, that directories took at least 51.8% of citations (published as a floor rather than a point estimate, because 36.0% of citation volume sat on domains the authors hadn’t classified and they declined to guess), that 254 distinct firm websites split the firm share, and that 34.2% of the firms the AI named had no page of their own among the sources cited.

Its most-cited domains were justia.com (9.7%), attorneys.superlawyers.com (8.07%), avvo.com (5.74%), lawyers.com (5.72%), expertise.com (4.91%), bestlawyers.com (4.27%), bestlawfirms.com (3.85%), lawyers.findlaw.com (2.16%), reddit.com (2.04%) and lawinfo.com (2.03%).

One structural observation survives even with those caveats, because it doesn’t depend on the dataset: in classic search, the same shape is already present. When I searched best employment law firm for startups California, the top results included trustanalytica.org, bestlawfirms.com, upcounsel.com, chambers.com and lawzana.com - three listicle or marketplace properties and two ranking directories - with individual firm sites appearing below them. (The search used an India region setting, so read that as indicative of the shape of the category rather than a ranking observation for any specific market.)

If directories already dominate the conventional results and the AI citation layer, the directory wall isn’t an AI artefact to be worked around. It’s the existing structure of this search category, carried forward.

The research layer has checkable errors

A second report in this space, the 2026 Legal AI Visibility Report from 5W Research and Haute Lawyer Network (April 2026), is worth discussing precisely because it illustrates what to watch for.

Two of its claims are checkable against primary documentation, and one is wrong. It advises firms to “unblock GPTBot, ClaudeBot, PerplexityBot, and Google-Extended in robots.txt”, describing this as “the single highest-ROI action in the entire report.” As set out earlier, OpenAI and Anthropic each document that search visibility is governed by a different agent from the training crawler, and Google scopes Search crawling to Googlebot and its preview controls rather than to Google-Extended. The instruction would change your training exposure without changing your search visibility.

That’s the useful lesson. When a report in your own category contains a technical claim you can check against vendor documentation in ninety seconds, check it. In a field this new, the error rate in secondary sources isn’t a minor issue - it’s the main thing to manage.

Your testimonials cannot become star ratings

Here is a constraint that catches nearly every professional-services firm, and one that rarely surfaces in writing about AI search.

Google’s Review snippet documentation (last updated 2026-09-08) contains a rule that reliably surprises product-focused marketers:

“If the entity that’s being reviewed controls the reviews about itself, their pages that use LocalBusiness or any other type of Organization structured data are ineligible for star review feature. For example, a review about entity A is placed on the website of entity A, either directly in their structured data or through an embedded third-party widget (for example, Google Business reviews or Facebook reviews widget).”

The LocalBusiness page repeats it from the other direction - aggregateRating and review are “only recommended for sites that capture reviews about other local businesses.”

If your law firm publishes its own client testimonials, marks them up, and embeds its Google Business reviews widget, those pages aren’t eligible for the star review feature. You can keep the testimonials on the page. You can’t use that markup to present your own site as an independent source of ratings about your firm.

This is structurally sensible. A review is only evidence if the reviewer isn’t the reviewed. It also means the review-markup playbook that works for a SaaS company doesn’t transfer, and that the review surface for professional services is necessarily external - directory profiles, the Business Profile itself, and independent platforms.

Two adjacent rules from the same page matter here. “Don’t aggregate reviews or ratings from other websites.” And:

“Don’t include fake or undisclosed incentivized reviews on your page or in your structured data markup. Examples include: Reviews that aren’t based on a genuine experience of a product or service; Reviews written in exchange for a benefit (such as money, discounts, vouchers, or free products) that don’t clearly and prominently disclose the incentivization.”

The Maps user-generated content policy describes how enforcement actually runs: “We use a combination of people and machine-learning algorithms to detect such content,” including scanning “for signals of suspicious user activity,” with contributions removed by automated models or escalated to trained operators. Attribution modeling on reviews is the kind of thing that reads as clever in a pitch deck and as a violation here.

Where the ethical line actually sits

For most professional-services marketing, the ethical line is discussed in terms of taste. It’s written down, and it’s narrower and more specific than the debate usually suggests.

The FTC’s Guides Concerning Use of Endorsements and Testimonials in Advertising, codified at 16 CFR Part 255 (source note 88 FR 48102, July 26, 2023), apply to you. The first thing to note is scope - many assume endorsement rules are about physical products:

“For purposes of this part, the term ‘product’ includes any product, service, brand, company, or industry.”

So a consulting firm’s case study, a law firm’s client quotation and a firm’s “best firm” badge are all in scope.

Material connections. The core rule:

“When there exists a connection between the endorser and the seller of the advertised product that might materially affect the weight or credibility of the endorsement, and that connection is not reasonably expected by the audience, such connection must be disclosed clearly and conspicuously.”

Note the qualifier: not reasonably expected by the audience. The rule isn’t “disclose all payments.” It’s “disclose payments a viewer would not expect.” And the Guides draw that line explicitly in the other direction too, at Example 10, where a company paid a non-profit association a reasonable fee for an evaluation: “Consumers would reasonably expect that marketers have to pay for this kind of certification. Therefore, there is no unexpected material connection … and the use of the seal without disclosure of the fee paid to the association would not be deceptive.”

Paid placement isn’t automatically unlawful. Undisclosed paid placement that the audience wouldn’t expect is. That’s a considerably more useful line than “never pay for anything.”

Expert endorsements. If your communication represents an endorser as an expert:

“the endorsement must be supported by an actual exercise of the expertise that the expert is represented as possessing in evaluating product features or characteristics which are relevant to an ordinary consumer’s use of or experience with the product.”

Example 5 in the same section is a cleaning service endorsing a cleanser. The service’s president made an expert endorsement and “the service must, in fact, have such experience, and have determined, based on its expertise, that the endorsed product’s cleaning ability is at least equal (or superior, if such is the net impression conveyed by the advertisement) to that of the leading competitors’ products available to consumers.”

For a firm, the analogous failure is a “best in the city” claim resting on marketing copy rather than on comparative work in that city and practice area. The comparison has to actually have happened.

Organizational endorsements. If your firm or a body endorses, the process matters:

“an organization’s endorsement must be reached by a process sufficient to ensure that the endorsement fairly reflects the collective judgment of the organization. Moreover, if an organization is represented as being expert, then … it must utilize an expert or experts recognized as such by the organization or standards previously adopted by the organization.”

A “best law firms” list published by a professional body, a trade publication or an association needs a stated, expert-run selection process behind it, or it’s an organizational endorsement without one.

Paid reviews. §255.2 Example 9 is about as direct as guidance gets:

“A manufacturer offers to pay genuine purchasers $20 each to write positive reviews of its products on third-party review websites. Such reviews are deceptive even if the payment is disclosed because their positive nature is required by, rather than being merely influenced by, the payment.”

Disclosure doesn’t cure payment-for-positive-outcome. It cures payment-for-review.

Intermediaries. This is the provision most often missed, and it covers the agency that sells you the manufactured mentions:

“Advertising agencies, public relations firms, review brokers, reputation management companies, and other similar intermediaries may be liable for their roles in creating or disseminating endorsements containing representations that they know or should know are deceptive.”

Bona fide independent evaluators. Example 14 sets out the line for rankings: if a company commissioned and paid for the analysis that produced its “fastest” claim, it “should clearly and conspicuously disclose its relationship to the testing company … If the ‘Data Speed Testing Company’ is not a bona fide independent testing organization with expertise … or it did not conduct valid tests that supported the endorsement message, the endorsement would also be deceptive.”

Transposed to your category: a ranking that selects firms by who paid isn’t a ranking. It’s advertising wearing a ranking’s clothing, and the FTC framework above is what makes that legible to a regulator.

And for lawyers, the ethics code catches something the FTC code does not

Federal advertising law focuses on disclosure: who paid, who endorsed, and what supports the comparison. It applies to law firms. Your jurisdiction’s professional conduct rules ask another question: is the statement itself accurate? That matters regardless of who’s speaking.

The American Bar Association’s Model Rule 7.1, “Communications Concerning a Lawyer’s Services”, reads in full:

“A lawyer shall not make a false or misleading communication about the lawyer or the lawyer’s services. A communication is false or misleading if it contains a material misrepresentation of fact or law, or omits a fact necessary to make the statement considered as a whole not materially misleading.”

• ABA Model Rule 7.1

Two things about that sentence matter for this guide, and I want to be precise about which part is the rule and which part is my reading of it.

The rule itself defines misleadingness by omission - not only by saying something untrue, but by leaving out “a fact necessary to make the statement considered as a whole not materially misleading.” The operative unit is the communication as a whole.

Here’s the problem when we write for AI answers. An assistant can lift the sentence that answers a question and leave out the qualifications three paragraphs below. You don’t control which fragment it selects, and you can’t correct a claim inside an answer that’s already been shown. You can control the sentence before you publish it.

So for firms in this category the practical rule is the same one the fintech guide reaches from a different direction: if you wouldn’t accept that sentence standing completely alone, as the only thing a reader ever sees, it isn’t finished. Put the limit in the claim. “Experienced in employment disputes” is a fragment. “Employment disputes heard in the state and federal courts in this metro area, between 2019 and 2025” is a claim that survives extraction.

Note also how this differs from the FTC material above. Rule 7.1 doesn’t ask whether you disclosed a material connection; it asks whether the communication is accurate and complete on its own terms. A page can be perfectly FTC-compliant - every fee disclosed, every comparison sourced - and still mislead when an assistant lifts one clause out of it. These are two different failure modes, and only one of them is about who paid.

And the search-platform line, separately

Independent of advertising law, Google states the AI-search version plainly in its 2026-07-10 guide, under the heading “Seeking inauthentic ‘mentions’”:

“our generative AI features can show what’s being said about products and services across the web … However, seeking inauthentic ‘mentions’ across the web isn’t as helpful as it might seem. Our core ranking systems focus on high-quality content while other systems block spam; our generative AI features depend on both.”

A campaign whose entire model is manufacturing mentions is relying on a surface Google has explicitly told you it won’t reward, while running a campaign the FTC has described as deceptive. Those are two independent reasons, and either one is sufficient.

Add the deceptive-authorship warning from the people-first content guidelines and you have three separate documented lines converging: don’t fabricate creator identity, don’t manufacture mentions, don’t manufacture positive reviews.

Measuring change without fooling yourself

Two first-party measurement surfaces exist and both have caveats you should read before you build a dashboard.

On the Google side, the 2026-07-10 guide names a dedicated report: “To measure how your content is performing in generative AI features on Google Search and Discover, use the Generative AI performance report in Search Console.” The existence of a separately named report is confirmed. Don’t assume it reports which specific pages were cited for which specific prompts. Treat its contents as something to inspect in your own property.

The same guide’s AI-features eligibility sentence is worth re-reading at this point: if inclusion in Search generative AI features is a precondition for display, then a site that hasn’t been included has no baseline to improve against, and a flat zero isn’t evidence that its content is weak.

On the Microsoft side, AI Performance in Bing Webmaster Tools (a public preview announced in February 2026) exposes Total Citations, Average Cited Pages, Grounding queries, page-level citation activity and visibility trends. Its authors are unusually careful about what those numbers aren’t:

“Average Cited Pages … does not indicate ranking, authority, or the role of any page within an individual answer.”

“Page-level citation activity … reflects how often pages are cited, not page importance, ranking, or placement.”

Read those two disclaimers as a pair and they rule out the obvious use of the dashboard. Citation counts tell you whether your pages are being referenced, and they deliberately refuse to tell you how well - and with no ranking or authority dimension exposed, the report can’t answer “am I winning”. On the local question, the post also notes that “businesses can register with Bing Places for Business to help ensure that key details such as address, hours, and contact information remain current and eligible for inclusion in AI-generated responses” - the Microsoft analogue of getting your local entity data clean.

Use citation counts to track change on your own site over time. I’d leave cross-firm comparisons out until a tool documents how it makes them. Citations and rankings measure different things. The platforms say that about their own reports.

Sources and References

• Google’s Guide to Optimizing for Generative AI Features on Google Search - Google Search Central. Last updated 2026-07-10 UTC.
• Creating Helpful, Reliable, People-First Content - Google Search Central. Last updated 2026-10-05 UTC.
• Google Search’s Guidance on Generative AI Content on Your Website - Google Search Central. Last updated 2026-10-01 UTC.
• AI Features and Your Website - Google Search Central. Last updated 2025-12-10 UTC.
• Local Business (LocalBusiness) Structured Data - Google Search Central. Last updated 2026-09-08 UTC.
• Review Snippet (Review, AggregateRating) Structured Data - Google Search Central. Last updated 2026-09-08 UTC.
• Organization Schema Markup - Google Search Central. Last updated 2026-09-08 UTC.
• Guidelines for representing your business on Google - Google Business Profile.
• How providers qualify for Local Services Ads - Google Local Services Ads.
• Maps user-generated content policy - Google Maps.
• Overview of OpenAI Crawlers - OpenAI.
• Does Anthropic crawl data from the web, and how can site owners block the crawler? - Anthropic. Published 2026-04-07.
• Perplexity Crawlers - Perplexity.
• Introducing AI Performance in Bing Webmaster Tools Public Preview - Microsoft Bing. Dated 2026-02 by URL path.
• 16 CFR Part 255 - Guides Concerning Use of Endorsements and Testimonials in Advertising - U.S. Federal Trade Commission, via eCFR. Source note 88 FR 48102, July 26, 2023; current eCFR text.
• Rule 7.1: Communications Concerning a Lawyer’s Services - American Bar Association, Center for Professional Responsibility, Model Rules of Professional Conduct. A model rule: states adopt and amend these rules, so the text binding you may differ. Read your jurisdiction’s version.
• ProfessionalService - Schema.org Type - schema.org.
• The State of AI Search Visibility for Law Firms 2026 - InterCore Research, 2026-07-20.
• The 2026 Legal AI Visibility Report - 5W Research / Haute Lawyer Network, April 2026.

This article is a marketing and search-visibility guide. It is not legal advice.