Why Do Only Some Accomplished Law Firms Appear in AI Recommendations?

What did the Visibility 360 study find?

Visibility 360 reported that 15 of 56 accomplished law firms, or 27%, appeared when AI systems were asked to recommend attorneys in specific practice areas. The firms sat in 14 US legal markets, and the queries ran on ChatGPT, Google AI Overview, Gemini and Perplexity. The study was released on April 2, 2026 under the title Why Only One in Four Top Law Firms Appear in AI Recommendations.

The firm behind the study sells authority-positioning and media services, and the release attributes the gap to missing credibility signals: media features, third-party recognition, expert commentary and structured answers to key questions. That is the authors' explanation. The release reports no test that isolates any one of those signals as the cause.

The headline is a count of firms. Fifteen of 56 is 26.8%, which the release rounds to 27%. Because the denominator is a small, hand-picked list, the figure describes that list and not law firms in general.

What did the Forward Push study measure?

Forward Push, a marketing agency that works only with law firms, ran the same best-lawyer questions on ChatGPT, Perplexity, Gemini and Claude over roughly six weeks between late April and early June 2026. The study covered 25 practice areas and 15 US cities and captured 44,411 answers. After dropping 6,238 lost to platform errors and quota limits, 38,173 clean answers remained.

Those 38,173 answers contained 96,988 mentions of more than 15,000 different firms. Dividing one by the other gives about 2.5 firm names per answer, a Firm AEO calculation and not a figure Forward Push states. The press release is the only public source Firm AEO could read in full, and the full report with city and practice-area tables is hosted by Forward Push itself.

The study counts answers and mentions, not accomplished firms. That makes it a measure of how the engines distribute attention, not of how many strong firms are missing.

Why can the same engines name some firms and miss others?

Engines build an answer from sources they can retrieve, and a firm is named when several of those sources describe it consistently. Forward Push read Perplexity's visible citations, more than 30,000 of them, and reported that 96% of Perplexity answers cited at least one legal directory while 68% cited the law firm's own website.

Credentials, verdicts and years of practice carry weight with a human reader and little weight with a retrieval system unless a retrievable page states them. A firm whose reputation lives in referrals and courtroom results can look thin online even when the record is strong.

Firm AEO's earlier posts cover the mechanics in detail: which directories each engine cites, how the retrieval step works, and why prompt wording changes who is named. This post stays on the question of what the two new studies add.

• •A firm can be accomplished and still have no profile on the directories an engine reads.
• •A firm can have profiles that disagree on address, phone number, practice areas or partner names.
• •A firm can publish its results on its own site only, with no page elsewhere that confirms them.
• •A firm can be named by one engine and absent from another, which is the next section's finding.

Do the four engines agree on which firms to name?

No. Forward Push reported that ChatGPT and Perplexity, given the same question in the same city and practice area, produced lists of named firms that overlapped only about 6%. A firm can lead on one engine and be a stranger on another, so there is no single AI result to win.

Behavior also differed in whether a firm was named at all. According to Forward Push, ChatGPT and Claude named an actual law firm in about 99% of answers and Perplexity in about 96%, while Gemini named a specific firm in only about 12%, defaulting to generic advice such as checking reviews.

For the 27% figure, this matters in a specific way. Visibility 360 counted a firm as appearing, and the release does not say whether a firm needed to show up on one platform or on all four. A firm could pass or fail depending on that definition alone.

How concentrated are the names the engines give?

Very concentrated. Forward Push reported that within any single city and practice area, the top three firms named captured about 34% of all mentions, while a long tail of roughly 64 other firms split the remainder, most named once or twice.

Firm AEO's arithmetic on that report: three firms sharing 34% average about 11% each, while 64 firms sharing the other 66% average about 1% each. Those averages are illustrative, since the real split is uneven, but they show the shape. Being one of 67 firms in the answer pool is not the same as being one of the three the engine returns to.

This is why a visible-or-invisible count understates the problem. A firm can be named occasionally and still capture a small share of the attention in its market.

Which sources sit behind the names?

For Perplexity, directories sit behind many of them. Forward Push reported that Super Lawyers appeared in 15.6% of the more than 30,000 Perplexity citations and Justia in 13.2%, with Best Law Firms, Best Lawyers, BCG Attorney Search and Chambers rounding out the top six sources.

That agrees in direction with the Citorian June 2026 test and with the 5WPR and Haute Lawyer index, which Firm AEO has covered separately. The percentages differ because the query sets, cities and practice areas differ, and the three should not be merged into one table.

One limit applies to every figure here. Perplexity is the only engine in the Forward Push set that shows its sources, so the source shares describe Perplexity. The press release does not report source shares for ChatGPT, Gemini or Claude.

What do the two studies not prove?

Neither study proves why a particular firm is missing, and neither is an independent audit. Both come from sellers of related services, both are press releases, and the full methods are not in the releases Firm AEO could read. A firm should treat the numbers as directional.

The most useful habit is to ask the same four questions of any vendor study before repeating its number.

• •What was the denominator: firms, answers, queries or citations? Visibility 360 used firms and Forward Push used answers, so the two cannot be converted into each other.
• •How were firms or prompts chosen? Visibility 360 does not state what made a firm accomplished, and neither release publishes the full prompt list.
• •What counted as appearing: one engine, all engines, a first-position name, or any mention?
• •Who sells what the study recommends? Both authors sell services that the findings favor, which is a reason to read closely and not a reason to dismiss the data.
• •Is the figure about the firm's market or only the sample? A 27% result for 56 selected firms does not predict the rate for any other firm.

How can a firm test whether its own record is visible?

A firm can test it in an afternoon by running its own best-lawyer prompts on each engine, logged out, and recording whether the firm is named, which sources appear, and which competitors are named instead. The result is one firm's sample, not a market statistic, and it answers the question the vendor studies cannot.

Firm AEO's monthly protocol uses three runs per prompt because engine answers vary run to run. The steps below are the short version, written as marketing tasks, and the firm's ethics counsel decides what any resulting claim may say.

• •List 10 to 20 prompts in the phrasing a client would use, naming the city and practice area, such as best family lawyer in a named city.
• •Run each prompt on ChatGPT, Perplexity, Gemini, Claude and Google AI Mode, logged out, three times, and date the log.
• •Record three things per answer: whether the firm is named, which sources are cited, and which firms appear instead.
• •Open each cited directory profile for the firm and check that the name, address, phone number, practice areas and attorney names match the firm's own site.
• •Mark every result the firm can prove with a page outside its own website, such as a directory profile, bar listing or news item, and every result it cannot.
• •Fix the gaps in that order, then run the same prompts again after the engines have had time to re-crawl.

What can a firm say publicly about these studies and its own results?

A firm can report what it measured and name the limits, and it should not turn a vendor figure into a claim about itself. ABA Model Rule 7.1 bars false or misleading communications about a lawyer's services, and an unsourced statement that AI recommends the firm, or a ranking the firm cannot substantiate, is the kind of statement the rule reaches.

Safe wording ties each claim to a dated test: the prompts, the engines and the date. Unsafe wording promises outcomes, such as saying the firm will be recommended, or treats a sample result as a general fact. Many states add their own advertising rules, and Firm AEO covers the common ones in its bar advertising post. The firm's ethics counsel decides.

None of this changes the measurement advice. A dated log of prompts and answers is useful internally and supplies the substantiation a public statement would need.