Do Directories Own AI Answers About Lawyers? Four Studies, Four Answers

If you are responsible for a law firm's marketing budget, someone has probably already shown you a slide that says AI assistants get their information about lawyers from directories. Depending on whose slide it was, the figure was 86%, or 96%, or something more modest. If it was a different presenter, you may have been told the opposite: that AI mostly cites firms' own websites, and that directories are a shrinking part of the picture.

Both camps have numbers. Both camps have something to sell. And a law firm deciding whether to renew a directory subscription, or whether to invest in its own site, deserves better than a choice between slides.

My proposition: the 2026 studies mostly aren't disagreeing. They are counting different things. Once you see what each one counts, the useful question changes from "do directories dominate?" to "for the questions my clients actually ask, which sources does AI lean on, and am I present in them?"

The four studies

I looked at four studies published in 2026 that try to measure where AI assistants get their information when people ask about lawyers. All four are US studies. All four are published by organisations with a commercial interest in the answer. That doesn't make them wrong, but it should shape how you read them.

Martindale-Avvo. The directory operator's research page states that "86% of AI search results for legal services cite our legal directory networks". The page names ChatGPT, Perplexity and Google AI Overviews, but it doesn't set out the sample, the questions, the dates or what "cite" means. The detail sits in a whitepaper behind a download form. Martindale-Avvo sells the directory listings whose visibility the study promotes.

5W, Legal AI Visibility Index (version 2). The PR and AI-visibility agency asked 24 questions three times each across four AI engines through their APIs, 288 answers in all, on a single day (26 October 2026). Directories and rankings supplied 30.5% of the citations. State bar associations were named in 57.3% of answers, and Martindale-Hubbell and Avvo in 47.6% and 46.2%; the confidence intervals overlap, so the study says it can't separate the three. The most-cited single domain was Wikipedia, at 6%. 5W notes that this version replaces a June version and that the two are not comparable. 5W sells AI-visibility services.

Forward Push. The law-firm marketing agency reports the largest sample: 38,173 answers analysed from ChatGPT, Perplexity, Gemini and Claude, across 25 practice areas and 15 US cities, collected between late April and early June 2026. Its directory figures are reported for Perplexity: Super Lawyers was 15.6% of citations and Justia 13.2%; 96% of answers cited at least one legal directory; and 68% cited a law firm's own website. It also reports that Gemini named a firm in about 12% of answers, against roughly 99% for ChatGPT and Claude. Forward Push sells marketing to law firms.

Taqtics. Another law-firm marketing agency counted 236 citations from 26 searches across eight US metro markets in June 2026, and sorted each by who owned the page. Firms' own websites took 67%, directories about 21%, earned press about 11%. The write-up doesn't say which AI assistants were tested. Its author describes it as a measurement, not a general rule.

Why the numbers look contradictory

Line the headline figures up and they appear to range from 21% to 96%. But they are at least three different measures.

Share of answers. Forward Push's 96% is the proportion of Perplexity answers that cited at least one directory. One directory link in an answer with ten citations counts as a yes.

Share of citations. Taqtics' 21%, 5W's 30.5% and Forward Push's 15.6% for Super Lawyers are proportions of all the links cited. On that measure a directory can appear in almost every answer and still be a minority of what gets cited.

Undefined. Martindale-Avvo's 86% is of "AI search results", a unit the page doesn't define.

The most revealing figures come from the same study. In Forward Push's Perplexity data, 96% of answers cited a directory and 68% cited a firm's own website. Those aren't competing claims. They describe answers that typically draw on both. An assistant asked for a good family lawyer in a city may cite a directory to establish who is credible and a firm's own pages to describe what it does.

The studies also differ in ways that would move the numbers even if they used the same unit. They ask different kinds of questions ("best lawyer in Houston" is not "what is the ABA"), test different engines, use APIs or consumer apps, run once or repeatedly, and sample different cities. 5W found the engines diverged sharply on specific names: Gemini named Martindale-Hubbell in 62.5% of its answers, Perplexity in 20.8%. Averaging across engines hides that.

Interpretation

Here is my reading, and I want to be clear that it is interpretation, not a finding.

First, "do directories dominate?" is the wrong question. On the evidence published so far, directories are present in a very large share of AI answers about choosing a lawyer, and firm websites are also cited often. The share depends heavily on how you count.

Second, the kind of question probably matters more than the averages. Questions that ask an assistant to choose ("best", "top", "recommended") seem likely to pull in third-party sources that signal credibility. Questions about how a legal process works plausibly draw more on explanatory content. None of these studies separates the two cleanly enough to give you a number, but it's a sensible hypothesis to test for your own practice areas.

Third, engines are not interchangeable. A finding about Perplexity is not a finding about ChatGPT, and almost none of this is a finding about Google AI Overviews, which 5W's study barely saw (4 Overviews in 72 searches).

Implications for law firms

Directory decisions shouldn't be made on headline figures. A subscription renewal deserves the same scrutiny as any other channel: which directories appear in answers to your clients' questions, in your jurisdiction, on the assistants your clients use.

Your own site still matters. Even in the study most favourable to directories, about two-thirds of Perplexity answers cited a firm's own website. Directories may help establish that a firm is credible; firm sites are often where the description of the firm comes from. In 20 years of working with law firms, I've rarely seen a channel question that turned out to be either/or.

Accuracy in directories becomes more important, not less. If AI assistants lean on directory profiles, an outdated practice-area list or a stale profile is no longer just a missed lead. It may be what an assistant repeats about you.

Australian firms should be cautious about importing US numbers. All four studies measured US markets and US directories. The directories, professional bodies and client behaviour that matter in Australia are different, and none of these studies measured them.

What we don't know

• How any of these figures translate to Australian, UK or other markets.
• Whether being cited in an AI answer leads to enquiries. None of the studies measures clicks, calls or instructions.
• How stable the figures are over time. 5W measured on a single day; AI products change models and retrieval often.
• What the full Martindale-Avvo and Forward Push methods are; I could read only the public pages and press releases.
• How consumer apps differ from the API outputs 5W tested.

Confidence summary

What to do this quarter

1. List the questions. Write down ten to fifteen questions your clients actually ask before they choose a lawyer, split between "who should I use?" and "how does this work?".
2. Check the sources. Ask them on the assistants your clients use, more than once, and note which sources are cited. Treat a single run as an anecdote.
3. Audit your profiles. For every directory that shows up, check that your profile is accurate, current and consistent with your website.
4. Read before you renew. When a directory or vendor shows you a figure, ask what was counted, on which engines, in which market, how many times, and who paid for the study.
5. Keep investing in your own pages. Make sure they describe what you do clearly enough to be quoted.

What to avoid

• Cancelling or doubling a directory spend because of a single headline percentage.
• Treating a US study as evidence about Australian clients.
• Buying an "AI visibility" product whose evidence is a study published by the same vendor, without reading the method.

Conclusion

The argument about whether directories own AI answers is mostly an argument about units. On the evidence available, the answer for most firms is that AI leans on directories and firm websites together, in proportions that depend on the question, the engine and the market. That's less exciting than a single number. It's also far more useful, because it tells you what to check.

FirmRanker, the research platform I founded, is designed to test exactly this kind of question: which sources AI systems rely on when they discover and recommend law firms. Its methodology explains how it measures that. At Practice Proof, we help firms decide where directories, their own sites and other channels fit in their wider digital and AI visibility work.

FirmRanker was founded by Dan Toombs. Practice Proof, also founded by Dan, provides digital and AI visibility services to law firms.