How do law firms get recommended by ChatGPT (and Gemini, Perplexity and AI Overviews)?

AI assistants recommend law firms they can verify. When someone asks ChatGPT, Gemini, Perplexity or Google's AI Overviews for a family lawyer in Parramatta or a commercial litigation firm in Brisbane, the engine does not rank pages the way Google traditionally did. It reads the sources it trusts, looks for firms that are described consistently across them, and names the ones with specific, checkable detail: named lawyers, practice areas, locations, credentials and third-party mentions. If your firm is vague on its own site, inconsistent across directories, or blocked from the crawlers, it does not get named. The fixes are mostly content and housekeeping, not new technology. Each factor is below, with what to do this month and a 30-day checklist at the end.

How each AI engine picks its sources

The four engines behave differently. Treat them as four related jobs, not one.

ChatGPT

ChatGPT answers from two places: training data fixed at a point in time, and live web search for anything current or local. The live search draws on the Bing index and OpenAI's own crawling. OpenAI publishes its user agents: GPTBot collects training data, OAI-SearchBot powers search in ChatGPT, and ChatGPT-User fetches pages when a user asks for a specific site. The rules are on OpenAI's bots page. The implication for a law firm: if you are poorly indexed in Bing and OAI-SearchBot cannot reach your site, ChatGPT is working from memory and from what other sites say about you.

Gemini and Google AI Overviews

Both run on Google's index, so ordinary Google SEO is the entry ticket. AI Overviews appear above the normal results for many legal questions and link to the pages they drew from. Google's guidance is that there is nothing special to do beyond its standard advice: pages need to be indexable and eligible for snippets. See Google Search Central on AI features. Gemini also grounds answers in Google Search when a question needs current information. The token to know is Google-Extended, which controls whether your content is used for Gemini training. Blocking it does not remove you from Search or AI Overviews. It is listed on the Google crawlers overview.

Perplexity

Perplexity is a search engine first. Every answer is built from live web results and every claim carries a citation. It runs its own crawler, PerplexityBot, and a user-triggered fetcher, Perplexity-User, both described in its crawler documentation. It favours pages that answer the question near the top, and it cites directories and review sites heavily for "best lawyer" questions. If your firm appears in Perplexity, it is usually because a page about you, not just your own site, said something specific.

Branded questions versus non-branded questions

These are two different problems and most firms only solve the first.

A branded question is "Is Smith & Co Lawyers any good?" or "Who are the partners at Smith & Co?" The engine looks you up. You win this when your site, Google Business Profile, Law Society listing and reviews all say the same thing, and third parties corroborate it.

A non-branded question is "Who is a good employment lawyer in Melbourne for an unfair dismissal claim?" Nobody has asked for you. The engine builds a shortlist from sources it already trusts for that topic: directories, review platforms, legal guides, news coverage and, if your practice-area page is specific and well cited, your own site. You win this by being present in the places the engine reads, and by having a page that answers the exact question better than the alternatives.

Ask both kinds of question when you test, and do not assume a strong branded result means you are showing up for non-branded ones.

What a quotable practice-area page looks like

Most practice-area pages are written for a partner to approve, not for a machine to quote. They open with the firm's commitment to clients and reach the actual service three screens down. Here is the structure we build.

1. Direct answer in the first 120 words. Say what the service is, who it is for, where you do it and what it involves. "We act for employees in unfair dismissal claims in the Fair Work Commission. The firm is based in Melbourne and takes matters across Victoria." A reader and an engine both know what the page is about before they scroll.
2. A named lawyer with credentials. Who runs this practice area, their admission year, accredited specialist status if held, and a link to their profile. Named people with checkable credentials carry far more weight than "our experienced team".
3. Specifics a competitor would not write. Courts and tribunals you appear in, typical timelines, fee structures where you can state them. Interchangeable content does not get cited.
4. An FAQ section. Five to ten real questions, each answered in two to four sentences, in plain language. These are the lines most likely to be lifted into an AI answer.
5. Structured data. Mark the firm up as LegalService or Attorney (both are schema.org types; Attorney is a subtype of LegalService), each lawyer as a Person with jobTitle, worksFor and sameAs links to their Law Society and LinkedIn profiles, and the FAQ section as FAQPage. Google no longer shows FAQ rich results for most sites, but the markup still tells every crawler exactly which text is a question and which is the answer. The spec is at schema.org/LegalService.

One page per practice area per office. A single "Our services" page listing twelve areas gives the engine nothing specific enough to quote.

Crawler access: the mistake we find most often

None of the content work matters if the engines cannot read your site. Two things to check.

First, robots.txt. It should not block GPTBot, OAI-SearchBot, PerplexityBot or Google-Extended unless that is a deliberate decision. Some firms blocked GPTBot in 2023 over training data concerns and never revisited it. If you want to stay out of training sets but remain findable, block GPTBot and allow OAI-SearchBot. They are separate rules.

Second, and more common, the firewall or CDN. Cloudflare, Sucuri and similar services ship with bot protection that challenges or blocks unknown crawlers, and AI crawlers often fail those challenges silently. Your robots.txt says "allowed", the request never completes, and nobody notices because the site works in a browser. Ask whoever manages hosting to confirm the AI crawler user agents are on the allow list and to check logs for 403 and 429 responses to them.

The directories and review sites that get cited

For non-branded legal questions in Australia, the engines lean on a short list of third-party sources. Being accurate on these is worth more than another blog post.

You will see other directories recommended. Most are paid lead-generation sites with little authority. We do not recommend paying for them.

Consistent NAP across everything

NAP is name, address, phone. It is the most common reason a firm's listings fail to connect into one entity in the engine's understanding. "Smith & Co Lawyers", "Smith and Co", "Smith & Co Legal Pty Ltd" and "Smith Co Lawyers Parramatta" are four different businesses to a machine. So are "Level 3, 10 George Street" and "L3/10 George St".

Pick one format for the firm name, each office address and each phone number. Apply it to the website footer, the LegalService schema, Google Business Profile, Law Society listings, LinkedIn and Facebook. If the firm has rebranded or moved, fix the old listings rather than creating new ones. Engines merge a corrected record far better than they guess between two conflicting ones.

How to measure it

There is no Search Console for ChatGPT or Perplexity yet, so measurement is manual and monthly.

1. Write down 15 to 20 questions your actual clients ask, pulled from intake calls and enquiry forms. Mix branded and non-branded. Include location.
2. Ask each in ChatGPT (search on), Gemini, Perplexity and Google, noting whether an AI Overview appears. Use a logged-out session so your history does not shape the answer.
3. Record three things per answer: whether your firm is named, which competitors are named, and which sources are cited. The sources list tells you where to focus next month.
4. Repeat monthly. The score that matters is how many of your 20 questions name your firm.

Expect slow movement at first, then steps as new citations get picked up. One of our finance clients went from no organic visibility to being recommended as a top option across five AI platforms within seven months. Law tends to be slower because the directory layer is more entrenched, but the mechanism is the same.

If you would rather have the baseline done for you, our AI visibility check runs this process across the four engines and reports which sources each one relies on for your practice areas.

Where this fits with the rest of your marketing

None of this replaces ranking in Google. The engines pull from pages that are already indexed and trusted, so weak conventional SEO limits everything above. We explain the overlap in GEO vs AEO vs SEO. SEO gets you in the room. The work above decides whether you get named.

For a law firm the work splits into three streams: practice-area content and schema (AEO and GEO optimisation), engine-specific checks for ChatGPT and Google AI Overviews, and the directory and review layer, which is the part most firms skip. Our legal marketing page covers how we run all three together.

Your 30-day checklist

Thirty days will not get a firm recommended everywhere. It clears the blockers and gives you a measurement you can repeat. What happens after that is doing the content and directory work consistently for the next two quarters.