Search in the AI era: how a law firm gets named in the answer

Search has changed shape. For twenty years the question was where your firm ranked on a page of ten links. Increasingly there is no page: someone types “recommend a conveyancing solicitor in Guildford” into ChatGPT, Perplexity or the AI box at the top of Google, and gets an answer that names two or three firms. Either you are one of them or you are not in the conversation.

This article is about how those engines decide which firms to name, why most law firm websites give them very little to go on, and what to do about it. It is written for the person who owns the firm’s marketing decisions, not for a developer, though there is a list at the end you can hand to one.

The list became an answer

A conventional search engine returns a ranked list and leaves the choice to the reader. An AI engine returns an answer and has already made the choice. That difference matters more than any ranking factor.

When Google shows an AI Overview, the firms it names are the ones it could read well enough to summarise. When ChatGPT is asked for a solicitor, it searches the web with its own crawler, reads the pages it finds, and composes a reply that cites the sources it understood. Perplexity does the same and shows its sources by default. Microsoft Copilot draws on Bing. None of them shows ten results and lets the reader scroll.

There is no page two in an answer. There is barely a page one. The engine picks a handful of firms, and it picks them for reasons that have less to do with backlinks and keywords than with whether it could read the site, work out who was behind it, and find a sentence worth quoting.

How an engine decides who to name

Strip away the vendor names and every AI engine is answering four questions about your website, in this order.

Can I get in? Every engine runs a crawler, and every crawler checks robots.txt before fetching a page. OpenAI’s GPTBot and OAI-SearchBot, Anthropic’s ClaudeBot, PerplexityBot, Google-Extended. A surprising number of law firm sites block some or all of them. Sometimes that was a decision. More often it arrived with a WordPress security plugin or a theme update in 2023, when blocking AI crawlers was briefly fashionable, and nobody has looked since. A blocked crawler does not see a worse version of the site. It sees nothing.

Can I read it? Crawlers fetch HTML. Most do not run JavaScript, or run it grudgingly. If your site is built on a modern page builder that sends an empty shell and paints the words in afterwards, the crawler receives the shell. The firm’s carefully written practice-area pages are invisible. You can test this yourself: open a page, choose “view source” rather than “inspect”, and search for a sentence from the page. If it is not there, an AI engine cannot see it either.

Can I tell who you are? This is where most firms lose. An engine reading a page written for humans has to infer that it belongs to a law firm, that the firm has a name, an address, a phone number, an area it serves. It will try. But the reliable way is structured data: a block of JSON-LD in the page that says, in a vocabulary every engine reads, that this site is published by a LegalService called Such-and-Such LLP at this address, serving these counties, and that the same entity appears on Companies House, LinkedIn and Google. Engines weigh corroboration heavily. A firm that exists in three places it trusts, all agreeing, is a firm it will name. A firm that exists only on its own website is a firm it will hedge about.

Is there a sentence I can use? The engine is composing an answer to a question. It needs a sentence from your site that answers it. “A boutique practice delivering bespoke solutions for discerning clients” answers nothing. “Such-and-Such is a firm of solicitors in Guildford acting for owner-managed businesses on commercial property, employment and disputes” answers several questions at once, and it is the sentence that gets quoted. The same goes for FAQs: a page that asks “how much does a commercial lease review cost?” and answers it in two plain sentences will be cited in reply to that question. A page that avoids the subject will not.

Why law firm websites are particularly bad at this

Law firm sites are, as a category, written to impress a human who is already on the page. That was a reasonable brief when a site’s job was to convert visitors Google had already sent. It is the wrong brief now, and the failure modes are consistent:

• The opening line is a slogan. “Excellence. Integrity. Results.” A model cannot do anything with it.
• Practice areas are described in the abstract. “Our corporate team advises on a broad range of matters.” Which matters? For whom? Where? The engine wants the specifics a partner would give in the first minute of a call.
• Location is implied, not stated. The office address is in the footer image. The counties served are nowhere. An engine asked for a solicitor in Kent cannot connect you to Kent.
• The site was built for looks. Animations, page builders, client-side rendering. Beautiful to a person, blank to a crawler.
• Structured data is absent or wrong. Either there is none, or a plugin emitted a generic WebSite node with no organisation attached, or the address in the schema disagrees with the one on the contact page.
• robots.txt was never read by anyone at the firm. It is the one file that decides whether any of the above matters, and nobody owns it.

None of this is a criticism of the firms. It is the natural result of websites being a marketing purchase, made once, and then left alone while the ground moved.

What to do this quarter

The good news is that most of the fixes are cheap, and the order is clear because each one is worthless without the one before it.

1. Find out where you stand. We publish a free Website AI Visibility Check. Enter your homepage address and it reads the page the way a crawler does, along with your robots.txt, sitemap and llms.txt, and scores the four questions above. Every finding comes with the fix. It takes ten seconds and does not ask for an email, though you can have the result as a PDF if you want to forward it.
2. Open the door. Fix robots.txt so the search crawlers you want are allowed in. If the firm has a considered position on model training, express it by blocking the training crawlers specifically rather than everything with “bot” in the name. Check no page you care about carries a noindex tag left over from a redesign.
3. Get the words into the page. If the check reports that only a few dozen words reach the crawler, that is the conversation to have with your web agency, and it is the single biggest change available. Server-side rendering or a static build; the visual design need not change.
4. Say who you are, to machines. JSON-LD for the organisation, with name, address, telephone, description, logo, the areas you serve, and sameAs links to Companies House, LinkedIn and your Google Business Profile. Then a WebPage node per page and a BreadcrumbList. Your agency can do this in an afternoon; the check will tell you when it is right.
5. Rewrite the first sentence of every important page. What the firm is, for whom, where. Then a short FAQ of real client questions, answered in plain sentences, marked up as FAQPage. This is the part that needs a partner’s time rather than a developer’s, and it is the part that gets quoted.
6. Publish the small files. An XML sitemap declared in robots.txt. An llms.txt at the root: a plain-markdown index of your most important pages with a line on each, written for a model to read. An RSS feed if you publish articles. A canonical link on every page.
7. Be corroborated. Claim and complete the Google Business Profile. Keep the LinkedIn company page current and consistent with the site. Collect genuine reviews. Make sure your name, address and phone number are identical everywhere they appear. Engines confirm you against these sources before they name you.

What not to do

Do not stuff. Adding “best solicitor London” forty times worked on nothing in 2015 and works on less now. Engines are language models; they recognise copy written for them and discount it.

Do not generate filler. Publishing thirty AI-written articles on subjects your firm does not practise makes the entity harder to identify, not easier. One well-written page per real practice area, in the voice of the partner who does the work, beats any volume of generic content.

Do not block out of fear. There is a legitimate debate about model training and copyright, and a firm is entitled to a position on it. But a blanket block put in place without that debate is a marketing decision made by a plugin. Have the conversation, then set the file accordingly.

Whose job this is

At most firms the website belongs to marketing, the server belongs to a web agency, robots.txt belongs to nobody, and the IT provider has never seen any of it. That gap is why the cheapest fixes on the list above go undone for years.

We think this sits on the same agenda as the rest of the firm’s technology. Whether the engines your prospective clients ask can find you is a question about crawler access, rendering, structured data and file hygiene, and it has consequences for the firm’s growth that a partner should hear about in the same conversation as security, compliance and cost. That is the conversation our quarterly vCIO sessions are for, and it is why we built the check.

Run it on your own site first. Then run it on the three firms you lose work to, and see who the engines are choosing.

Being found is one half of AI in a law firm. Being ready to use it safely inside the firm is the other, and our AI Readiness Check covers that half in about six minutes.