Every conversation I’ve had with a legal marketing team this quarter has come back to the same question: what does AI search mean for our website?
I wanted to look into this properly. And who better to talk to than the LLM itself. So, I sat down with Claude (Anthropic’s AI) and asked it to talk me through what’s actually happening to law firm websites and the things legal marketers need to be aware of.
HELEN FOORD: So, everyone is talking a lot about “getting ranked for LLMs” but what, in your opinion, is the real position we’re facing? Has anything really changed – I mean, surely good copywriting is good copywriting.
CLAUDE: Yes and no. The principles of good copywriting haven’t changed: clarity, specificity, real claims with evidence. What’s changed is the cost of vague writing.
Law firm web copy used to be skimmed by a human prospect already warmed up by a referral or a directory ranking. The copy’s job was to confirm credibility, and waffle and jargon slid by because nobody read it closely.
LLMs don’t skim. They extract. “Leading firm with deep expertise” is invisible to them. “Acted for X on Y in Z” is extractable. The copy that was always weak now actively fails to exist.
HF: So can you explain in a nutshell how search has changed?
CLAUDE: The model has flipped. Search used to be: type a query, get a list of links, click around, decide. Now it’s: ask a question, get an answer. The list is gone. Someone else has done the comparison for you.
Two things follow. First, your website is no longer the front door. It’s downstream of being mentioned in the answer. Second, the answer is built from a wider pool than just your site: directories, reviews, articles, social. Around 85% of the sources behind AI brand mentions are third-party sites.
You’re being talked about, not searched for.
HF: And I can imagine there’s a change in traffic as well? Is it true that fewer people are clicking through to websites now?
CLAUDE: Yes, and the data points one way. Clicks are down across the board. AI Overviews take up most of the top of the Google results page, and people get their answer from ChatGPT or Claude without leaving the chat. For law firms, fewer GCs will land on your website to “have a look.”
But the people who do click through are further along. They’ve already shortlisted you in the answer. The visit is no longer reconnaissance, it’s verification. That changes what the page has to do. The traffic figure looks worse. The conversion picture often doesn’t.
HF: So does this mean that law firms are getting more qualified enquiries than they were?
CLAUDE: Not automatically, no. The funnel shift only delivers warmer enquiries if you’re appearing in the answer to begin with. Plenty of firms aren’t, because they haven’t done the third-party work LLMs draw on: directories, PR, reviews, legal media coverage.
For firms that are cited, yes, the visitor who clicks through is closer to deciding. But the website still has to close, because they arrive half-decided.
The gap is widening between firms that have done the work and firms hoping traffic will bail them out.
HF: Interesting. So, what developments have we seen in this field over the last couple of months in particular?
CLAUDE: Two shifts stand out.
First, SEO and AI citations are decoupling. Ahrefs this year found only 38% of AI Overview citations come from pages ranking in Google’s top 10, down from 76%. And 80% of LLM citations come from pages that don’t rank in Google’s top 100 at all. Ranking and being cited are no longer the same thing.
Second, the platforms fragment. Only around 11% of domains are cited by both ChatGPT and Perplexity. One strategy doesn’t fit all.
The counterweight: AI-referred visitors convert at roughly 4.4 times the rate of organic search.
HF: I was interested to see that you went off and searched lots of sources to give me that answer. What can you tell me about how LLMs decide on sources and what that means for law firms?
CLAUDE: LLMs run several sub-queries in parallel, retrieve content, then weight it. Three signals tend to dominate: clarity of the answer on the page itself, corroboration from other trusted sources, and recency.
For law firms, that means the sources that matter most are not always your own: Chambers, Legal 500, the legal press. These are exactly the kind of authoritative, regularly updated third-party voices LLMs lean on. Directory submissions and legal PR aren’t BD ticks anymore. They’re feeding the machine that decides whether your firm gets named.
Your website still matters. But what people say about you matters more.
HF: Right, so it’s not just down to the copywriting on the page, it’s also about your wider marketing communications strategy.
CLAUDE: Yes, and the strategic implication runs deeper than people realise. For years, law firms have run directories, website, PR, BD, and social as separate workstreams, often with different teams and different agencies. LLMs collapse those silos. They read your firm as a single entity across every surface, and they reward coherence.
If your Chambers ranking says one thing, your website says another, and your partners’ LinkedIn says a third, the signal blurs. You don’t get cited.
The firms that win in AI search are the ones whose story holds together across the whole comms mix.
HF: Has this had an impact, then, on how law firm practice area pages need to be written?
CLAUDE: Yes, three big changes.
First, ditch the credentials opener. Around 44% of LLM citations come from the opening 30% of a page. “We are a leading practice” is wasted real estate. Open with what you actually do, for whom, and where.
Second, make claims extractable. Named deals, jurisdictions, sectors, values. LLMs can quote “recovered $2.3bn in an LCIA arbitration.” They can’t quote “successful disputes practice.”
Third, structure the page around questions a prospect would actually ask. FAQs and direct-answer sections get cited more than dense paragraphs.
Practice pages built like brochures don’t survive this shift.
HF: What about the role of lawyer biographies then? Is that something we need to worry about for LLM search?
CLAUDE: Yes. Bios sit at the heart of how LLMs answer “who’s good at X” questions, and most law firm bios are badly suited to the job. The standard opener is a wall of awards and rankings, then a few generic lines about practice areas, then a list of credentials. None of it is extractable.
What LLMs need: who this person acts for, what kind of work, where, and in what sectors. Front-loaded. Specific. Named where you can name. Consistent with their Chambers entry, LinkedIn, and Lexology contributions.
The pedigree paragraph can stay. Just stop opening with it.
HF: Are there other pages in a law firm website we need to think about next?
CLAUDE: Two stand out.
First, insights and thought leadership. This is where the firm has a voice on substantive legal issues, and it’s the content most likely to be picked up by Lexology, Mondaq, and JD Supra, which amplifies LLM exposure further. Treat insights as long-term assets, not one-off PR. Make them findable, dated, and tied to the practices they support.
Second, deals and case studies. They’re concrete by nature: named values, jurisdictions, structures. Most firms either don’t have them or bury them three clicks deep. Move them up. They’re some of the most quotable content on your site.
HF: I’d like to talk a little more about ‘sales’ in the sense of how law firm websites can better support conversion of enquiries. I’m aware that LLMs are starting to do more agentic shopping type things. Is this something law firms can learn from?
CLAUDE: Yes, two big lessons.
First, agents abandon friction instantly. Humans conditioned by AI search are getting closer to that pattern. Law firm websites are routinely worse at conversion than any decent e-commerce site, and the gap is becoming visible.
Second, agentic shopping rewards specificity about how engagement works. Most firms hide their process behind “get in touch.” Show it. Explain how scope, fees, and conflicts get worked through. Anticipate the questions a prospect would ask before instructing.
Law firms still treat the website as a brochure. The site needs to start selling.
HF: I have another question. Do all LLMs search the same way or do firms need to look at each of them differently when planning their GEO and AEO strategy?
CLAUDE: They differ, but probably less than the panic suggests. ChatGPT pulls from Bing, Perplexity from Google, Gemini and AI Overviews from Google’s own stack. Source preferences vary too: Gemini favours Google reviews, ChatGPT references certain media sources more heavily.
But the overlap in citations between platforms is genuinely low. Around 11% of domains are cited by both ChatGPT and Perplexity. So one strategy doesn’t blanket all of them.
The practical answer: get the fundamentals right (extractable content, third-party authority, consistent positioning), then check your analytics to see which platforms actually drive your traffic. Optimise from there.
HF: You’ve said check your analytics but what’s the best way to learn about the platforms and what’s driving traffic?
CLAUDE: Three steps, in order.
First, the free stuff. Set up a custom channel group in GA4 to track referrals from ChatGPT, Perplexity, Gemini, Claude, and Copilot. Set up Bing Webmaster Tools too. Bing now shows you the actual queries bringing Copilot traffic, which is the only direct visibility any platform gives – but this is really big news considering how most law firms and big corporates are wedded to Copilot as an LLM.
Second, run the prompts yourself. Ask ChatGPT and Claude what your prospects would ask. Note what comes back. Note who gets cited.
Third, once you know what you’re tracking, consider a monitoring tool. Profound, Peec AI, or Otterly are established names. Don’t buy before you’ve done the first two.
HF: Interesting. And what about keywords and searches? I’m guessing the approach we had for optimising for keywords doesn’t work now?
CLAUDE: It still works as a starting point, but it isn’t the destination anymore.
The average ChatGPT query is around 60 words, compared with four on Google. People give LLMs full context: who they are, what kind of matter, in which jurisdiction. You can’t target that as a single keyword.
The shift is from keywords to topics and questions. Instead of optimising for “energy disputes lawyer,” map the specific questions a prospect would ask, and build content depth across them. Own a recognisable territory rather than chasing isolated terms.
HF: I’m interested to discuss how LLMs assess trust. Many clients choose lawyers based on referral and recommendation. How does an LLM know to mention or include a law firm based on trust?
CLAUDE: LLMs can’t replicate referrals because the conversations happen privately. What they can do is read the public residue of trust: who’s named as expert by sources the model treats as authoritative.
For law, that residue is concentrated in places like Chambers and Legal 500, which exist precisely to translate confidential peer and client feedback into published rankings. Legal press, awards, and consistent thought leadership feed the same signal.
This is why directory work has become more valuable, not less. The methodology that makes directories trusted by clients is the same thing that makes them trusted by LLMs.
HF: I can see there’s a lot of editing work for those law firms with existing websites. If you were to suggest the top three action points, what would they be?
CLAUDE: One. Rewrite the front 30% of every key page. Practice areas, sector pages, lawyer bios. Lead with specifics, not credentials. This is where LLMs do most of their citation work, and where most law firm copy is at its weakest.
Two. Audit consistency across surfaces. Website, Chambers entry, Legal 500 submission, LinkedIn, Lexology. LLMs read them as one signal. If they say different things, you blur.
Three. Build the conversion path. Surface people. Clarify how engagement works. Make the next step obvious. If AI search is sending you warmer prospects, your website needs to be ready to close.
HF: And what about those law firms contemplating a brand new site? What would you suggest for them?
CLAUDE: A new build is a rare chance to start from the right place. Three things to bake in.
One. Architect the site around the questions prospects ask, not the firm’s internal structure. Cross-sections of sector and practice often matter more than practice taxonomy. Build the IA accordingly.
Two. Get the technical scaffolding right from day one. Schema markup, structured data, an llms.txt file, clear date stamps. Retrofitting this later is painful. Building it in is cheap.
Three. Design the content programme alongside the site, not after launch. The site is one node. Insights, directories, and PR feed the same machine.
HF: Are there any questions you think I should have asked you and haven’t? Anything you’d like to add?
CLAUDE: One thing worth adding: this shift is more meritocratic than SEO ever was.
In Google-era SEO, big firms with big budgets and decades of domain authority won by default. LLMs care less about that. They care about topical authority: who genuinely owns the question being asked. That’s good news for specialist boutiques and chambers, who have always had real expertise but limited reach.
The flipside is that performative thought leadership is finished. Publishing for the sake of publishing won’t get you cited. The content has to actually say something. That sorts firms quickly into two camps.
HF: What do you mean by “say something”?
CLAUDE: Content that takes a position. Most legal thought leadership rehearses what’s already known: summaries of new legislation, lists of considerations, balanced overviews. It’s safe and it’s invisible to LLMs. There’s nothing to extract.
Content that says something names a specific view, draws on real practice experience, and could provoke disagreement. “Most firms are reading the new Procurement Act as giving more flexibility on frameworks. We disagree, and here’s why” is extractable. “The new Procurement Act is significant” isn’t.
If a competitor could publish the same article verbatim, it wasn’t saying anything.
HF: Interesting. So if this is leveling the field between big and small law firms, can we expect to see more of the smaller law firms (who perhaps have less internal politics and a clearer tone of voice) rising to the top of results?
CLAUDE: Yes, in the niches. No, across the board.
Smaller firms can dominate specific topics where they have genuine depth: energy disputes in offshore wind, ATE funding, art law, a single regulatory niche. Less internal politics and clearer positioning help them move faster on the kind of content LLMs reward.
But scale still helps for general queries. The Magic Circle has decades of accumulated citations and press coverage that LLMs read as authority. That won’t vanish overnight.
The middle squeeze is where it gets uncomfortable. Mid-sized firms without a clear niche or top-tier brand depth are the most exposed.
HF: And finally, what can you see on the horizon? Where do you think this subject is going over the next six months and year? And what lessons are there for law firms from this?
CLAUDE: Three things are coming in the next twelve months.
First, agentic AI moves from novelty to default for B2B research. In-house teams and procurement will increasingly delegate “shortlist three firms for this mandate” to an AI agent. The agent does the research, synthesises, and presents options. The firms that make those shortlists win the conversation before a human even gets involved. The firms that don’t will not know they were excluded.
Second, the platforms will give us more visibility. Microsoft already shows Copilot search queries in Bing Webmaster Tools. OpenAI, Anthropic, and Google will follow. Within a year, monitoring AI citation will be table stakes, not an early-mover advantage.
Third, the discipline becomes continuous. Citation drift is already running at 40-60% month-to-month. This isn’t a project. It’s an ongoing programme, with quarterly reviews of how the firm appears across platforms, what’s changing, and what the data is telling you to adjust.
For law firms, three lessons follow.
One: start now. The firms that wait until this is “proven” will be twelve months behind. Closing that gap will be expensive.
Two: stop treating digital, BD, directories, and PR as separate disciplines. They feed the same machine. Run them that way.
Three: be honest about what your firm actually owns. Generic positioning was always weak. AI search makes it terminal. The firms that will win are the ones that know exactly what they are and aren’t, and have the confidence to say so.
What strikes me about this conversation is how little of it is actually about AI. It’s about specificity and about being honest. About firms knowing exactly what they’re good at, and saying it in a way that their targets would, were they asking the questions.
It seems to me that AI search isn’t doing anything new… it just exposes the firms that have been hiding behind generic language for years, and rewards the ones that have been doing the harder tone of voice work.
If your website reads like a brochure, the next twelve months will be uncomfortable. If it reads like the lawyers that actually do the work, and the clients you do it for, you’re already closer to winning than you might think.
