A growing share of legal research now ends inside an answer that never sends a click. Being named in that answer is a different discipline from ranking beneath it — and nobody can buy their way into it.
Generative engine optimisation is the work of making a firm the one an AI assistant names. When someone asks ChatGPT, Gemini, Perplexity or Google’s AI Overviews who to call about a case, the system composes an answer from sources it trusts. Being one of those sources depends on three things: whether the model can identify your firm as a distinct, real organisation; whether your content is written so a clean answer can be lifted from it; and whether independent sources elsewhere on the web corroborate what you say about yourself.
Models repeat what several independent sources agree on. A claim that exists only on your own website rarely survives the trip.
None of this is buyable. There is no placement, no bid and no ranking dashboard, which is exactly why the market is full of confident promises. What there is instead is a set of conditions that make citation measurably more likely, and the discipline to keep checking whether it is actually happening.
Four reasons, in roughly the order they need fixing.
The model cannot confidently tell that the firm on your website, the one in a directory listing and the one mentioned in a news piece are all the same organisation. Inconsistent names, addresses, practice descriptions and missing structured data all break that connection.
Content written as a continuous persuasive essay gives an assistant nothing self-contained to quote. Passages that lead with a direct answer and then support it are what get extracted — which is why a well-built FAQ outperforms a long article here.
If your firm is discussed nowhere except your own domain, there is nothing to confirm. Directory profiles, legal publications, local press, podcasts and forum discussion are the corroboration layer, and it is the part most firms have none of.
A different phone number in three directories, an old office address still listed, a practice area on one profile and not another. Each inconsistency lowers confidence in the entity, and confidence is what decides whether you get named.
Five stages. The third and fourth are where most of the work lives.
Get the firm consistently and unambiguously described everywhere it appears — name, locations, practice areas, people, contact details. This is unglamorous reconciliation work and it is the foundation everything else rests on.
Schema that states plainly what the organisation is, what it does, where it operates and who its people are, so the machine-readable version of the firm matches the human one.
Pages written to answer a question directly and then support the answer. Real questions, answered in full, in language a model can quote without mangling.
The corroboration layer: legal directories, publications, local press, podcasts and industry discussion. Slow, relationship-driven work, and the part that separates firms who get cited from firms who merely publish.
A fixed set of prospect questions run monthly across each assistant, logging whether the firm is named and which sources are cited instead.
Not tiers or upgrades — this is the work.
Name, address, practice areas and attorney details made consistent across every source that describes the firm.
Organization, service, person and location markup, kept accurate as the firm changes rather than set once and forgotten.
Pages and FAQ sets built around the questions prospects actually ask, formatted for clean extraction.
Directory accuracy, publisher outreach and industry presence — the sources a model reads to decide who is real.
Monthly manual testing across ChatGPT, Gemini, Perplexity and AI Overviews against a fixed question set.
What the assistants say about your firm when asked directly, including anything wrong that needs correcting at the source.
Conventional search has an accountability layer: impressions, positions and clicks, reported by the search engine itself. AI answers have nothing equivalent. No vendor publishes citation data, and the same question asked twice can produce different answers depending on phrasing, region, session and model version.
So we measure it the only honest way available, which is manually. Each month, a fixed list of the questions a prospective client would genuinely ask — who to call for a specific case type in a specific city, how a kind of claim works, what something costs — run across each major assistant. We record whether your firm is named, and every source cited instead.
That list of competing sources is the most useful thing the exercise produces. It is your outreach target list, written by the model itself.
Two things follow from this that we would rather say now than in month four. Results are directional rather than precise, and month-to-month movement can be noise. And nobody — us included — can guarantee that an assistant will name your firm, any more than we could guarantee a Google ranking. What we can do is the work that makes it likelier, and show you the evidence either way.
Directional by nature, and reported as such.
It follows how much research a client does before calling. Pick yours.
GEO — generative engine optimisation — is the work of getting a firm named and cited inside AI-generated answers, in ChatGPT, Gemini, Perplexity and Google's AI Overviews. SEO aims at a ranked list of links; GEO aims at being the source the model draws on when it composes a reply. The underlying work overlaps heavily, but the target is different: rankings reward the best page, citations reward the best-established and most corroborated entity.
Because a growing share of legal research now ends inside an answer that never sends a click. Someone who asks an assistant which firm to call for a truck accident in their city may act on that reply without visiting a single website. If the answer names three firms and yours is not among them, your ranking was never consulted. Existing search visibility helps, but it does not automatically carry over.
Three things, roughly in order. The model has to be able to identify your firm as a distinct entity — consistent name, location, practice areas and details across the web. It has to find content it can lift a clean, self-contained answer from. And it has to find other sources that agree with you, because models strongly prefer claims corroborated across several independent places. That last point is why a claim appearing only on your own site rarely gets repeated.
No, and be wary of anyone who says otherwise. These systems produce different answers to the same question depending on phrasing, session, region and model version, and none of them expose a ranking you can buy or control. What can be committed to is the work that demonstrably makes citation more likely, and honest measurement of whether it is happening.
Manually and repeatedly, because there is no analytics product for this. Each month we run a fixed set of the questions a prospective client would actually ask across each major assistant, record whether the firm is named, and log which sources are cited instead. That competing-source list is the most valuable output, because it becomes the outreach target list — those are the places that need to mention you.
Only with a clear reason, and know which bot you are blocking. Several vendors run separate crawlers for training and for live search retrieval. Blocking the training crawler while allowing the retrieval one is a coherent position: you stay citable in answers without contributing to model training. Blocking both makes citation far less likely. For a firm trying to be found, blocking everything is usually the wrong trade.
No. It sits alongside it, and much of the foundation is shared — structure, schema, entity consistency and genuinely useful content serve both. What differs is emphasis: GEO leans harder on third-party corroboration and on formatting content so a passage can be extracted cleanly. We run them together because separating them means paying twice for the same groundwork.
It is early, and we would rather say so than oversell it. The honest position is that the foundational work — entity establishment, structured data, factual consistency, third-party presence — is worth doing regardless, because it strengthens conventional search at the same time. What we will not do is bill for a dedicated AI-search programme as though the measurement were as mature as Search Console. It is not, and we would rather scope this properly than fashionably.
A free growth plan: we run the questions your prospects would ask across each major assistant, show you whether your firm is named, and list every source cited in your place.
Two different conversations. Pick the one that fits and we'll take you straight to the calendar.
