Google's AI Overviews now appear on 23.6% of legal queries, and on 57.9% of the question-style searches a worried client actually types, on an Ahrefs analysis of 146 million results pages. Where an Overview appears, click-through to the results beneath it falls by close to 60%, and 68% of Google searches now end without a click to any website at all.

Every tool in the conventional kit assumes that click. The practice area page exists so a visitor lands on it, the client alert so somebody reads it and writes in. Take away the visit and you take away the mechanism, while the dashboards carry on looking respectable, because no firm can measure the enquiries it was never mentioned to.

What replaces the click is brief and rather final. A client with a problem asks ChatGPT, or Gemini, or Perplexity, and gets back a paragraph that answers her question and names two or three firms. She writes to one of them that week. She never sees a list of ten, and nor do the firms left off it.

Generative engine optimisation, or GEO, is the work of understanding how a machine arrives at those two or three names.

Why a machine cannot use your website to choose a lawyer

A system asked to name a competent M&A lawyer in Chicago has your website and four hundred others. Yours says you are a full service firm with deep expertise and a client centric approach. So do the other four hundred, in roughly those words.

The claim is true enough. It does not separate you from anybody, and a system that must produce three names needs a source that separates firms rather than flattering them equally.

Very few sources do. Somebody independent has to have made a judgement, applied it consistently, and published the result where a machine can read it. In practice that means the ranking tables, and the concentration is severe. The 2026 Legal AI Visibility Report from 5W and the Haute Lawyer Network found seven properties holding effectively the entire citation layer for legal queries: Chambers, The Legal 500, Super Lawyers, Best Lawyers, Martindale, Avvo and Justia.

The list splits in a way that decides where your effort goes. Chambers and The Legal 500 rank firms and practice groups, run on submissions and researcher interviews, and govern the corporate and disputes end of the market. Avvo, Justia, Super Lawyers, Martindale and Best Lawyers work at the level of the individual lawyer, draw on bar records, peer review and client ratings, and govern the consumer end. A plaintiffs' firm in Phoenix and an Am Law 50 firm in New York are not competing for the same tables, and work aimed at the wrong half of the list is wasted.

Size does not buy an exemption. The same report tested the largest firms in the country and found that Cravath, Kirkland and Skadden cannot outrank the directories for their own practice areas. Search for Cravath's M&A work and the top six results were two Chambers profiles, two of Cravath's own practice pages, a Legal 500 profile and a Chambers Global profile. The firm is present, and it is present as an entry in somebody else's table.

Being ranked is not quite the end of it either. A machine can use only what it can open, and a profile behind a login is invisible whatever the masthead says.

One question sets your strategy, and the answer differs by practice area

Before committing a dollar, establish one fact for each of your main practice areas. Has anybody independent already published a ranking for this kind of work, in this place?

Where they have, the engine defers to it. Ask about securities litigation in New York, or personal injury in Phoenix, and the existing tables effectively write the shortlist. A firm that is not ranked will not be named, whatever its website does.

For genuinely niche work, nobody has. The engine then has nothing to defer to, and assembles its answer from ordinary search results instead. Ask about a narrow ERISA fiduciary question, or CFIUS review of a semiconductor joint venture, and careful writing on an open website still wins cleanly.

Strategy therefore gets decided practice area by practice area, and most firms sit on both sides of the line without having checked which is which.

The arithmetic of waiting

Citation authority behaves like any other scarce position. It gets more expensive as more people want it, and the same report puts the increase at roughly 50 to 80% compounded annually over the next two years.

Run that forward and it stops being an abstraction. A firm that starts now at $500,000 and establishes authority in its category by 2028 pays around $1 million in total. A firm that begins in 2028 needs $3 to $5 million for the same position, and if the category has filled up by then, may not be able to buy it at any price.

Those are projections rather than invoices, and I would treat the precise figures with the caution any forecast deserves. The direction is the part worth acting on. Positions in a small number of tables are being taken, they are not being vacated often, and the cost of arriving late compounds.

Where to start this week

Ask ChatGPT, Gemini and Perplexity the five questions your best clients actually arrive with, and write down which firms each one names. It takes an afternoon, and little else tells you your starting position.

Then check whether an independent ranking exists for each practice area, because that answer tells you whether your effort belongs in submissions or in writing. I have mapped that out directory by directory in a longer guide to GEO for law firms.

Start with the tables that already carry a page about you. Avvo builds profiles from state bar records whether a lawyer engages with it or not, so a great many firms have entries they have never seen, holding old addresses and missing half the practice areas. Claiming and correcting those costs nothing, and it repairs the most widely read version of you.

Where a ranking does exist, the practical question is which research cycle you can still make. Chambers and The Legal 500 both run annual cycles with deadlines set jurisdiction by jurisdiction, and a firm that misses one waits a year for the next.

Then make your own pages state facts rather than adjectives. Bar admissions, courts, jurisdictions, matter types. Model Rule 7.1 requires that what you say about yourself be neither false nor misleading, which is a low bar for a firm writing down what it verifiably does, and an awkward one for a firm writing down how excellent it is.

None of this is glamorous, and it is closer to filing than to advertising. Bing Webmaster Tools and Google Search Console have both begun reporting how often a site is cited or shown in AI answers, so for the first time the effect is becoming measurable rather than inferred.