What we did. We ran the Bar Test —asking AI what a real client would ask— on the Madrid law firm market, in Spain, across three practice areas: tax, corporate and M&A. Findings:
- Every query returns the same three or four names, not a long list. AI does not offer options: it offers a short recommendation.
- A concentration pattern appears: the same boutique can be recommended in two different practice areas, while the vast majority of firms appear in none.
- The firms recommended are not the national giants, nor necessarily the best. They are the ones AI can recognise with confidence —through authority away from the website, through technical legibility, or through both.
The operational conclusion: in your category, AI has already started handing out recommendations, and very few firms are winning them.
The setting: a client no longer opens Google
A finance director needs a law firm for a tax inspection. The owner of a family business wants to sell it and is looking for M&A advice. Two partners in dispute need a corporate lawyer. In 2026, all three do the same thing before asking for a meeting: they ask ChatGPT, Perplexity or Google with AI.
And AI does not return ten links for the client to choose from. It returns a short answer with three or four names. If your firm is not among them, the client never finds out it existed. It is not that you lose the comparison: you never enter it.
The experiment: nine questions across three practice areas
We did what that client would do. For each of the three practice areas —tax, corporate and M&A— we asked AI to recommend boutique firms in Madrid, in the plain language of someone hiring, not in lawyers' jargon. We noted which names came back.
Two methodological decisions matter. First: we named no target firm in the question. AI recommends organically and we observe who comes up; the visibility measured is therefore real, not prompted. Second: we excluded the national giants (Garrigues, Cuatrecasas, Uría and the like). Not because they do not appear, but because the relevant question for a boutique is not «can I beat Garrigues?», but «can I beat the boutique my own size that AI already recommends?».
What AI answered
Across the three practice areas, the behaviour was identical in form: a short recommendation of three firms, different in each category apart from one revealing exception. In tax, three boutiques specialising in taxation came up. In corporate, another three, none of them overlapping with the previous ones. In M&A, three more.
So far, what you would expect. What is interesting is what happened between categories.
The pattern between categories
Of all the firms that appeared across the nine recommendations, one came up in two different practice areas: Araoz & Rueda, recommended in both tax and M&A. And here is the revealing part: we scanned its website and its structured markup for AI is minimal. Even so, AI recommends it in two categories. Why? Because its authority away from the website —decades of track record and a presence in the sector directories the engine cross-checks— is solid enough that it is recognised despite a technically mute website. We do not cite it as advertising —we have no relationship with the firm—, but as the clean example of the real pattern: AI recommends whoever it can recognise, and recognition comes through authority, through technique, or through both.
This changes how the problem reads. The easy narrative is «AI does not recommend law firms yet, it is early». False. AI already recommends, confidently and with specific names. What is happening is that the distribution is concentrated: a few legible firms take presence across multiple categories, and the majority —perfectly good firms to a human— are invisible in all of them.
For a partner, the implication is blunt and direct: you are not competing for visibility against the whole market. You are competing against the three or four firms in your category that have already done the technical work. And every week that passes, that position hardens.
Why a few do and most do not
The temptation is to assume the firms recommended are the best ones. That is not it. They are the ones AI can recognise with confidence, and that depends on two things: the firm's authority away from its own website (track record, mentions, presence in the directories the engine cross-checks) and the technical legibility of its website (Schema.org markup declaring what it is and what it does, consistent NAP, verifiable links confirming the entity is real). Araoz & Rueda wins on authority almost alone. Most firms have neither of the two in legible form —and that is why they appear in no category.
That is where your lever is. If your firm already has genuine authority —track record, cases, team— but AI does not cite you, the problem is not reputation: it is that your website does not let AI read it. Releasing that technical layer is a matter of weeks. We set it out in detail in why your firm does not appear in ChatGPT, Perplexity or Claude.
What this means for your firm
Three practical readings of this Bar Test:
- The channel is already handing out places, and there are few. Do not wait for it to «mature». By the time it is obvious to everyone, the places in your category will be taken.
- Your competitor is not the giant. It is the boutique your own size that AI already recognises —through authority, through technique, or through both. That is the position you can contest.
- It is within reach. Because the problem is infrastructure and not reputation, it is corrected at code level, without redesigning the website. If your firm already has genuine authority —track record, cases, team—, making it legible to AI is a matter of weeks.
Check it yourself in three minutes
You do not have to take this article on trust. Open Perplexity or ChatGPT and type the exact query your ideal client would type: «best [your practice area] firm in [your city] for [the type of matter that pays you best]». Read the names that come back. If yours is not there, you have just seen what your next client sees.
Frequently asked questions
How do I know whether AI recommends my firm?
Open ChatGPT, Perplexity or Google with AI and type the query a client of your profile would type: for example, «best [your practice area] firm in [your city]». If your name does not appear among the firms recommended, your firm is invisible on that channel. This is what we call the Bar Test.
Why does AI recommend some firms and not others?
It is not a question of legal quality or prestige. It is a question of data infrastructure: generative engines only recommend entities they can read and reconcile with confidence through structured markup (Schema.org), NAP consistency and verifiable links. An excellent firm with poor infrastructure is invisible; a competent one with solid infrastructure takes the answer.
Is this the same as SEO?
No. SEO competes for position in a list of links on Google. GEO (Generative Engine Optimization) competes to be cited inside the single answer AI gives. A firm with impeccable SEO can be completely invisible in ChatGPT or Perplexity, because these are different systems with different criteria.
Can my firm's situation be changed?
Yes. Because the cause is infrastructure and not reputation, it is corrected at code level, without redesigning the website or rewriting the content. In a firm with genuine prior authority, the result is usually visible in weeks, not years.
Update (July 2026): we ran the test on ourselves
We do not ask for anything we have not tested on ourselves. We applied this same Bar Test to our own website over 60 days and documented it without touching up the results. The outcome, in two numbers: we went from zero citations in AI answers to more than 300, and from no engine recommending us to Gemini now including us when someone asks for a consultancy for law firms — still only one engine in five. The lesson is confirmed on ourselves: the technical part is won in weeks; getting AI to recommend you to a stranger is authority, and that takes months. The full experiment, here.
Would you like to know which name appears instead of yours?
The GEO audit includes a bespoke Bar Test of twenty queries across the five generative engines, a dossier with the 30/60/90-day implementation plan and a presentation video. We show you exactly who is winning the recommendation from you, and why. From €1,500.
Request a diagnostic