Strategy · GEO
The problem isn’t that AI recommends tools that don’t exist. It’s that it recommends real tools in the wrong context. In the experiment this article documents — 3 prompts sent to 4 AI platforms in clean sessions during the week of 24 March 2026, totalling 12 independent experiments — three actors occupied leading positions in responses about legaltech for law firms in Spain. All three are real, active products. None of the three is what the AI said it was when it placed them in first position.
That is GEO in reverse: the mechanism that should surface the right actors works equally well at miscontextualising the wrong ones. GEO — Generative Engine Optimisation — is the set of content and digital architecture decisions that determine whether a company appears in AI platform responses, how it appears, and with what argument. Unlike traditional SEO, the criterion is not ranking position in a results list: it is direct citation in the model’s answer.
Sector context — Spanish legaltech market. If you’re already familiar with the ecosystem, you can skip this section.
The market for AI-powered legal tools for law firms in Spain organises into four groups. The editorial incumbents — Aranzadi LA LEY (Thomson Reuters), Wolters Kluwer and Lefebvre — dominate by document volume and history, having integrated generative AI on top of their own databases. The second group covers CLM and contract automation: Bigle Legal is the most relevant Spanish player, alongside international tools such as Luminance, Harvey and Spellbook at the high end. The third group are emerging AI-native legal platforms: Prudencia.ai, Maite.ai and Lexiel target the small firm or individual lawyer who needs a legal co-pilot in Spanish. The fourth group — large firms with proprietary tools such as Garrigues (Gaia), Cuatrecasas (CELIA) or Pérez-Llorca (Leya) — falls outside the scope of this experiment.
The segment analysed is firms of 5 to 50 lawyers, primarily in commercial, employment and corporate practice — the highest-volume segment in Spain, and the one most likely to consult AI platforms before adopting a new tool.
The mention map: who appears on each platform and how often
The table below covers all actors identified across the 12 experiments.
✓L Explicit leadership position
✓ Mentioned
✓L* Leadership in wrong context
— Not present
| Actor | ChatGPT | Perplexity | Gemini Free | Gemini Thinking | Total |
|---|---|---|---|---|---|
| vLex / Vincent AI | ✓ | ✓ | ✓ | ✓ | 8/12 |
| Bigle Legal | ✓ | ✓ | ✓L | ✓L | 7/12 |
| Aranzadi / LA LEY | ✓ | ✓ | ✓ | ✓ | 6/12 |
| Wolters Kluwer | ✓ | ✓ | ✓ | ✓ | 5/12 |
| Prudencia.ai | — | — | ✓ | ✓ | 4/12 |
| Maite.ai | — | — | ✓ | ✓ | 3/12 |
| LexDoka | ✓L* | — | — | — | 2/12 |
| Lexiel | ✓ | — | — | — | 2/12 |
| Parallel | — | — | ✓L* | — | 1/12 |
| MiDespacho.Cloud | — | ✓L* | — | — | 1/12 |
3 prompts × 4 platforms = 12 independent experiments. Clean sessions, week of 24 March 2026. ✓L* = leadership position in wrong context — detail in the section below.
The most striking data point before examining the errors: vLex/Vincent AI leads the map with 8 of 12 possible mentions. It is a legal database platform in the process of integrating AI — its client profile and pricing model differ substantially from those of Bigle Legal, Prudencia.ai or Maite.ai — but AI platforms cite it consistently across all four tools tested. The three actors with verified products for the target segment never appear with the same force on the same platforms: Bigle Legal features across all four but rarely as an explicit leader; Prudencia.ai and Maite.ai are absent from ChatGPT and Perplexity — the two platforms with the highest volume of professional queries.
Three platforms, three different mechanisms for miscontextualising a real actor
ChatGPT: the article’s argument, without verifying the product
In Prompt 2 of the experiment — “What are the best legaltech solutions for law firms in Spain in 2026?” — ChatGPT placed LexDoka in first position in the CLM segment.
LexDoka is a real product. Its core is a multi-agent AI for audit work — with specialised agents covering standards such as NIA-ES, PGC and ICAC — and a secondary product, LexDoka Legal, with published pricing (€80–200/month) and 500+ legal teams declared on its website. However, its brand identity, its LinkedIn presence (tagline: “AI for auditors”, 2–10 employees, 609 followers) and all of its commercial communications point to audit firms, not law firms.
The source that generated ChatGPT’s position was an article published in July 2025 in Revista Jurídica DERECHO ABOGADOS, titled “LexDoka redefines contract management with its all-in-one intelligent software”. ChatGPT used that argument. It did not verify what the product actually was, or which client profile it was designed for.
Gemini: the company’s own blog as a category trap
In the same Prompt 2, Gemini Free placed Parallel in first position in the commercial contract CLM segment.
Parallel has a real and verified presence in the Spanish legal sector. Cuatrecasas — one of Europe’s largest law firms by revenue, with over 1,700 professionals across 27 offices — uses its platform to manage the client onboarding process (KYC). The results are documented in their published case study (document in Spanish): 40% reduction in KYC process follow-up time, 25% faster matter opening, 3 hours saved per professional per week, and over 1,000 client onboardings completed without email chains. Parallel’s CEO is a former tax lawyer from Cuatrecasas itself.
But Parallel does not sell commercial contract management. It sells document process automation for KYC, client onboarding and AML (anti-money laundering). The reason Gemini placed it in CLM is that Parallel’s blog published an article titled “LegalTech: 8 types of solution that can help you”, which described CLM as one of eight legaltech market categories.
Gemini used that article to associate Parallel with the category. It did not distinguish between “company that writes about CLM” and “company that sells CLM”.
A second mechanism in Gemini is worth noting: LexGoApp, a legal tool comparison site for Spanish lawyers with four or five well-structured articles, appears as a source in at least 4 of the 12 experiments. In practice, a single domain is determining which local actors Gemini recommends when a lawyer searches for tools.
Perplexity: the page title, without verifying the geographic market
In Prompt 2, Perplexity placed MiDespacho.Cloud in first position.
MiDespacho.Cloud is a real, active product with verified clients. It offers AI-powered firm management, case tracking, a client portal and marketing tools. The problem is that it operates exclusively in the Latin American market: its pricing is in US dollars ($19.95/month) and Mexican pesos, and its testimonials correspond to users based in Mexico.
The mechanism was the most mechanical of the three: the actor’s own website — with a page title optimised for “the best technology solution for legal firms in 2026” — was indexed among the 10 sources Perplexity used. The content was in Spanish. The legal terminology was generic. That was sufficient for first position. A law firm in Madrid following that recommendation would be contacting a provider that does not operate in their market.
What the three cases have in common: AI doesn’t fabricate, it miscontextualises
The three errors are different in type — of use (LexDoka), of segment (Parallel), of geography (MiDespacho.Cloud) — but they share the same underlying mechanism: content well-placed in a sufficiently authoritative source moves positions without the model verifying whether the context is correct.
This distinction matters methodologically. The standard criticism of AI in this type of analysis is that it “invents companies”. What this experiment documents is different and harder to detect: AI positions real companies in incorrect categories because existing content — a press article, a blog post, a page title — associates them with that category, and the model does not cross-check that association against the actual product.
The inverse implication is direct: any company with content in Spanish about legaltech can appear in responses aimed at the Spanish law firm market, regardless of its actual product, its segment, or its geography. The filter is not the product. It is the content.
The data point that captures the gap
The three actors with verified products for the target segment — Bigle Legal, Prudencia.ai and Maite.ai — do not appear with the same force on any of the three platforms for any of the three prompts. Prudencia.ai, which has the most differentiated product argument in the experiment and published pricing from €69/month, is invisible on ChatGPT and Perplexity. Bigle Legal, present across all four platforms, receives the explicit leader label in a single experiment out of twelve. The fragmentation is not random. It corresponds to something concrete and actionable — but it is not in the product.
If miscontextualisation happens without anyone planning it, what can content built with precision do for the actors that do have the right product?