GEO in 2026: what your website needs to get the AI recommendation

On July 9, OpenAI released its new GPT-5.6 model family (Sol, Terra and Luna), with stronger agentic capabilities to research, compare and act on a user’s behalf, as reported by TechCrunch and Axios. More and more of your prospects — including your own — aren’t searching Google anymore: they’re asking an AI assistant to research and recommend. If your website isn’t built for that assistant to read it, understand it, and cite it, you simply don’t exist on that channel. That’s what GEO (Generative Engine Optimization) answers: the discipline that decides whether you show up in the answer or not.

Why this stopped being a theoretical debate

Recent launches confirm a trend that’s been building for months: general-purpose AI models are no longer limited to answering questions — they browse, compare sources and act on behalf of whoever’s asking. When a prospect asks ChatGPT or Gemini “recommend an accounting firm in Barcelona” or “which custom software agency works well in my industry,” the model doesn’t make up an answer at random: it browses, cross-checks a handful of sources, and cites two or three. If your site isn’t among them, you lose the conversation before it starts.

This doesn’t replace Google or classic SEO — people will keep searching and clicking — but it adds a new channel with partly different rules, and most small and mid-sized businesses haven’t touched it yet.

GEO isn’t magic: it’s giving the model what it needs to cite you

A generative engine prioritizes content it can cite without taking a risk: verifiable claims, backed by concrete data and self-contained context, on pages that load fast and are unambiguous to parse. In practice, that means a handful of concrete things:

Self-contained, citable content

Every relevant page — your service, your industry, your company profile — should answer “what does this company do, and for whom?” in the first paragraph, without the model having to infer it from the rest of the site. Sentences with concrete entities (name, city, industry, a real number) get cited far more often than generic filler like “an innovative solution for your business.”

Structured data (Schema.org)

Organization, Service, FAQPage or Product markup isn’t a technical nicety — it’s the most direct way a model has to confirm what you are, what you offer, and at what price, without having to interpret it from visible text.

FAQs written the way people ask an AI

An FAQ section that answers the way someone would actually ask ChatGPT — “how much does…?”, “how does… work?”, “is it safe to…?” — with self-contained answers that name your company, gets cited far more often than a generic three-line FAQ.

llms.txt and clean technical architecture

Just as robots.txt guides crawlers, an llms.txt file at your site’s root summarizes your business in a format language models can parse. Add solid Core Web Vitals, correct hreflang if you operate in multiple languages, and clean sitemaps: whatever makes a Google crawler’s job easier makes an LLM’s job easier too.

The case of professional services

In law firms, accounting practices, consultancies and management firms we see the same pattern over and over: the website communicates well to a human who already understands the industry, but assumes too much context for a model to understand it in isolation. A tax advisory firm whose site says “we manage your accounting with rigor” gives an AI assistant nothing citable; one that says “tax advisory in Barcelona specializing in businesses under 20 employees, fixed quote from the first meeting” does. You can see the full approach on our AI for professional services page: the same precision we apply to RAG assistants built on a firm’s own documentation applies, point for point, to how your website reads from the outside.

How we apply this at Evicron

This isn’t theory we leave in a slide deck: this very website ships with llms.txt, structured data and citable content by design. On Plata Garage’s site we publish 9 different types of structured data; on Forcadell’s trilingual real-estate portal, with over 2,300 listings, we keep hreflang correct across Spanish, Catalan and English so both Google and generative search engines serve the right version. All of this is part of our SEO and AI-optimized websites service: measurable technical SEO plus the GEO layer most studios still aren’t applying, starting at €3,500 with delivery in 4-6 weeks.

What to do this week

You don’t need to rebuild your entire site to get started:

1. Audit your opening paragraph on every key page: does it answer what you do, for whom, and where you’re based, without depending on the rest of the site?
2. Check whether you have minimal structured data (Organization, at least) and add it if you don’t.
3. Rewrite your FAQ as real questions, not marketing headlines.
4. Publish a basic llms.txt: five lines describing your business is enough to start.

If you’d like us to audit how an AI assistant sees your business today and what’s missing for it to cite you, get in touch: the first consultation is free, and you’ll walk away with a concrete diagnosis, not a generic checklist.