Your next client no longer starts on Google. More and more B2B decision makers open ChatGPT, Perplexity or Gemini and ask directly: "which B2B prospecting agencies are there in Spain?" or "which tool do you recommend for X?". The assistant answers with three or four names. If yours is not there, you do not exist for that buyer: there is no page two.
Public data from several analytics platforms points in the same direction: referral traffic from ChatGPT and Perplexity to corporate websites has multiplied over the past year, and firms like Gartner have long anticipated a meaningful decline in traditional search volume in favor of AI assistants. The trend is clear even if exact figures vary by industry.
The good news: earning a place in LLM answers is neither magic nor luck. It has a name (GEO), it has a method and, because it is new, almost nobody in Spain and LATAM is working on it seriously yet. This guide explains how to do it.
What is GEO (Generative Engine Optimization)?
GEO (Generative Engine Optimization) is the discipline of preparing your digital presence so that language models (ChatGPT, Perplexity, Gemini, Claude) mention, cite and recommend your company when a user asks them a question relevant to your business. If SEO optimizes for appearing in a list of links, GEO optimizes for appearing inside a written answer.
They share foundations (quality content, authority, technical health), but the goal and the metric change:
| Aspect | SEO | GEO |
|---|---|---|
| Outcome | Position in a list of links | Mention or citation inside an answer |
| Content unit | The page and the keyword | The citable passage: definition, data point, list |
| Who decides | A ranking algorithm | A model synthesizing multiple sources |
| Metric | Clicks and rankings | Brand mentions and LLM referral traffic |
Why does GEO matter in B2B?
Because the B2B buyer has changed counters. According to industry studies, most of the B2B buying process happens before talking to a salesperson, through independent research. That research, which used to be a dozen Google searches, is increasingly a conversation with an assistant that summarizes, compares and recommends.
The consequences for your pipeline are three:
- Whoever arrives, arrives convinced: a buyer to whom ChatGPT has already explained who you are and why you fit enters the conversation with half the work done. It is scarce traffic, but with extremely high intent.
- The filter is invisible: on Google you can see where you rank; with an LLM there is no official console telling you why it cited your competitor and not you. If you do not measure this, you do not know you are losing.
- The window is now: models consolidate what they read consistently across many sources. Building that footprint takes months, and whoever builds it first in their category will be the "default name" for years.
How do LLMs decide whom to cite?
There is no public algorithm, but available research and practice point to four recurring factors:
- Entity authority: models favor entities they understand well: who you are, what you do, where you operate. A consistent identity (same name, same description, same contact data) across your website, LinkedIn, directories and press weighs more than any trick.
- Structured data: JSON-LD (Organization, Service, FAQPage, Article) translates your website into machine language. It does not guarantee the citation, but it removes the ambiguity that prevents it.
- Citable content: LLMs cite clean definitions, ordered lists, sourced data and direct answers to specific questions. A paragraph that defines a concept well in two sentences is more likely to be cited than a thousand words of filler.
- Consistent third-party mentions: if only you say you are relevant, the model hesitates. If industry directories, media, podcasts and comparison pages say it too, the model takes it as fact. Mentions (even without a link) are the new link building.
Technical checklist: how to appear in ChatGPT, Perplexity and Gemini
This is what your website needs in order to be readable, citable and verifiable by a generative engine. At Desorbitante we apply this very checklist to our own website, so it is tested in production:
- Static, indexable HTML: content must live in the HTML the server returns, not be rendered only with client-side JavaScript. Many AI crawlers do not execute JS: if your text is not in the source, it does not exist for them.
- A robots.txt that allows AI bots: verify you are not blocking GPTBot, ClaudeBot, PerplexityBot or Google-Extended. There are websites investing in GEO with the door shut by an inherited disallow.
- Linked JSON-LD: Organization with sameAs pointing to your profiles, Service for each solution, Article and FAQPage on the blog. All connected with @id, not loose fragments.
- llms.txt at the root: a text file summarizing who you are, what you do and your key URLs, designed for a model to read in one pass. It is an emerging standard and takes an hour to implement.
- Clean definitions at the start of each section: answer the heading's question in the first or second sentence. Models extract passages, not whole pages.
- Real FAQs: a frequently asked questions section with the question phrased as a human would ask it and a direct 2-4 sentence answer. It is the format assistants cite most.
- Sourced data: attributed figures ("according to industry studies", with a link when one exists) instead of vague claims. Citation-based engines like Perplexity reward what is verifiable.
How do you measure GEO?
The absence of an official console does not mean it cannot be measured. Three concrete practices:
- Mention sampling: define 15-20 questions your ideal customer would ask ("best X agencies in Spain", "how to solve Y") and run them monthly through ChatGPT, Perplexity and Gemini. Record whether you appear, in what position and what is said about you. It is manual, but it is your conversational market share.
- LLM referral traffic: in your analytics, segment visits arriving from chatgpt.com, perplexity.ai, gemini.google.com and copilot.microsoft.com. They will be few compared to Google, but watch their conversion rate: it tends to be noticeably above the site average.
- Declared attribution: add "how did you hear about us?" to your forms and listen during first calls. "I found you by asking ChatGPT" is already showing up in real deals, and it is the data point that convinces leadership to keep investing.
GEO and SEO: one system, not two budgets
Treating GEO and SEO as separate projects is an organizational mistake. They share 80% of the work: a healthy technical architecture, content that answers real questions, authority built through mentions. What changes is the emphasis: GEO rewards citable clarity and entity consistency over keyword volume.
Properly understood, GEO is one more piece of your demand generation: content that educates the market and, on top of that, feeds the models that market consults. And it connects with an idea we have defended before when comparing demand and leads: trust is built before the buyer raises their hand. LLMs are now part of that "before".
The minimum operating routine: every new piece of content ships with a citable definition, an FAQ and JSON-LD; mention sampling is reviewed quarterly; and every PR or partnership action is also evaluated by its footprint in sources the models read.
Frequently asked questions
Does GEO replace SEO?
No. Google still drives the vast majority of web traffic, and generative engines themselves feed on well-ranked content. GEO and SEO share foundations; the smart move is operating them as a single system with two outputs.
How long does it take a company to appear in ChatGPT?
Engines with browsing (Perplexity, ChatGPT with search) can cite you within weeks if you publish citable, indexable content. Consolidating yourself as a recurring answer in your category is 6 to 12 months of consistent work.
What is the llms.txt file?
It is a plain text file at the root of your domain summarizing your company, your services and your main URLs in a format readable by language models. It is an emerging standard, with no guarantee of universal adoption, but its cost is minimal and it leaves your canonical information one fetch away.
Should I block or allow AI bots on my website?
If you sell B2B and want to be recommended, allowing them is the coherent choice: blocking GPTBot or PerplexityBot erases you from the answers. Blocking only makes sense for paid content or sensitive intellectual property.
How do I know if LLM traffic is already reaching my website?
Open your analytics tool and filter referral traffic by chatgpt.com, perplexity.ai and gemini.google.com. If you use GA4, create a segment with those domains; on most B2B websites it already shows up, even if small.
The question is already being asked: decide who answers it
Right now, in some office in Madrid, Mexico City or Bogota, a decision maker is asking an AI assistant about vendors like you. The answer they will receive is being written today: with your HTML, your structured data, your FAQs and what third parties say about you.
The action plan fits in one sentence: make your website readable for machines, your content citable for models and your brand consistent across the sources those models read. Start with the checklist in this guide, measure mentions every month and treat each LLM answer as what it is: the new first result.
