Published by Sue Johns-Chapman
Published Date 26.05.2026 
Generative visibility isn’t won on your website. It’s won in the citation graph. And most brands are still optimizing in just one territory.
The short list has moved
For two decades, marketers fought for a single prize: a spot on the consumer’s mental short list. Advertising bought reach in the channels target audiences trusted, hoping repeated exposure would convert into recall when a purchase decision came up.
That short list has moved. When a B2B buyer asks ChatGPT, “What’s the best CRM for a 200-person sales team?” or a homeowner asks Perplexity, “Which smart lighting system works best with Apple Home?”, the AI returns a curated short list of three to five brands. The user didn’t build it. We did, collectively, through every page the model has scraped, every article a brand has earned, every review a brand has ignored.
Many brands have not yet realized that they are no longer competing only to be one of the obvious choices in the minds of potential customers, but also to be one of the options recommended by AI models.
GEO is fought beyond your website, across the entire ecosystem
Traditional SEO worked on a closed surface: your domain. Technical hygiene, on-page semantics, internal linking and backlink authority. Everything pointed at one URL space you owned and controlled.
Generative visibility doesn’t work that way. Muck Rack’s Q4 2025 study found that 89% of all AI citations come from earned media, not from brand-owned content. SparkToro’s 2026 analysis confirms that roughly 85% of brand mentions in AI answers originate on third-party pages. Ahrefs recently tested schema markup and found it did not move AI citations at all.
Translation: if your entire AI search strategy lives on your website, you are optimizing about 15% of the signal surface. That is an uncomfortable number for organizations that have spent a decade building SEO machinery aimed at their own domain. It is also a strategic opening for those willing to look up from it.
Semantic consensus: what LLMs actually score
LLMs don’t rely on a single source, especially not your own website. Instead, they look for agreement across multiple sources to identify consensus. When Forbes, TechCrunch, a respected trade publication, and an analyst report describe a brand with consistent attributes, the model reads that convergence as authoritative consensus.
The Stacker/Scrunch research is clarifying: a single Tier-1 placement yields a citation rate of roughly 7.7% in relevant prompts. Distribute the same story across four or five authoritative publications and the citation rate climbs to 34%. Same brand, same news, four times the visibility, because AI systems treat data replicated across multiple sources as confirmation, not redundancy.
BrightEdge’s AI Catalyst analysis adds another layer. Across five AI engines, source overlap between any two engines ranges from 36% to 59%, while brand overlap ranges from 35% to 55%. Engines disagree about what to cite. They agree about who to recommend. That alignment is the fingerprint of semantic consensus.
Map the entities. Then occupy the sources.
This is the part of the job that traditional marketing teams still tend to overlook.
Start by defining the entities and attributes you want LLMs to associate with your brand: categories, use cases, adjacent technologies, and differentiators. Then, audit the citation graph of your category. Amsive’s research found ChatGPT leans heavily on Wikipedia, Perplexity on Reddit and YouTube, Microsoft Copilot on Forbes and Gartner. The mix changes by vertical: in beauty, social and creator sources dominate; in B2B SaaS, analyst reports and trade press carry more weight; in connected hardware, manufacturer ecosystems, specialist review media and community forums do most of the corroboration work.
Once you know which sources the model is reading for your category, the strategy writes itself: be present in those sources, described consistently, in the language of the entities and attributes you have chosen.
We applied exactly this approach with Signify / Philips Hue, our finalist case for Best Use of Search in Retail/Ecommerce at this year’s European Search Awards. Rather than treating AI visibility as an on-site optimization problem, we mapped the smart lighting category’s citation graph across AI engines, identified the entity associations that mattered most to commercial intent — ecosystem compatibility, design authority, energy efficiency — and aligned earned and owned media to reinforce those associations at the source layer the models were actually reading.
PR’s overdue promotion
Corporate communications has spent a century in the service of institutional goals: reputation, executive visibility, crisis response, and investor narrative. PR was almost never asked to drive customer acquisition because that was marketing’s job, measured in clicks and conversions.
This division of roles no longer works.
AI engines turn earned media into a direct input for purchase recommendations. Stacker’s research found that earned media distribution lifts AI citations by a median of 239%. This is no longer merely a matter of brand recognition but is clearly linked to the company’s actual ability to attract customers, which is attributable to the work of the communications teams.
This means that at least part of every serious PR and digital PR programme should now be allocated to organic findability. Of course, not at the expense of institutional priorities, but alongside them.
But… isn’t this black hat GEO?
If you’re thinking of LLM poisoning, think again. This is not about tricking the models.
LLM poisoning means deceiving a model with manipulated or false information to alter its outputs. What we are trying here is the opposite: ensuring the sources LLMs already trust describe your brand in a way that is consistent with the positioning you have genuinely chosen. It is the old discipline of brand coherence applied to a new distribution layer.
Rand Fishkin’s January 2026 SparkToro study confirmed that individual AI responses are randomised: identical prompts return different lists across runs. But visibility percentage across many runs is stable, measurable and shaped by the citation graph. We do not try to trick the model. We try to influence the consensus it is summarising.
A playbook for European marketing leaders
Five steps you can take today to build your AI brand authority tomorrow:
- Map your target entity-attribute associations. What should an LLM say about you, and next to which entities?
- Audit your category’s citation graph. Identify the 10–15 sources AI cites most in your commercial prompts. That is your real battleground.
- Concentrate on Tier-1 plus vertical press. Two placements in authoritative outlets outperform twenty in low-authority ones. Consistent category citation typically arrives at 3 to 5 well-distributed Tier-1 hits.
- Maintain a freshness cadence. Content updated within the last 30 days earns notably more AI citations.
- Measure visibility percentage, not rankings. Run each prompt multiples times across platforms. Track presence, sentiment and accuracy, not position.
The short list of the future is being written now
Entity authority is slow to build and slow to decay. The brands that align their PR with their findability strategy now will be the brands AI recommends later.
The short list has moved from the consumer’s mind to the model. The only question worth asking now is whether your communications strategy has moved with it.
Author: Fernando Maciá Domene. Bio: Fernando Maciá is the founder and CEO of Human Level and one of the pioneers of SEO, online marketing and web analytics in Spain, working in the field since 2001. He is the author of twelve books on SEO and digital marketing, including the first Spanish-language book on SEO (2006). He has lectured at over 450 seminars and courses across Spain and Latin America, and speaks regularly at international search and marketing conferences.
