AI Visibility in B2B: there is no universal playbook

For a year and a half you have heard the same recipe for winning AI visibility in B2B: raise your Domain Authority, fill the site with content, optimize the domain, and ChatGPT will eventually cite you. Almost every GEO agency sells it. The problem is not that it falls short. It is that it measures the wrong lever. Ahrefs analyzed 75,000 brands and measured what correlates with showing up in Google's AI answers (its AI Overviews): by far the strongest signal is brand mentions spread across the web -a correlation of 0.664-, well above backlinks, which sit at 0.218. Your Domain Rating lands in between, at 0.326: it matters, but only half as much as mentions.

AI visibility in B2B: AI recognizes entities, not domain strength

A model does not crawl your link graph when it decides who to cite. It does not count how many links point to your domain. It looks at whether your brand shows up, consistently, on sites it treats as trustworthy. That is why a decade of link-building weighs less than being mentioned in the right places: the model measures recognition, not references.

The backdrop confirms it. Muck Rack analyzed more than 25 million links cited by ChatGPT, Claude and Gemini, and 84% of the citations come from earned media -content with editorial or third-party backing, not the material you control with direct commercial intent. That share has moved between 82% and 89% across its three editions since July 2025: it is not the whim of a model update, it is how these systems choose where to pull what they say from. Your product page, your landing, the post written to sell: none of that is what the model lifts as a source.

So far, the easy conclusion would be "drop the website and go chase mentions." That would be just another single recipe. And single recipes are exactly what this is here to dismantle.

Does AI cite a brand the same way across every model?

No. And that is the piece almost no one puts on the table. Semrush analyzed 325,000 prompts and 89,000 LinkedIn URLs cited in AI search engines, and the pattern is not one: LinkedIn is cited unevenly depending on the engine -14.3% in ChatGPT Search, 13.5% in Google AI Mode and just 5.3% in Perplexity-. What makes you visible in one may move nothing in another.

That is where "do this and AI will cite you" falls apart, whatever the "this" is. There is no single answer to where to invest in AI visibility in B2B that works for every company, because it depends on the engine your buyer asks in. A consultancy whose clients ask in ChatGPT and a SaaS whose technical buyers live in Perplexity should not run the same strategy, no matter what the trending headline says. Same sector, different allocation. Anyone selling you the universal playbook is selling you smoke.

What it costs your pipeline to be left off the day-one shortlist

This is where a CMO should start doing the math, because this is not measured in reach, it is measured in pipeline. 94% of B2B buyers use AI models during the buying process (6sense). And the figure that turns it into money: 95% end up buying from one of the vendors that were already on their shortlist on day one (6sense).

Put the two numbers together. If the initial list is increasingly formed by asking an AI, and you do not appear on it because you optimized the channel that model does not cite, you are not fighting the deal at a disadvantage. You are out of 95% of the business before the conversation even starts. And the worst part is that you never see it coming: that pipeline does not enter your CRM as a lost opportunity, because it never entered at all.

Before you move a single euro of budget, there is one question to answer with data and not with intuition: which of your entities -your domain, your brand, your people's profiles- each model cites when your category comes up. The Antropus visibility report measures it engine by engine, so you know where you show up and where you do not before deciding where to spend.

The Entity-Model Matrix: your B2B AI visibility budget, allocated with data

Out of all this comes a model, and it fits in three lines, all three of them numbers.

The first is which of your entities AI cites when someone asks about your category: it is not the same for it to name you, your domain, or your competitor. The second is that breakdown split out by model, because ChatGPT and Perplexity do not agree and an average lies to you while wearing the face of a data point. The third, the one almost no one calculates, is the real weight of each model among your buyers -not among the general public- because your budget follows whoever buys from you, not OpenAI's market share.

At Elevam we use the HSA Protocol -developed by Asier López Ruiz- to measure Share of Response by entity and by model before deciding where to invest: a systematic sweep of prompts against the four leading models, repeated over multiple iterations so the result does not hinge on a single stray answer. It is what turns "we think we should post more on LinkedIn" into "in the model where your clients buy, your brand ranks below your competitor, and that is why the budget goes here." Without that measurement, the Entity-Model Matrix is a nice idea. With it, it is a decision you can defend in front of a committee without breaking a sweat.

And there is the uncomfortable part. A brand with real authority outside the models may not yet exist inside them. You can be a benchmark in your sector and still have ChatGPT skip your name when a buyer asks who does it well. That is exactly the problem: visibility in generative AI is not inherited from your outside reputation. It is built -and, before it is built, it is measured.

Frequently asked questions