The AI search industry wasn't built for B2B buying

In November 2023, six researchers published a paper that gave a new discipline its name: Generative Engine Optimization (GEO). They set it up as a way of optimising, and defining, visibility metrics for AI-generated answers. A whole industry has since been built on that definition.

I wanted to see how that industry is made up, and how well it's geared to support the way B2B buyers now buy. What surprised me most was how little of it is built with B2B in mind.

As at 3 October, I'd mapped 173 firms and tools selling AI visibility and discoverability, found through AI answers, directories, agency listicles and a database sweep. It isn't a census - plenty more agencies have GEO on a services page - but it's a fair picture of who a buyer would come across. And because a sweep built on visibility language will mostly find visibility, I also ran a second search, this time for firms selling AI-mediated buying, vendor selection or getting shortlisted.

Here's how the 173 break down:

What stood out about the 145 service firms was that about two-thirds are agencies that sold SEO, PR, content or performance marketing first, and have added AI as a bolt-on or a repositioning. So are all but eight of the 35 who work in B2B.

The rest of this issue looks at those 35 more closely.

Visibility is the gate, not the destination. Being named only gets you onto the list. To make the shortlist, the AI has to frame you as a fit for the buyer's problem and prefer you over the alternatives - and the buyer still chooses after that. Most of the 35 measure only the gate: 25 of them frame their work on visibility alone.

Their reports count citations, mentions and share of voice - the metrics of search, carried straight over into AI. The lens came with the metric, back in that 2023 paper, and most of the firms I mapped still look through it. The other ten have started talking about shortlists and buyer journeys, but look at what they actually sell and it's usually a citation audit.

The next metric along is recommendation rate: how often the answer prefers you, not just names you. A few of the 35 now frame their work around it, and a handful of firms outside my list track it. But it's embryonic, and no two I found define it the same way.

The AI-native specialists - firms with no SEO history to unlearn - have picked up the new language fastest: six of the eight are among those ten. One goes further than anyone. It maps the buyer's whole AI conversation, stage by stage, and shows where they drop out. But it still stops where the conversation ends.

This is where B2B and consumer AI search part company. A consumer can see an answer and buy. A B2B purchase goes through a committee and weeks of checking, so being recommended gets a vendor into contention, but the vendor can still lose the deal on what the buyer finds next.

Here's where the problem compounds. Recommendation is barely measured; what happens after it isn't measured at all.

B2B buyers will check the answers that come from AI - though rarely by clicking the citation: Pew Research tracked US Google users and found they clicked a link inside an AI summary in just 1% of visits. But they do go straight to your site, read third-party reviews, look for case studies, speak to peers who've bought from you, or talk to your sales team. If what they find doesn't back up what the model said, the recommendation dies there - and no visibility report will show you that.

None of the 35 specialists working with B2B clients follows the buyer to that point. Nor did a targeted search I ran on 2 October turn up any firm that does.

Most of the data needed to close this gap already exists. The tools agencies use already track what the AI says about a brand and which sources it draws on. Vendors hold the rest: their own site, their review profiles, their sales conversations. What's missing is anyone putting the two side by side.

If you already have an agency, use this to question what you're getting. If you're just about to choose one, put it in the brief - and be wary of "best agency" lists, because several of the ones I found were written or sponsored by agencies that then appear in them.

Ask three questions:

What does the report count? Mentions, citations and share of voice tell you whether you're visible. Recommendation rate, however it's defined, tells you whether the answer prefers you. Both stop at the answer and don't look at the downstream decision process that follows.

What happens when the buyer acts on it? Does anyone check what the buyer finds next - your site, the reviews, the case studies, the sales conversation - and whether it backs up what the model just said?

What are you judged on? If success is citations gained, the work ends where the buyer's decision starts.

The second question is the one that sorts them. If the answer is no, you're buying visibility, whatever their homepage says.

If you want a quick read on where you stand, reply with your company and category. I'll run three of the prompts your buyers would use and tell you whether you show up and how you're described. First five replies, back to you within a week.

Next issue, I'll follow the buyer one step further: to the publishers, review sites and communities that feed the AI's answer and get checked after it - and what that's worth to the businesses that own them.

John
Decoding Growth (formerly Get2Growth)

Method: 173 firms and tools mapped as at 3 October 2026. 33 of the 35 specialists were read on their live sites, the other two from directory profiles. Firms are described by pattern, not named.