A founder asked me last month why his brand wasn't being recommended by ChatGPT. He'd already paid an SEO agency to "fix" it. They'd added schema markup, written an llms.txt, lodged him on a few directory sites. Nothing changed. The model still didn't know who he was.
I looked at the work. The technical stuff was fine. That wasn't the problem.
The problem was that his brand had nothing for the model to recommend. Not a positioning. Not a clear answer to "who is this for, and why them and not someone else." His website said he did "innovative solutions for growing businesses." His About page told me where the founders went to school. The model couldn't cite him because there was nothing specific to cite.
That's the branding problem most agencies aren't solving yet. And I think it's the most interesting space in marketing right now.
Two camps, both wrong
Almost every agency talking about AI visibility falls into one of two camps.
The first camp is the SEO and digital marketing shops. They've reframed AEO as a technical optimisation problem. Schema markup. llms.txt. "Get listed in trusted sources." Submit to directories. They'll audit your site, hand you a checklist, and bill you for the implementation.
That work isn't wrong. It just isn't enough. Schema markup makes a brand legible to a model. It doesn't make the brand worth recommending.
The second camp is traditional brand studios. Most of them don't think about AI visibility at all. It's not on their map. Brand work is identity, voice, positioning, applications. AI feels like a tech problem somebody else handles after the brand is built. They'll do beautiful work and never ask the question of how the model is going to read it.
Both camps are right about half the picture and missing the other half. And it's the missing half that decides whether the brand gets cited or doesn't.
What AI is actually doing
When someone asks an AI platform "who's the best brand consultant for SaaS founders in the UK," the model doesn't run a search and rank ten links. It synthesises an answer. It picks a small number of names. Sometimes one. The names it picks are the ones the model has the strongest, most specific signal about.
Specific is the key word.
If your positioning is "we help businesses grow through design," the model has nothing to do with that. There are ten thousand agencies who could be described that way. The model can't recommend you over them because there's no reason to. So it picks somebody whose positioning gives it something to point at.
The brands AI cites are the ones whose answer to "what do you do and who for" is sharp enough that the model has a slot to drop them into. That's a brand strategy problem. It's the same problem that made positioning work pre-AI, just with a higher cost for getting it wrong.
In the old world, soft positioning meant slow growth. You'd still get found via SEO, ads, referrals, networking. You could compensate with budget. In the AI world, soft positioning means invisibility. There's no second page. There's no scroll. There's the answer, and there's not being the answer.
Why the SEO framing falls short
If you treat AI visibility as a checklist of technical signals, you can do all of it and still not get cited.
I've seen brands with perfect schema, well-structured llms.txt files, citations across the trusted-20 domains, and they're still being beaten by competitors with worse technical hygiene but a tighter story. Because when the model goes to draw on what it knows about you, "well-structured" isn't useful unless what's structured is worth saying.
Technical SEO is the floor. It gets you eligible. It doesn't get you picked.
The agencies selling AEO as a pure technical service are doing the same thing SEO agencies did in 2009 when they were stuffing keywords into footers. The tactics work for a moment. Then the platforms get smarter, the noise gets filtered, and the brands that win are the ones with something real underneath the optimisation.
Why the brand studio framing falls short
The other side of this is brand studios who'd never set foot in an SEO conversation.
I sympathise. I came up through that world. The instinct to keep brand work pure, to focus on the strategic and creative craft and let the marketing technology people handle the technology, is a defensible one. I held it for years.
But it leaves a hole. If you write a positioning line, a brand story, a set of messaging pillars, and you've never asked how a language model is going to read them, you're working in a register that's quietly going out of date. The brand book has to make sense to a person reading it, a journalist quoting it, and a model citing it. Three audiences, one document. Most brand books I've seen lately are written for one out of three.
That doesn't mean writing for AI. It means writing with the awareness that AI is reading. Specific claims. Named verticals. Real differentiation. The same hallmarks of good positioning, just with a sharper consequence for vagueness.
Where the gap actually is
The studios doing AI visibility work right now aren't a third type of agency. They're brand studios who've absorbed the AI question into the way they already work.
When I take on a project, the AI visibility thinking shows up in three places. It shapes how the positioning gets pressure-tested. It shapes what goes on the site and how it's structured. And it shapes the longer-term narrative the brand needs to be telling in public so that when models look for context about the company, there's something to find.
None of that is a separate service. It's how the work gets done now. But you can only do it if you're doing brand strategy in the first place. Schema markup on a brand with no point of view is decoration. Brand strategy that ignores the model is leaving the most important new audience unaddressed.
I don't think most agencies will close this gap quickly. SEO shops can't grow brand strategy capability in a quarter. Brand studios won't add AI thinking unless somebody on the team makes it their business to. So it stays open. And the brands working with whoever closed it first will be the ones cited in their categories for the next decade.
The question isn't whether your agency does schema markup. It's whether your agency knows what the model should say about you when it gets asked. Most don't. That's the work.