Most AI search plans stop at "get our brand mentioned." The leverage is in two moves almost nobody makes: engineering which brands the model groups you with, and changing what the category gets judged on.
AI Search & GEO · 8 July 2026
Winning in AI search · Part 2 of 3
Most AI search plans stop at one idea: get our brand mentioned more. It's the right place to start. It's also where almost everyone stops, and where the actual leverage begins.
There are two moves past "be visible" that hardly anyone makes. We think they're the whole game. But before either of them works, one thing has to be true.
There's a lot of noise suggesting AI search is some separate discipline. It isn't.
Ahrefs looked at AI Overview citations and found that 38% of cited pages also rank in the top 10 organically, down sharply from 76% a year earlier. Read past the headline and the mechanic is the interesting bit: AI Overviews lean less on the single ranking page and more on whichever sources keep showing up across the fan-out of related queries. So lifting head-term rankings through content refreshes, gap work and digital PR isn't optional. It's the floor.
Every agency will pitch you that floor. Fine. It's table stakes. Here's what they won't pitch.
Models build their "which NZ bank" or "which provider" answer from how brands appear together across the web. If the open web keeps discussing four names as a set, the model learns the set. Anyone outside it gets quietly dropped at the moment of intent, however good their own website is.
So being cited once isn't the goal. Being woven into how the category gets discussed is.
That's deliberate work, not luck. It means seeding content and earning mentions that place your brand beside its peers: the same comparison articles, the same roundups, the same Reddit threads and LinkedIn posts that establish who the real players are. You're not waiting to be found. You're manufacturing the company you keep. For a home services brand fighting three local incumbents, or a beauty label up against the names everyone already lists, that placement is the difference between being a considered option and not existing.
Once you're in the set, change what the set is being judged on.
This is the move we run in brutally competitive verticals, and it's not just "be present". It's reframing the criteria. Surface a dimension of comparison the others can't match. Make the lens itself work for you.
A worked example: for a challenger bank, "NZ-owned versus Australian-owned" is exactly that kind of lens. Once you're in the AI's consideration set, the content surrounding your brand can make ownership a criterion (profits staying onshore, decisions made locally) rather than a fact buried on an about page. The model picks it up because it's stated consistently across sources it trusts. First-party angles work the same way: publish your real lending-decision timeframes, your actual fees, and you hand the AI something specific to repeat that nobody else can.
At Compare the Market, the entire car-insurance strategy depositions the idea that there's one "best" policy and replaces it with "it depends on you". That reframe is now baked into how the category gets described back to shoppers.
A quick caveat from experience: in regulated industries this is a Legal and Risk conversation, not a free-for-all. The angles have to be true and defensible. Done right, that constraint is a feature, and there's more on that in part three.
When a brand asks how to show up in ChatGPT and you treat it as an on-page problem, you start fiddling with pages. As iPullRank put it in their analysis of Google's AI search guidance, the real answer "often has very little to do with your website".
That reframes digital PR and earned mentions from a nice-to-have into a direct AI-visibility lever. A favourable mention on a respected finance page, or an honest comparison the other banks can't replicate, shapes what the AI repeats. And it's the same authority that's always driven SEO: the publications models cite are the publications that pass link equity. The two compound.
There's a structural reason to take this seriously now, not later. The US v. Google antitrust trial dragged the plumbing of search into open court, with OpenAI's head of product testifying about barriers to entry and data access, and the through-line is that distribution and source authority are being contested at the platform level. The brands building earned authority across the web now are the ones the next generation of answer engines will reach for.
Getting cited is the first win. Being the brand the AI names first, in a framing your competitors can't easily counter, is the one worth building toward. That's relevance engineering applied to your whole marketing mix, not a content tweak.
Part two of a three-part series. Next: how to win the retrieval layer and measure it under imperfect attribution. Start at part one if you missed it.