Discovery and consideration are moving into AI answers - but the highest-value commercial terms still don't trigger them. Here's where the fight actually is, and why "write better content" misreads the shift.
AI Search & GEO · 29 June 2026
Winning in AI search · Part 1 of 3
Over the next year or two, search stops being one thing.
Blue links still matter at the bottom of the funnel. Someone who types "apply for home loan" is ready to act, and Google will keep handing them a page. But the part of search where people actually decide who to trust is moving into AI Overviews and answer engines. Discovery, consideration, the early reading-around: all of it. At I/O 2026, Google called it a new era for AI Search and shipped the biggest change to its search box in 25 years. That's not a forecast. It already happened.
Most banks, insurers and retailers will treat this as a content-quality project. Write a bit better, refresh a few pages, carry on. We think that's a mistake, and the reasons are specific enough to be worth laying out.
Here's the thing nobody selling AI panic wants to say: the highest-value commercial terms don't have AI Overviews on them.
"Credit cards." "Business banking." "Savings accounts." "Apply home loan." No AI Overview present. Those are the queries where advertisers spend tens of millions and Google earns hundreds of millions, and Google has every reason to protect that real estate. So the AI answers appear one rung up the funnel instead, on "home loan interest rates", "best savings accounts in NZ", the comparison terms people use while they're still working out who's worth talking to.
That distinction matters because it tells you where to fight. Lose the comparison layer and you never make the shortlist. Win it and you're in the room when the commercial query finally gets typed. Which is still, for now, a paid search and classic SEO game.
One more shift worth naming: the commercial SERP changed in late 2025 too, and not only because of AI. Google made ads look more organic. We watched rankings hold steady while the numbers underneath them moved, purely because the page around the ranking had been redrawn. If your reporting only tracks position, you'll miss it.
A single question doesn't stay single anymore. Google fans it out into a cluster of related searches and assembles an answer from across the lot. So the unit you're optimising for isn't a keyword. It's the whole web of questions around a topic.
In practice that means building for how people actually phrase things: "best", "cheapest", "versus". "Best credit cards." "Cheapest home loan rates." If your pages only answer the polished head term and ignore the messy real ones, you're invisible across most of the fan-out. This is as true for an ecommerce catalogue as it is for a lead-gen funnel.
You'll hear that AI isn't worth the attention because it drives roughly 1% of traffic. That reading is wrong in two ways.
Overall traffic is down. That part's real. But AI shows up as almost no referral traffic for a simple reason: people get the answer inside the AI, then go direct or search your brand on Google. The visit that AI caused gets logged as direct or branded. You genuinely cannot measure how much demand it's shaping, and that blind spot is the actual risk, not the headline percentage.
When someone asks an AI "what's the best home loan rate in NZ", the brand it names gains consideration before anyone visits a website. The brand it leaves out is gone from the moment of intent and never sees the data to prove it. We'll get into how to instrument that in part three. For now the point is just: the gap is the threat.
AI answers are only as good as what they retrieve. The head of ChatGPT, Nick Turley, made this plain on The Verge's Decoder: the right product, in his view, is language models wired to "ground truth". That's why OpenAI bolted search onto ChatGPT in the first place.
So the model has to pull its answer from somewhere. Might as well be you.
That's not a slogan, it's a brief. It means earning the citations, structuring content so it can be retrieved, and building the off-site presence that tells the model who counts. The work compounds, and it sits right alongside everything else in a digital marketing programme rather than off to the side.
And the bigger shift is already visible in the same I/O announcement: Google confirmed search agents that run in the background, scanning the web on a user's behalf. The endgame isn't "be cited". It's "be recommended", and eventually "be chosen" by an agent comparing rates while the customer sleeps. For anyone in banking and finance, that's the next two years, not the next decade.
This is part one of a three-part series on winning in AI search. Next: why getting cited is only half the job.