Something quiet has changed in the way people find businesses, and most owners have not noticed yet. A year ago a customer looking for a supplier typed a few words into Google and worked down a list of blue links. Today a growing share of them open ChatGPT, or read the summary Google writes at the top of the page, and never scroll to the links at all. Ofcom now puts UK adult use of AI chatbots at around 27 percent, and BrightLocal reports that close to half of consumers now ask AI for a business recommendation, a habit that barely existed two years ago. The behaviour is moving faster than the marketing advice around it.
This matters because the answer these tools give is not neutral. When someone asks an AI assistant who the good web designers in Bristol are, or which accountant handles limited company work well, the model names a handful of businesses and leaves the rest out. Being named is the new front page. Being left out is the new page two, except there is no page two any more, because the person has their answer and has closed the tab.
Why your rankings no longer tell the whole story
For years the goal was simple. Rank on the first page for the terms your customers use, and the clicks follow. That logic is breaking down in front of us. Ahrefs measured clicks to the top result falling by around a third once Google places an AI Overview above it, and barely one percent of people click a source link inside an AI summary. You can hold your hard-won position at the top of the results and still watch the traffic thin out, because the customer got what they needed without ever leaving the page.
The instinct is to treat this as a threat, and for businesses that do nothing it will be. The opportunity is that the field is wide open. Very few of your competitors understand how these systems choose who to mention, so the ones who act now will own the answers for their category while everyone else is still arguing about keywords.
How AI decides who to name
An AI assistant does not rank pages the way a search engine does. It reads across the web, builds a picture of who is credible on a subject, and assembles an answer from the sources it trusts. Two things follow from that.
The first is that clarity beats cleverness. These models favour content that states things plainly and answers the actual question a person asked. A page that circles a topic for eight paragraphs before saying anything useful gives the model nothing to lift. A page that answers the question in the first two sentences, then backs it up, is far easier to quote. When we rebuild content for AI visibility, we are often not adding words. We are cutting the throat clearing and putting the answer where the machine can find it.
The second is that evidence earns citations. Analysis of what actually gets quoted shows that adding real statistics lifts the chance of being cited by around a third, and quoting authoritative sources lifts it further still. Vague confidence does nothing. Specifics do the work. This is why the businesses that win here tend to be the ones willing to share real numbers, real methods and real prices, rather than the ones hiding behind soft language and a contact form.
There is a reputational layer underneath all of this that you cannot fake. These models pay attention to how often and how consistently a business is described across the wider web. Your own site matters, but so does whether your name, your location and what you do line up everywhere else a machine might read them. A business that is described one way on its site, another way in a directory and a third way on an old profile looks uncertain to a system whose whole job is to decide what is true.
What we actually do about it
When we take this on for a client, we start by asking the assistants directly. We put the questions a real customer would ask into ChatGPT, Google and Perplexity, and we see who gets named. It is a blunt and honest test, and it usually lands harder than any report, because the client watches their competitor get recommended in plain English while they go unmentioned.
From there the work is structural. We make sure the pages that answer real buying questions exist and say something worth quoting. We add the marks in the code that tell a machine what your business is, where it operates and what it charges, so there is no ambiguity to resolve. We tidy the way you are described across the web so the story is the same everywhere. And we build the kind of specific, evidence-led content that these systems reach for, because a page that makes a clear claim and proves it is a page an AI is happy to stand behind.
None of this replaces good search practice. A fast, well-built site that ranks properly is still the foundation, and it always will be. What has changed is that ranking is no longer the finish line. The finish line is being the answer.
The window is open now
The reason to move on this in 2026 rather than 2027 is that the models are forming their view of every market as we speak, and early, credible sources carry disproportionate weight. The businesses that become the trusted answer for their category now will be hard to displace later, in the same way the sites that took search seriously fifteen years ago still enjoy the advantage today.