Search is changing, and the change is now measurable. When Google places an AI summary at the top of its results, the top organic listing loses well over half of its clicks. Ahrefs has measured a 58% drop in click-through rate (opens in a new tab) for the top-ranking page on searches that trigger an AI Overview, up from 34.5% just eight months earlier. Independent browsing-data studies, including one from the Pew Research Center (opens in a new tab), point the same way. People increasingly get their answer on the results page and never click through.
It is worth saying what has not happened, too. The loudest forecasts called for traditional search to collapse by a quarter in 2026 (opens in a new tab). At the halfway point of the year, nothing on that scale has arrived. The real shift is quieter and more durable: not the end of search, but a change in how people find things and which sites get surfaced when they do.
So the headline is not "search is dying". It is "the rules for getting found are changing". The discipline that has grown up around those new rules is called generative engine optimisation, or GEO: the work of getting your business cited and recommended inside AI-generated answers, rather than ranked in a list of blue links.
Plenty of people are now selling GEO as a fresh bag of tricks to bolt onto your site. We think that is the wrong way to look at it, and chasing those tricks will cost businesses money on tactics that do not last.
Here is the more useful read. AI search does not reward sites that game it. It rewards sites that were built properly in the first place. The work that earns a citation is mostly the same work that makes a site genuinely good. That is not a consolation prize. It is the opportunity.
The wrong lesson
The temptation is to treat AI search like early SEO: find the lever, pull it, win the traffic. Stuff in some keywords, add an FAQ block, sprinkle the right phrases, and wait for the engine to notice.
The evidence says this does not work. The foundational academic study on the subject, GEO: Generative Engine Optimization (opens in a new tab) from a team at Princeton, tested a range of content strategies across multiple AI engines. Keyword stuffing, the classic SEO shortcut, performed worse than doing nothing. What actually lifted visibility was adding relevant statistics, quoting credible sources, and citing authoritative references. In short, the things that make a page more trustworthy to a human reader are the things that make it more quotable to a machine.
Ahrefs, which has run some of the largest studies on AI citations (opens in a new tab), found the same pattern from the other direction. Word count has almost no correlation with whether content gets cited. Long does not win. Clear wins. Their data points to answer-first structure, a high density of named entities, genuine freshness, and brand mentions across the wider web as the things that move the needle. Brand mentions, in fact, outweighed backlinks as a signal by roughly three to one.
None of that is a trick. It is editorial discipline and credibility, measured.
What the evidence actually rewards
Strip away the jargon and the durable signals are consistent across the research and across platforms.
Answer the question directly, and early. AI engines retrieve a passage from your page and often quote it close to word for word. The passages they choose are the ones that answer a real question in plain, declarative sentences near the top, rather than burying the point under evocative copy. Write the answer first, then the supporting detail.
Be specific and credible. Statistics, named entities, real examples, and citations to sources an engine can verify all raise the odds of being quoted. Vague, adjective-heavy copy gets skipped.
Keep it current. Content that is genuinely updated when the facts change is cited more than content that has sat untouched for two years. Changing the date without changing the substance does not count.
Build authority off the page. Being mentioned, referenced, and discussed elsewhere now matters more than the old link-counting game. Reputation is a ranking signal, not just a vanity metric.
Google's own guidance lands in the same place. Its advice on succeeding in AI search (opens in a new tab) is, almost word for word, the advice it has given for years: make unique, useful content for people, give them a good page experience, and make sure the content can actually be accessed. There is no separate AI index and no secret AI algorithm to exploit. The fundamentals did not get replaced. They got more important, because there is now far less room on the page for mediocre content to scrape by.
The engineering layer most teams skip
Most GEO advice stops at the writing. That is where it goes wrong, because a large share of AI search failure is not editorial. It is technical.
If an AI engine cannot crawl your site, none of your content matters. We regularly see sites that block AI crawlers in robots.txt without realising it, ship their main content only after JavaScript runs so the engine sees an empty page, or carry no structured data at all, leaving the machine to guess what the page is even about. You can write the most quotable paragraph on the internet. If the engine cannot reach it or parse it, you are invisible.
This is the layer that separates a content exercise from real engineering, and it is the layer we care most about. Getting it right means treating your site as a machine-readable product, not just a brochure.
When we recently rebuilt the PodcasterPlus (opens in a new tab) marketing site, we moved it off WordPress and onto Svelte, and built the AI-readiness in from the start rather than bolting it on later. In practice that meant a few specific things. Pages are server-rendered, so an engine sees the full content immediately rather than a blank shell. Structured data describes the product, its features, and the entities on each page in a format machines can read directly. The robots.txt is deliberate about which crawlers are welcome, instead of accidentally shutting the door on the ones that matter. And the site ships an llms.txt file, a plain-text map that tells AI engines what PodcasterPlus is and where the canonical information lives.
Here is the nuance most GEO pitches skip. Structured data tells an engine that you exist and what your entities are. It does not get quoted in an answer. Prose does. So the technical layer and the content layer have to work together: the schema makes you legible, and the writing makes you quotable. We write the pages to answer the questions people actually ask, in plain sentences, near the top, so that when someone asks an assistant for the best podcast tool with a particular feature, there is a clean sentence ready to be lifted.
To be honest about it, you do not strictly have to write this way. Google has said its systems can pull a passage from an ordinary, multi-topic page without you chopping everything into question-and-answer chunks. So answer-first writing is an enhancement, not a rescue for something broken. We do it anyway, because it also makes the page clearer for the human reading it and easier to act on. The two audiences want the same thing.
What this means if you run a business
You do not need a GEO budget or a new agency on retainer. You need to stop treating AI visibility as a separate campaign and start treating it as a property of a well-built site.
In practice, that is a short list worth checking. Can AI crawlers actually reach your content, or is robots.txt or a heavy JavaScript front end quietly hiding it. Does your most important content render server-side. Do your key pages answer the obvious questions clearly and near the top, in language a person would actually use. Is there structured data describing what you do. Are you being mentioned and referenced beyond your own domain. And is anything important genuinely kept up to date.
Answer yes to most of those, and AI search is not a threat to you. It is a filter that happens to favour the businesses that did the work. The companies that lose are the ones who treated their website as a cost to minimise. The ones that win treated it as a product.
That is the part we find genuinely optimistic. For years, building a site properly, with clean architecture, real content, and an eye on how machines read it, was hard to justify to a finance director because the payoff was diffuse. AI search has made the payoff concrete. Good engineering and honest content are no longer just good practice. They are how you get found.
If that sounds like more of an engineering problem than a marketing one, you are reading it right. That is exactly how we think it should be approached.
Not sure how your own site would score against that list? Get in touch and we will tell you how it looks to both people and the engines now doing the searching.