Generative Engine Optimisation (GEO): What the 2026 AI Search Data Means

A new report claims the era of ten blue links is over, and that the brands winning now are the ones written into the answer itself. The data is worth your attention. The urgency is worth a second look.

For the busy leader: the one-minute version

Search has stopped handing out links and started handing out answers. AthenaHQ's State of AI Search 2026, built on millions of responses across seven AI models, reports that several leading publishers have lost more than half their search traffic since AI answers arrived, and that a small number of brands now dominate what the machines say.

The shift it describes is real and it matters. The report is also a sales document, written by a company that sells the tools to measure the problem. So read it for the data and keep a cool head about the alarm.

The practical message survives the scepticism. Make your best knowledge public, structured and useful, then check whether AI systems are actually quoting you. That is the whole game.

For twenty-five years, the deal between a business and a search engine was simple. You published something useful, Google ranked it, and if you ranked well enough, people clicked through to your site. Ten blue links, a race to the top of them, and traffic as the prize. That deal is now being quietly rewritten, and most business owners have not been shown the new terms.

AthenaHQ's State of AI Search 2026 is one attempt to describe those terms. It draws on millions of AI-generated responses across seven leading models, including ChatGPT, Google's AI Overviews and AI Mode, Gemini, Copilot, Grok and Perplexity, with the source data consolidated between December 2025 and March 2026. It is a substantial piece of work. It is also, and this matters, a report written by a firm whose business is selling generative engine optimisation software. That does not make the findings wrong. It does mean the conclusions lean, understandably, towards "act now or lose". My job here is to separate what the data shows from what the vendor would like you to feel about it.

The end of the blue links

The headline claim is rather blunt. Search engines have moved beyond the ten blue links and now deliver answers directly on the results page, and according to the report several leading sites have seen traffic fall by more than 50% after AI features rolled out. If you run a business that depends on people finding you through Google, that number should stop you in your tracks.

It should also prompt a question the report does not fully answer: more than 50% for whom? The figure describes several leading sites, not the average business, and the pain is not spread evenly. A publisher that lived on quick factual answers, the kind an AI can now serve without a click, has lost its lunch. A specialist firm selling a considered service may barely feel it. Averages hide as much as they reveal, and "up to" is not "typically". The direction of travel is not in doubt. The size of the hit depends entirely on what you do and who you serve.

The report frames this optimistically. Fewer visitors, it argues, but the ones who arrive are better educated and closer to buying, because the AI has already done the explaining. That is plausible. It is also the glass-half-full reading. The same facts support a gloomier one: fewer touchpoints mean fewer chances to shape how someone understands your market before they decide, and an answer engine sitting between you and the customer is a middleman you do not control. Both readings are honest. The report only tells you one.

What the data actually shows

Underneath the framing sits some useful benchmarking, and this is where the report earns its keep. A few numbers are worth holding on to, all of them from the State of AI Search 2026.

On average, a given brand is mentioned in 16.3% of AI responses to discovery prompts. The leaders in each category reach 56.5%, roughly three times the market average. Share of Voice, the proportion of brand mentions a company captures against its competitors, shows the same concentration: the top brand in a category averages 33.6%, the second 19.5%, and the third 13.3%. In other words, the winner does not take all, but the winner takes a great deal, and the drop-off after third place is steep.

This is the part I would take most seriously, because it is a pattern, not a projection. AI answers reward incumbency and visibility. The rich get richer. It is the same uncomfortable dynamic the 2026 AI adoption data keeps showing: advantage compounds for those already moving. If you are already the recognised name in your field, the machines will tend to repeat that back to enquirers. If you are not, you are largely invisible, and being invisible in an answer engine is worse than ranking tenth on a page, because there is no page to be tenth on.

One more figure is worth keeping, and this one plays straight into the vendor's pitch, fairly enough. AI systems cite a brand's own domain in only around 16% of responses. Roughly 84% of the time, the machine is answering about you using someone else's words, not yours. That is the real problem generative engine optimisation sets out to solve, and it is a real one.

Where the machines actually look

If AI is mostly quoting other people, the obvious question is: which people? The report's source data is the most quietly revealing part of the whole document. The most cited sources across all sectors are, in order, reddit.com (21.9% of citations), youtube.com (10.3%), en.wikipedia.org (7.3%), linkedin.com (6.5%) and forbes.com (5.3%).

The two biggest sources feeding the world's AI answers are a discussion forum and a video platform. Not your polished brochure site. Not the press release. A Reddit thread where strangers argue about what actually works. The machine has decided the most trustworthy account of your product is often a conversation you were never part of.

There is real variance underneath the average, too, and it undermines any idea of a single tidy playbook. Grok draws over half its citations, 51.8%, from Reddit alone. Perplexity and ChatGPT also lean heavily on it. Google's AI Overview spreads its net far wider. Some models cite as few as six distinct domains per answer, others as many as twenty-seven. Optimising for one is not the same as optimising for all, a nuance the "just do GEO" narrative tends to smooth over.

The report also maps where AI crawlers enter a website. Blog pages account for 37.5% of entry points, homepages 19.2%, and product pages 13.8%. And it categorises the content the machines prefer to quote: informational material leads at 36.2%, followed by comparative or best-option content at 23.2%. The lesson is old-fashioned and slightly deflating for anyone hoping for a secret. AI likes clear explanations and honest comparisons, published somewhere it can actually read them.

A new scoreboard, and why it needs watching

The report argues, reasonably, that the old measures of success are losing meaning. If nobody clicks, clicks are a poor scorecard. In their place it proposes a new s

The logic is sound. You cannot manage what you do not measure, and if buying decisions are increasingly shaped inside an AI answer, you need some way of knowing whether you appear in it. Treat it as a question of AI governance as much as marketing. There is, though, a self-interested edge worth naming plainly. Each of these metrics happens to be one that generative engine optimisation software, the kind AthenaHQ sells, exists to track. New metrics create new dashboards, and new dashboards create new subscriptions. That does not make them useless. It does mean you should adopt them because they answer a real question for your business, not because a report told you they are the future.

Is Generative Engine Optimisation a New Discipline, or SEO in a New Suit?

The report never quite asks itself the obvious question. Strip away the terminology, and how much of Generative Engine Optimisation is actually new?

Read the recommendations and you have seen every one of them before. Publish open, structured, trustworthy content. Answer the real questions your audience asks. Build authority through credible sources and fresh data. Monitor, measure and adjust. Any SEO specialist from 2015 would recognise the lot. Good content marketing has always been about being useful and being findable. For the most part, GEO is that same discipline, pointed at a machine that summarises rather than one that ranks.

That is not a criticism but a relief. It means you are not starting from zero, and it means the snake-oil version of GEO, the one selling secret tricks to game the algorithm, can be ignored. The new part is narrow, and it matters. You are no longer optimising to win a click. You are optimising to be included, accurately, in an answer that someone else's model writes on your behalf. Being the source rather than the destination is a genuine change of posture. It rewards firms that treat their expertise as something to give away in the open. It quietly punishes the ones who lock their best thinking behind a contact form.

It is like the difference between owning the shop on the high street and being the name the knowledgeable local recommends when a stranger asks where to go. The high street still matters. The recommendation now matters more.

What this actually means for you

Strip the report down to what a sensible owner-manager should do on Monday morning, and it comes to four honest moves. None of them requires you to panic, and none of them requires a new software subscription to begin.

• Publish your expertise in the open. Your best knowledge belongs on public, well-structured pages, not hidden behind a form. If a machine cannot read it, it cannot quote you. The blog and the homepage are where AI looks first, so that is where your clearest thinking should live.

• Write to answer real questions. Think about the exact questions a customer would type, or speak, into ChatGPT, and answer them properly. Comparisons, definitions and honest best-option guides are what the machines reach for most.

• Earn a presence where the machines look. If Reddit, YouTube, Wikipedia and industry publications are feeding the answers, showing up credibly in those places matters as much as your own website. You cannot fake this, and you should not try. You earn it by being worth mentioning.

• Check whether it is working, then decide. Ask the major models the questions your customers ask and see whether you appear and whether what they say is accurate. Do this by hand before you pay anyone for a dashboard. If your visibility really warrants continuous tracking, the tools exist. Reach for them second, not first.

  
  

That final point carries the whole argument. The report is right that a shift is underway and right about most of the practical response. Where I would push back is on the tempo. If you would rather not work this out alone, 360 Strategy provides AI consulting in Scotland to help leaders make these moves deliberately, not fearfully. "Those who act early win" is true of almost every change, and it is also exactly what you would say if you were selling the shovels.

Ready to find out whether AI is quoting you? Book a Clarity Call.

Questions worth considering

Before you commission a GEO project or dismiss the whole thing as hype, answer these properly.

• If a customer asked ChatGPT to recommend a firm like yours today, would your name come up? Have you actually checked, or are you guessing?

• How much of your useful expertise is public and readable, and how much is locked behind a form or trapped in a PDF?

• When an AI answers a question about your market using someone else's words, whose words are they, and are they accurate?

• Is your traffic really falling, or are you reacting to a headline number that describes someone else's business?

  
  

The closing thought

The State of AI Search 2026 is a good report with a commercial accent. Take the data seriously. Take the deadline it implies with a pinch of salt.

The shift it describes is real, and it rewards something reassuringly old-fashioned: being useful, in the open, to people who are trying to make a decision. That was good practice when the prize was a click. It is worth more now the prize is being the answer.

Being part of the results is no longer enough. Whether that alarms you or motivates you depends less on the technology and more on whether your expertise has ever really been findable in the first place.

Mark Evans MBA is founder of 360 Strategy, a growth strategy and AI consultancy based in Scotland.

Frequently asked questions

What is GEO?

Generative Engine Optimisation, sometimes called Answer Engine Optimisation, is the practice of making your content likely to be included and cited in the answers produced by AI systems such as ChatGPT, Gemini and Google's AI Overviews. Where SEO aimed to rank a page, GEO aims to make your expertise part of the machine-generated answer itself.

Is SEO dead?

No, and be wary of anyone who says so. Traditional search still drives significant traffic, and the same fundamentals, useful content, clear structure and genuine authority, underpin both SEO and GEO. What is changing is the prize. Visibility inside an AI answer is becoming a second front you cannot ignore, not a replacement for everything that came before.

Do I need to buy GEO software?

Not to begin with. You can test your own visibility for free by asking the major AI models the questions your customers ask and seeing whether, and how accurately, you appear. Paid monitoring tools become worthwhile once you have confirmed the problem is real for your business and you need to track it continuously. Start with the manual check.

Where does the data in this article come from?

All figures are drawn from AthenaHQ's State of AI Search 2026, which the publisher states is based on millions of AI-generated responses across seven leading models, with source data consolidated between December 2025 and March 2026. The interpretation, and the scepticism, are my own.

Source: AthenaHQ, State of AI Search 2026: How AI Search Engines Decide What Gets Seen. Figures described as "up to" or referring to leading brands or sites are drawn from the top performers in the dataset and should not be read as market averages.