GEO: Why Being Cited by AI Now Matters More Than Ranking

Search stopped being a list of links and became an answer. If your brand is not in the answer, it is not in the conversation, and no amount of ranking fourth will fix that. This is what generative engine optimisation is for, and it is not a rebrand of SEO.

What GEO actually is

GEO, or generative engine optimisation, is the work of getting your content cited inside AI-generated answers on Perplexity, ChatGPT, Google AI Overviews and Copilot, rather than ranked in a list of blue links. SEO competes for a position on a results page. GEO competes to be the source the answer is built from.

The distinction sounds academic until you look at what it changes. A ranking rewards the page. A citation rewards the paragraph. An AI system does not read your article and decide it is the best article; it looks for the specific block of text that answers the question it was asked, lifts that, and attributes it. So the unit of optimisation shrinks from the page to the section, and the question stops being “is this comprehensive” and becomes “can any single part of this be quoted on its own and still make sense”.

Why it matters now

Because the click is disappearing, and the numbers on that are no longer speculative.

The zero-click rate, the share of searches that end without the user visiting any website, has climbed from 50% in 2019 to 64.82% in 2026. When an AI Overview is present, it reaches 83%, against 60% when one is not. That is the whole game changing shape: the majority outcome of a search is now that nobody goes anywhere.

The effect on the sites that do rank is measurable and consistent in direction, though the studies disagree on magnitude, which is worth being honest about. Ahrefs, studying 300,000 keywords, found a 34.5% drop in click-through rate for the top-ranking result when an AI Overview appeared, and its updated February 2026 figure puts that at 58%. Amsive, working across 700,000 keywords, found a more conservative 15.49% average decline across all positions. Take the range rather than the headline: somewhere between a sixth and a half of the traffic that used to arrive from a first-place ranking no longer does.

The most useful measurement is not from an SEO vendor at all. Pew Research Center followed the actual browsing of 900 US adults across 68,879 Google searches, 12,593 of which returned an AI summary, and watched what people did next. Users who saw an AI summary clicked a traditional search result in 8% of visits. Users who did not see one clicked in 15% of visits, nearly twice as often. That is the traffic cost of an AI answer appearing above you, observed rather than modelled.

And then the number that should genuinely change your plans. In the same study, users who saw an AI summary clicked a link inside that summary in just 1% of visits.

The uncomfortable implication

Read those together and the usual GEO sales pitch falls apart. Being cited in an AI answer produces a click roughly one time in a hundred. Anyone selling GEO as a traffic channel is selling something the evidence does not support, and if you buy it on that basis you will judge it a failure within a quarter.

So why do it at all?

Because being cited is a different kind of asset. When someone asks an AI which branding agencies in Nairobi are worth talking to, the answer they receive either includes you or it does not, and that answer is now the shortlist. It arrives with the machine’s authority rather than an ad’s, it is not marked as marketing, and the person reading it did not have to compare ten directory listings to get there. You are not buying a visit. You are buying presence at the moment of consideration, which is the thing brand budgets have always been trying to buy and rarely got this cleanly.

It also means the metric has to change. If you measure GEO by sessions you will conclude it does not work. Measure it by whether you appear, in what position, for the questions that precede a purchase.

What the work involves

Three things determine whether an AI system can cite you, and they are not the three things SEO agencies usually lead with.

1. Can it get in

AI crawlers are separate from Googlebot and they are named separately: GPTBot and OAI-SearchBot for OpenAI, PerplexityBot, ClaudeBot, Google-Extended for AI Overviews and Gemini grounding, Applebot-Extended for Apple Intelligence. If any of those is disallowed in your robots.txt, your visibility on that platform is zero regardless of how good the content is.

This is the most common problem we find, and it is usually nobody’s decision. A security plugin, a caching layer or a host’s optimiser rewrites robots.txt during an update, the AI crawlers get swept up in a blanket block, and nothing visibly breaks. The site keeps ranking. It just quietly stops existing inside AI answers.

Worth knowing before someone offers to solve it for you: the popular idea that you can allow citation while blocking training does not hold up technically. Most platforms use the same crawler for both, so blocking the bot blocks both, and there is no standard mechanism to separate them.

2. Can it extract an answer

Structure, not volume. A definition in the first 200 words for a “what is” question. Numbered steps, verb first, for a “how do I” question. A table, not a paragraph, for anything comparative. Explicit question-and-answer pairs where people ask questions. Each section written so it survives being quoted out of context, because that is precisely what will happen to it.

The commonest failure is a genuinely good 3,000-word article that buries its answer in the eighth paragraph. It ranks. It never gets cited. The fix is almost always restructuring rather than writing more.

3. Will it trust you

AI systems do not extract the most relevant answer, they extract the most credible one. Named authors with real credentials beat anonymous pages. Attributed statistics beat vague claims: “Pew Research Center found users clicked a link inside an AI summary in 1% of visits, from 12,593 AI-summary searches in March 2025” is citable, “studies show most people do not click” is not. Original data beats commentary, because the model cannot find your data anywhere else. And schema markup matters more than it ever did for rankings, because it states as data what a model would otherwise have to infer from prose.

Where to start

In this order, because the first item can invalidate all the others.

1. Read your robots.txt. Confirm GPTBot, PerplexityBot, ClaudeBot and Google-Extended are not disallowed. Re-check it after any plugin or host update.
2. Pick ten questions a customer asks before they buy from you, phrased as they would actually type them.
3. Ask those ten on Perplexity and ChatGPT. Write down who gets cited. That list is your real competitive set, and it is often not the companies you thought.
4. Take the three questions you should own and restructure the pages that answer them: answer in the first 200 words, self-contained sections, a table if it is a comparison, sources named.
5. Add schema. FAQPage where you have questions, Article on your writing, Organization so a model knows what kind of entity you are and where.
6. Re-test monthly and record movement.

On that last point, expect it to be partly manual. There is no standard API across the AI platforms for citation tracking, so honest measurement means testing an agreed set of queries by hand and logging the results, with Google Search Console covering AI Overview impressions. Anyone offering you a fully automated citation dashboard across every platform is rounding up.

The short version

SEO earns the ranking. GEO earns the citation. The second one is where the attention is going, the traffic case for it is much weaker than the pitch usually admits, and the presence case is much stronger. Get your crawlers unblocked, structure your answers so they can be lifted, name your sources, and measure whether you show up rather than whether you got a click.

Sources

1. Ahrefs, AI Overviews Reduce Clicks by 34.5% (300,000-keyword study; updated February 2026 figure of 58% for the top-ranking result) - ahrefs.com/blog/ai-overviews-reduce-clicks
2. Amsive, 700,000-keyword analysis finding a 15.49% average CTR decline across all positions, reported in Zero-Click Search Statistics 2026 - digitalapplied.com/blog/zero-click-search-statistics-2026-complete-data
3. Zero-click rate 50% (2019) to 64.82% (2026); 83% with an AI Overview present against 60% without - arfadia.com/blog/zero-click-search-statistics-2026
4. Pew Research Center, Google users are less likely to click on links when an AI summary appears in the results, 22 July 2025. Browsing data from 900 US adults, March 2025; 68,879 Google searches, of which 12,593 produced an AI summary. Traditional result clicked in 8% of visits with an AI summary against 15% without; a link inside the summary clicked in 1% of visits - pewresearch.org

Figures were current at the time of writing, July 2026, and the studies above disagree with each other; where they do we have given the range rather than the most dramatic number. Two limitations worth stating plainly, since this piece argues for naming sources: the Pew data is US-sampled and was collected in March 2025, so it describes a market a year ahead of ours and a product that has changed since. The direction is not in doubt. The exact percentages, for a Kenyan audience in 2026, are an indication rather than a measurement.