"Generative Engine Optimisation: how to get cited in AI responses"

Suppose someone asks ChatGPT which company can help them improve the visibility of their online shop. The answer is polite, comprehensive and well-reasoned, containing three names. Yours isn’t among them. Not because you rank poorly on Google, but because nobody has ever given any thought to how an AI system retrieves, reads and cites your pages.

That’s what Generative Engine Optimisation is all about. This article answers one question: how do you arrive at that answer? You can read about what AI Overviews do to your traffic and your click-through rate, just how bad the impact is, and how to adapt your content portfolio accordingly in AI Overviews and Your SEO Strategy. This is about the other side of the coin: being quoted.

What exactly GEO and AEO are

Generative Engine Optimisation (GEO) involves optimising your content and online presence so that AI-driven platforms mention, quote or recommend your brand when someone asks a question. These platforms include ChatGPT, Google AI Overviews, Google AI Mode, Gemini, Microsoft Copilot and Perplexity.

You’ll also come across Answer Engine Optimisation (AEO), and sometimes LLMO or simply AI SEO. They largely describe the same objective. AEO is the most precise term when your specific goal is a citation; GEO is broader and covers any form of visibility in generated output. In this article, we’ll use GEO as an umbrella term and AEO when we’re specifically referring to citations.

The difference compared to traditional SEO lies in the success metric. SEO ensures that your content is crawled, indexed and found, and secures you a position on the results page. GEO determines whether that same content is selected when a model compiles an answer, and secures you a position within that answer. In short: SEO gives you the right to be found; GEO gives you the right to be recommended. A page can rank very highly and never appear in an AI response. The reverse is just as likely to happen.

That calls for different KPIs. With SEO, you look at rankings, impressions and clicks. With GEO, you look at citations, brand mentions, your share of AI answers compared to competitors, and the quality of the traffic coming in via source links. Without a solid SEO foundation, this is of little use, so if the basics aren’t in place yet, that’s where you start. In the guide to search engine optimisation explains what that foundation entails.

Why you need to do something about this now

The Netherlands is right in the thick of it. AI Overviews began appearing here in May 2025, and in October 2025, Google officially rolled out AI Mode to Dutch users: a full search mode featuring follow-up questions and contextual sources, not just a summary above the results.

The figures on the impact are stark. Research by Ahrefs in December 2025, based on 300,000 keywords, shows that the click-through rate for the number one position drops by 58% as soon as an AI Overview appears above it. The proportion of search queries ending without a single click rose from 56% in 2024 to 69% in 2025. Gartner predicts that by 2026, a quarter of organic search traffic will shift to AI chatbots and virtual assistants rather than traditional search clicks. We’ll explore what this shift means for your traffic, and which content is most at risk, in AI Overviews and Your SEO Strategy.

The most telling figure isn’t about Google but about the market. Only 20 per cent of organisations have started implementing AEO, whilst 70 per cent expect it to have a significant impact on their digital strategy within one to three years. That gap between knowing and doing is precisely where you can build a head start.

There’s also some good news, which is rarely mentioned. Visitors who land on your site via an AI-generated referral are more valuable than average. Research shows that visitors referred by AI convert on average 4.4 times as effectively as standard organic traffic and spend 68 per cent more time on the site. This makes sense: the model has already done the preliminary research, so anyone who clicks through is genuinely interested in finding out more.

Finally, a fair counterpoint, because that’s part of the picture. Google still processes around 373 times more search queries than ChatGPT (SparkToro and Datos, 2024). Traditional search is by far the largest channel and will remain so for the time being. The conclusion, therefore, is not that you should stop doing SEO, but that you should build another layer on top of it.

Rankings and citations are becoming increasingly disconnected from one another

This is the key insight of the whole subject, and at the same time the point on which the most contradictory figures are circulating. They are less contradictory than they appear, as long as you specify which system was measured and when.

There is a clear time series for Google AI Overviews. In July 2025, 76% of the cited URLs still came from the organic top 10. By February 2026, this had fallen to 38%, with 31% of citations coming from pages ranked 11th to 100th and the remaining 31% from pages outside the top 100. The same pattern can be seen at the GEO agency Brandlight, which observed the overlap between Google’s top rankings and AI citation sources falling from 70% to below 20%. This decoupling is therefore not a one-off but a trend that is continuing with each measurement.

Within Google, however, ranking still makes a difference. Pages in position 1 have around a 58 per cent chance of being cited in an AI Overview, compared with 14 per cent for those in position 10. A high ranking is no longer a guarantee, but it is an advantage.

Standalone chatbots work differently. Fewer than 10 per cent of the sources cited by ChatGPT, Gemini and Copilot appear in Google’s organic top 10 results for the same query. That figure is lower than the percentages above because it relates to different systems: AI Overviews run on the Google index, whilst Perplexity uses its own index and ChatGPT draws on the Bing index plus its own crawled data. So do not compare those figures with one another, but read them side by side.

And then there’s the issue of scarcity. Large language models cite an average of 2 to 7 domains per answer. Whereas a search results page has ten organic listings and a second page beyond that, an AI answer is a short list of a handful of names. Competition has therefore become fiercer, not less intense.

That is both encouraging and unsettling. Encouraging because you can gain visibility without first making it into the top 10. Unsettling because a strong SEO dashboard gives you a false sense of security.

How AI systems retrieve sources

To understand why one page is cited and another isn’t, it helps to know what happens from a technical perspective. Most AI search systems use Retrieval-Augmented Generation (RAG). The model first retrieves relevant documents from external sources and combines them with its own knowledge to form an answer. So it doesn’t answer your question off the top of its head; it searches first and then writes.

When the page is fetched, it is tokenised: the HTML is broken down into chunks which the model places in a vector space, where meaning is expressed as the distance between points. The user’s query ends up in that same space, and whatever is close by is considered as a source. The practical consequence is that if your headings lack a semantic hierarchy, your entities aren’t named consistently, and a paragraph touches on three topics at once, your page becomes a vague cloud rather than a clear focal point. The model will then omit it, not out of unwillingness but because it cannot determine exactly what you are answering.

Research has been carried out into what actually matters. In 2025, SE Ranking analysed 400,000 URLs and found that traditional tricks such as keyword stuffing have no effect on AI visibility – and can sometimes have a negative impact. What does work is factual density: authoritative quotes, concrete statistics and references to primary sources. This can increase the visibility of lower-ranked websites by up to 40 per cent. Research from Princeton University suggests a similar order of magnitude for targeted GEO techniques.

For challengers, that is the most interesting finding to emerge from this whole topic. The link landscape is dominated by players who have been there for twenty years, but the battle for citations is won through structural clarity and information density. You can still sort that out this month.

How does each platform choose its sources?

Not all AI platforms work in the same way, and these differences determine what content you create.

Google AI Overviews

AI Overviews are summaries displayed above the standard search results, with links to the sources used. They are selected based on traditional SEO signals combined with informational value: authoritative content, a sound technical structure and a direct focus on providing answers. Google itself is clear about the technology: there are no special tags to enable AI features. The fundamentals remain the same; it is simply that the success signal has shifted from position to citation.

ChatGPT Search

ChatGPT bases its search keywords on Bing’s web index, as well as its own crawled data. The system places greater emphasis on exact matches and directness than Google does. A clear question-and-answer structure works particularly well here: pages containing a specific question and a to-the-point answer are more likely to be reproduced almost verbatim.

Perplexity

Perplexity is a research assistant that displays source cards alongside every answer. It uses its own index, not Google’s, so the selection of sources does not automatically overlap with the top 10. Recency carries more weight here than on other platforms: an article from last month will often outrank an older article with the same content. E-E-A-T signals such as an author’s name, professional context and links to external profiles are actively used to assess a source’s credibility.

The key conclusion is more important than the individual characteristics: a page that ChatGPT likes to cite may not even appear in Perplexity. And vice versa. So optimise broadly. Good structure, up-to-date content and demonstrable authority work across all three platforms simultaneously, whilst a trick that works on one platform will yield nothing on the other two.

Start by allowing the AI crawlers in

Of all the points in this article, this is the action that offers the quickest route to results, and it will take you just a quarter of an hour. Check your robots.txt file for blocks on GPTBot and OAI-SearchBot (OpenAI), PerplexityBot and ClaudeBot (Anthropic). Anyone who blocks these bots will simply not be considered as a potential source. No citation, no mention, no discussion.

This happens more often than you might think. Many sites added a block sometime in 2024 or 2025 due to concerns about content usage, or are hosted by a provider that blocks AI bots by default at server level. So don’t just check your robots.txt file; also check whether your firewall or CDN is blocking these user agents.

If you want to protect certain content, do so selectively. Restricting access to a customer portal is justifiable. Closing off your public marketing and information pages whilst at the same time complaining that you’re not mentioned anywhere is not.

What to change in your content

Put the answer at the top. Open each section with a direct answer of 40 to 60 words, and only then provide context and supporting arguments. No preamble, no atmospheric introduction. Each H2 or H3 block then functions as a self-contained, quotable unit that a reader can pick up without needing to understand the rest of the page. This pattern is known as ‘answer-first’ writing and is the most underrated technique on the entire list.

Enter the query behind the search. People type different things into a chat window than they do into a search bar. Not ‘refinance a loan’, but ‘can I refinance my loan without a penalty?’. Use headings that ask the actual question and answer them directly below.

Increase the density of facts. Specific figures, dates, definitions and references to primary sources make content citable. A paragraph containing no facts offers the user nothing that will help them further. Include the source in the running text, as that is precisely the part that will be quoted.

Think in terms of entities, not keywords. AI systems work with people, organisations, concepts and the relationships between them. Ensure that your brand is consistently recognisable as an entity: use the same name everywhere, provide structured organisational data, maintain a complete Google Business Profile, and ensure your brand is listed on platforms that the model already recognises.

Build up your authority on the subject rather than writing individual articles. A single good page won’t convince a model. A coherent cluster comprising a solid main page surrounded by supporting articles, all linked to one another, demonstrates that you’ve truly mastered a subject.

Keep it up to date. Perplexity places great emphasis on the publication date, but Google and ChatGPT also regard fresh content as more reliable. Update your strategic pages with new data and ensure that the ‘dateModified’ in your schema matches what you have actually updated.

Make sure you’re visible outside your own website. AI systems don’t just crawl your domain. YouTube now accounts for 18.2% of all AI Overview citations that come from outside the top 100 search results. Videos, podcasts, industry platforms and LinkedIn are all taken into account.

If you want to appear in search results for queries containing a place name, this involves local optimisation: your Google Business Profile, reviews and location pages. We’ll go into more detail on this in Local SEO for your area.

Schema markup is infrastructure, not a button

Structured data makes it clear what your content is about, who wrote it and how it relates to known entities. Use JSON-LD: Google recommends this format, and machines can process it more easily because it is neatly separated from the HTML.

The types that matter most for AI visibility:

• Article, with ‘author’ as a Person and with ‘datePublished’ and ‘dateModified’ correctly entered
• FAQPage, for Q&A blocks
• HowTo, for step-by-step guides and instructions
• Organisation, with ‘sameAs’ links to authoritative external profiles for entity recognition
• Product and Offer, for the visibility of online shops in AI responses

There is some confusion surrounding FAQPage, so let’s be clear: as of May 2026, FAQPage rich results will no longer appear on the standard search results page for most websites. Their value for AI Overviews, Perplexity and ChatGPT remains unchanged. Its purpose has shifted from SERP enrichment to serving as a citation source, and that is no reason to remove the markup.

And now for a word of caution. Schema is a correlative factor, not a direct ranking factor, and certainly not a button that activates AI responses. Google has confirmed that there is no special markup that activates AI features. The figures you see on this are correlations: pages with FAQPage schema appear 3.2 times more frequently in AI Overviews, whilst fully completed Product and Reviewmarkup achieves a citation rate of 61.7 per cent compared to 41.6 per cent for generic markup, and analysis by AirOps shows 2.8 times higher citation scores for pages that combine clean structure with good schema implementation. Good reasons to do it, but the mechanism is indirect: better machine readability, stronger entity recognition, clearer E-E-A-T. If you apply schema to poor content, you’re still left with poor content. The fact that only 12.4% of websites implement structured data at all does, however, make it a low-cost advantage. Validate your markup with the Google’s Rich Results Test.

Authority that you cannot write yourself

AI systems don’t just learn from your website; they learn from everything that’s been written about you. In fact, they have a distinct preference for earned media – independent sources that write about you – over what you say about yourself. That’s exactly what a critical reader would do too.

The figures all point in the same direction. Domains with a high number of brand mentions on platforms such as Quora and Reddit are, on average, four times more likely to be cited than domains with hardly any activity on those platforms. Domains with profiles on Trustpilot, G2 or Yelp are three times more likely to be selected as a source by ChatGPT.

In practical terms, this means: digital PR, guest articles in trade journals, interviews, inclusion in round-ups, active review collection and consistent listings in industry directories. Don’t call it link building, because the aim is different. This is reputation management for systems that derive your reputation from what others publish about you.

What doesn’t work

There is a lot of advice circulating on this topic that costs money and time without delivering any results. The most persistent piece of advice concerns llms.txt, sometimes in combination with a separate Markdown version of your pages for language models. In its own documentation on optimising for AI features, Google has confirmed that Google Search does not use that file and that a separate Markdown layer is not necessary. If you come across this advice on a blog or in an audit report, you now know that it is incorrect. You’re better off investing those hours in the structure of the pages you already have.

Two other pitfalls. Keyword stuffing doesn’t work here and can backfire, as the SE Ranking study showed. And abandoning traditional SEO principles is the most costly mistake of all: technical aspects, content and authority remain the foundations on which everything rests. AI Overviews are based on the same index and the same quality assessments as standard search results. Anyone who neglects the basics will also perform poorly in AI-generated answers.

How to measure AI visibility

This is where most teams fall short, because AI visibility isn’t included in your standard reporting. You can achieve excellent results in ChatGPT and Perplexity without ever seeing it reflected in Google Analytics. Only 16 per cent of brands systematically monitor AI search performance, which explains why so few organisations know whether their efforts are paying off.

Since 3 June 2026, Google Search Console has featured a separate performance report for generative AI, which shows how often links to your website have been displayed in AI Overviews and AI Mode. This is a real step forward, as previously you couldn’t tell whether a drop in traffic was down to traditional SEO or AI. Do, however, be realistic about what it is: a visibility report. Click-through and conversion data per AI channel are still missing. Some accounts also have a filter for AI Overviews under ‘Result type’ in the Search Results section.

In addition, these are the methods that yield the best results:

• A fixed prompt set. Come up with ten to twenty questions that your target audience actually asks, and submit them each month to the platforms that matter to you. Make a note of whether you’re mentioned, whether there’s a source attribution, what the tone is like, and which competitors are featured. The latter is the AI equivalent of Share of Voice, also known as Share of Model.
• Track brand searches. If search volume for your brand name increases, this is often a reflection of mentions in AI-generated answers that you’re otherwise unaware of.
• Identifying AI traffic in GA4. ChatGPT adds a distinctive marker to links in its replies, allowing you to identify that traffic as a separate source.
• Compare the conversion rate with the volume. Fewer clicks with stable conversion rates don’t mean you’re failing; it just means that visitors who are just browsing have found what they’re looking for elsewhere, whilst serious visitors still end up with you.
• Specialised tools. Ahrefs Brand Radar This is followed by brand mentions relating to Google AI Overviews, AI Mode, ChatGPT, Perplexity, Gemini and Copilot. Semrush has a similar index, and platforms such as Peec AI, Otterly.ai and Profound go into greater depth at the prompt level.

There’s one check that’s consistently overlooked: is what AI systems say about you actually correct? Around 30 per cent of marketers report that AI systems describe their brand inaccurately. A service listed incorrectly or an out-of-date price is more damaging than not being mentioned at all, and the only way to find out is to check for yourself.

This isn’t a quick fix

We’d be doing you a disservice if we presented this as a checklist to tick off on a Friday afternoon. It takes weeks or even months for a revised article to be assessed differently, especially if your authority still needs to grow. Platforms that index in real time, such as Perplexity, respond more quickly than models that also rely on training data. And because models are updated regularly, citation patterns can change abruptly. So plan for ongoing optimisation, not for a project with an end date.

Small and new brands can certainly succeed here, but not by taking on the big players head-on. Choose a specific niche that the big brands are ignoring, cover it comprehensively with factual, well-structured content, and build on that. They have brand awareness and a wealth of external mentions. You have specialisation, speed and depth. In a response that mentions only a handful of domains, that’s a real advantage.