SEO and visibilitygenerative engine optimization

What is Generative Engine Optimisation (GEO)?

Generative Engine Optimization (GEO): origin of the term, differences from SEO, how AI chooses its sources and what an SME in French-speaking Switzerland can do.

Mehdi B.

Co-founder · SEO & digital strategy

What is Generative Engine Optimisation (GEO)?
In brief

Generative Engine Optimization (GEO) refers to the practices that make content more likely to be picked up and cited in the answers of generative engines (ChatGPT, Perplexity, Google’s AI Overviews). The term comes from a research paper published in 2023. Our position: GEO extends good SEO, it relies on indexable, sourced and signed pages, and nobody can guarantee a citation.

Generative Engine Optimization (GEO) is the set of practices that make content more likely to be picked up, and cited with a link, in the answers written by generative engines: ChatGPT, Perplexity, Microsoft Copilot, Gemini or Google’s AI Overviews. SEO seeks to place a page in a list of results. GEO looks at what happens when a machine reads several pages, draws an answer from them and chooses which ones to mention.

At Smart Impact, we hold a simple position on this subject. GEO is not a separate profession that makes SEO obsolete: it extends well-done SEO, with stronger requirements for precision, sources and the identity of the author. On the other hand, there is no recipe that guarantees a citation, and we are wary of offers that promise one.

Where the term Generative Engine Optimization comes from

The term was not invented by an agency. It appears in a research paper titled “GEO: Generative Engine Optimization”, filed on arXiv on 16 November 2023, then accepted at the KDD 2024 scientific conference. Four of its six authors come from Princeton University and IIT Delhi; the other two are independent researchers based in Seattle. Their starting point can be put in a few words: engines that synthesise several sources into a single answer leave very little hold to those who produce the content, and there was no method to measure and then improve the visibility of a source in these answers.

To test it, the team built a benchmark called GEO-bench, made up of 10,000 queries spread across 25 domains. They then rewrote the source pages using nine different methods, and measured what share of the generated answer relied on each source, taking into account its position in the text. The most frequently quoted result is a visibility gain of up to 40%.

The detail is nonetheless more instructive than this headline figure. The most effective methods consist of adding citations, statistics and references to identified sources. Conversely, keyword stuffing, a reflex inherited from the early days of SEO, did worse than leaving the text as it was. A merely “authoritative” tone, without additional evidence, brought no significant gain in the main results.

Another result directly concerns SMEs. The strongest gains were observed on sources that started from fifth place among those provided to the engine: the method of citing sources increased visibility by 115.1% there, whereas it reduced it by 30.3% for the source placed first. In other words, a lower-ranked but better-documented page can gain space in a generated answer.

These figures call for caution, however. The engine tested was a laboratory prototype, which took the top five Google results and had GPT-3.5 write the answer. The authors repeated the test on Perplexity, but on a sample of only 200 queries. It is a serious demonstration, not a user manual for the assistants your customers use in 2026.

GEO, SEO, AEO and AI Overviews: what each term covers

These acronyms overlap, and the market readily plays on that. SEO, or search engine optimisation, aims at the ranking of a page in a search engine’s results. AEO, for Answer Engine Optimization, refers to the work that allows content to be picked up as a direct answer to a question, for example in a featured snippet or by a voice assistant. GEO extends this logic to engines that write a synthesis from several sources. In practice, AEO and GEO overlap heavily, and we treat them as a single discipline.

AI Overviews, finally, are not a discipline but a Google feature: a generated summary at the top of certain results pages, accompanied by links. Google also offers AI Mode, which handles more complex questions in several steps and returns a longer answer, also accompanied by links.

Term Nature What we are trying to achieve
SEO Discipline A place in the traditional results
AEO Discipline Being the direct answer to a specific question
GEO Discipline Being picked up and cited in an answer written from several sources
AI Overviews, AI Mode Google Search features These are surfaces where SEO and GEO play out, not a goal in themselves

For Google, the boundary is also thinner than people say. Its documentation on AI features in Search is explicit: there are no additional requirements to appear in AI Overviews or AI Mode, no special optimisation needed, no machine-readable file or text file “for AI” to create, and no particular schema.org structured data to add. A page must be indexed and eligible to be shown with a snippet in Search, like any other result. On Google, GEO therefore starts with healthy SEO, and our article on how to structure a site to be cited correctly by AI details the technical side.

How generative engines choose and cite their sources

An assistant that cites pages does not draw them from its memory alone: it fetches them from the web at the moment of answering. At Google, the documentation describes a mechanism called query fan-out. The system runs several related searches on subtopics and data sources, which lets it display a wider range of links than classic search. AI Overviews and AI Mode may use different models, so their links vary for the same question. At OpenAI, a dedicated crawler, OAI-SearchBot, visits sites so they can appear in ChatGPT’s search features.

Practical consequence: ranking first on the main query is no longer enough to explain who gets cited. Ahrefs analysed 863,000 results pages and 4 million URLs cited in AI Overviews, in a study published on 2 March 2026. Only 37.9% of the cited pages were among the top ten results for the query, and 31.0% did not even appear in the top hundred. In July 2025, the same company measured about 76%, but with a different detection method. Ahrefs attributes the gap partly to fan-out: Google cites pages that stand out on the sub-questions, not only on the question asked.

The type of site matters too. The Pew Research Center tracked the browsing of 900 American adults in March 2025, or 68,879 Google searches. Wikipedia, YouTube and Reddit were the most cited sources, in generated summaries as in traditional results, and together accounted for 15% of the sources in summaries. Government sites ending in .gov accounted for 6% of the sources in summaries, against 2% of traditional results.

That said, citing is not the same as citing correctly. In March 2025, the Tow Center for Digital Journalism published in the Columbia Journalism Review a test of 1,600 queries submitted to eight generative search tools, including ChatGPT Search, Perplexity, Microsoft Copilot and Gemini. The task was to identify the source of excerpts from news articles. Collectively, the tools were wrong in more than 60% of cases, from 37% for Perplexity up to 94% for Grok 3, and more than half of the answers from Gemini and Grok 3 pointed to invented or broken URLs. A company therefore needs to check not only whether it is cited, but also what is attributed to it.

Hand placing a document-shaped piece into a white jigsaw puzzle.

Why the subject already concerns SMEs

Generated answers reduce clicks to websites, and this is now measured. In the Pew Research Center study, internet users who saw a generated summary clicked on a traditional result in 8% of visits, against 15% without a summary. They clicked on a link placed in the summary itself in only 1% of visits. Bain & Company, for its part, surveyed 1,117 people in December 2024: about 80% of respondents rely on zero-click results for at least 40% of their searches, and the firm estimates that this behaviour lowers sites’ organic traffic by 15% to 25%.

Google defends a different reading. In its May 2025 advice to publishers, it states that clicks coming from AI Overviews are of higher quality, because these visitors spend more time on the site, and it recommends tracking conversions rather than click volume alone. Both findings can coexist. This data, however, concerns American internet users, and to our knowledge no equivalent measurement has been published for French-speaking Switzerland.

What an SME in French-speaking Switzerland can do in practice

Nothing that follows is a trick. These are practices that generative engines and classic engines reward for the same reasons, and that also serve the human reader.

A page that answers a specific question

Take a ten-person fiduciary firm in Lausanne. Its future clients do not only type “fiduciaire Lausanne”; they ask an assistant how to move from a sole proprietorship to a Sàrl, or what to prepare before the annual closing. A page that deals with just one of these questions, with the answer in the first paragraph followed by the conditions and exceptions, gives the engine a passage it can reuse without distorting it. Google is going in this direction: its May 2025 advice stresses original, “non-commodity” content, because people ask longer and more specific questions in AI search. Important content must also remain as text, and not be locked inside an image or a scanned PDF.

Dated figures and named sources

This is the soundest lesson from the 2023 paper. A figure with its origin and year, an attributed quotation, a reference to an official text (a federal ordinance, a cantonal directive, a standard) give the engine a reason to rely on your page rather than another. Keyword stuffing has the opposite effect.

Identified authors

In its guide to helpful content, Google asks whether it is evident to your visitors who wrote a piece of content, and whether pages carry a byline where one would expect it. It expressly encourages accurate author information, and states that among the criteria it groups under the acronym E-E-A-T, trust is the most important. In practice, an article signed by the person who actually does the job, with a page presenting their background, is worth more than a text attributed to “the team”.

Structured data that matches the text

Structured data, the invisible markup that describes a page to search engines, remains useful for presenting your company, your addresses or your articles. Google reminds us of two rules: everything in the markup must be visible on the page, and there is no special markup for AI. Markup that announces reviews, opening hours or services that are absent from the text is precisely what to avoid.

Choosing which AI crawlers you let through

Each AI provider has its own crawlers, and your site’s robots.txt file lets you allow or block them one by one. At OpenAI, OAI-SearchBot is used for search: a site that blocks it no longer appears in ChatGPT search answers, except as a simple navigational link. GPTBot, for its part, collects content that may be used to train models. Blocking it does not remove your site from ChatGPT search results, since that visibility depends on OAI-SearchBot. OpenAI states that a robots.txt change takes about 24 hours to be taken into account for search, and that ChatGPT-User, which visits a page at a user’s request, may not follow these rules.

At Google, the logic is similar. AI Overviews are part of search, so it is Googlebot that matters, and snippet directives such as nosnippet or max-snippet limit what can be reused. The Google-Extended token lets you refuse the use of your content to train Gemini models; according to Google, it has no effect on inclusion or ranking in Google Search. Our default recommendation is therefore to let Googlebot and OAI-SearchBot through, and to treat training as a separate business decision. The Tow Center test is a reminder that blocking is not watertight: some tools reproduced content from publishers that had nonetheless blocked their crawlers.

As for the llms.txt file, it is a proposal published in September 2024 by Jeremy Howard, still open to contributions, which aims to give AI agents a summarised entry point to a site. It is not a standard, and Google states plainly that no text file for AI is needed to appear in its AI features. We do not advise against it, but we do not present it as a lever.

Being present where assistants go to look

Some of the citations are not about your site but about pages that talk about you. Pew’s data shows the weight of Wikipedia, YouTube and Reddit; in Ahrefs’ study, YouTube accounts for 5.6% of the addresses cited by AI Overviews. An SME in Neuchâtel will not have a Wikipedia article, and that is not a reasonable goal. By contrast, an up-to-date Google business profile (Google lists it among the information to look after), a member page in a professional association, an article in the regional press or a video explaining your trade are all pages an engine can come across while looking for an answer.

How to check whether your company is cited

No tool gives a complete view today, but three checks are enough to know where you stand.

The first is manual. Draw up a list of around fifteen questions your customers really ask, then submit them every month to ChatGPT, Perplexity, Microsoft Copilot, Gemini and Google’s AI Mode. Each time, note whether your company is cited, whether what is attributed to you is accurate and which sites appear in your place. Answers vary from one tool to another and from one day to the next, which is why the trend over several months matters more than a single attempt.

The second goes through Google Search Console. According to Google’s documentation, traffic from AI Overviews and AI Mode is included in the performance report, under the “Web” search type. It is not isolated: you will see the overall effect, not a breakdown by feature.

The third goes through Google Analytics. According to OpenAI’s help centre for publishers, ChatGPT automatically adds the utm_source=chatgpt.com parameter to the links it sends, which makes it possible to isolate these visits in acquisition reports.

What is not established

Several claims circulate without foundation, and it is better to set them aside clearly. Google does not require any llms.txt-type file. No tag or format guarantees a citation: Google even states that meeting its technical requirements does not guarantee that a page will be crawled, indexed or displayed. The gains measured by the 2023 paper were measured on a prototype, and studies on AI Overviews citations diverge strongly depending on the method, as the gap between Ahrefs’ two successive measurements shows. No one, Smart Impact included, can promise that an assistant will cite your company. What we can do is bring together the conditions that make such a citation possible and check what actually happens.

It is one of the areas covered by our AI and automation support, which we always carry out alongside traditional SEO.

Frequently asked questions about GEO

Does GEO replace SEO?

No. On Google, AI features impose no requirement beyond those of traditional search: a page must be indexed and able to be shown with a snippet. GEO adds requirements for accuracy, sources and bylines, on a foundation that remains that of SEO.

Do I need to create an llms.txt file?

It is not necessary for Google, which states that no text file for AI is required. llms.txt remains an open proposal since September 2024, not a standard adopted by search engines.

Does blocking GPTBot prevent my site from appearing in ChatGPT?

No. According to OpenAI, GPTBot relates to model training, while appearing in ChatGPT search depends on OAI-SearchBot. You can block the former and allow the latter.

Can an agency guarantee that ChatGPT will cite my company?

No, and an offer that promises it should put you on your guard. Engines choose their sources themselves, their answers vary and their citations are not always accurate. Serious work focuses on the means: useful, sourced, signed pages that are accessible to the right crawlers, followed by regular monitoring.

If you want to know how your company appears in these answers today, it is a good starting point for the first conversation with our team.

Sources

GEOSEOAI
Mehdi B.
About the authorMehdi B.

Co-founder of Smart Impact. Hooked on the web from the start, he launched his first project in 2006: an online music magazine still running today. With close to 20 years of SEO experience, a Swiss federal diploma in marketing and a solid geek streak, he turns clients' ideas, sometimes hazy ones, into concrete digital projects with his team.

LinkedIn
Further reading