GEO, short for generative engine optimization, is the work that makes your site visible inside answers written by artificial intelligence, whether that’s Google’s AI Overviews, ChatGPT or Perplexity, while classic SEO goes after a place in the list of links. Google switched these generated answers on in France on 22 July 2026, and Switzerland has had them since March 2025. For a business owner the practical consequence comes down to two points: GEO extends the search work you already know rather than replacing it, and the definition of success changes because a citation without a click is now worth something.
What is GEO exactly?
GEO covers the writing and structure choices that raise the odds of an AI engine pulling your content into its answer and naming you as a source. The term comes from a research paper presented by Pranjal Aggarwal and his co-authors at the ACM SIGKDD conference in 2024, which put a name on a very concrete problem for anyone who publishes online.
A generative engine doesn’t behave like the list of links you’re used to. It breaks your question into several sub-questions, goes looking for pages that answer each of them, then writes one single reply by taking pieces from what it found, with a few sources shown beside or beneath the text. Your page stops being a destination somebody picks from a list and becomes raw material that the machine cuts up and reassembles in its own words.
The researchers built a benchmark called GEO-bench to measure how much the writing itself matters, and they showed that some approaches lifted a source’s visibility inside the generated answer by up to 40%, with wide variation depending on the subject area. That number gets quoted a lot and misread almost as often, so it’s worth stating plainly: it doesn’t mean GEO wins you 40% more traffic, it means that on that particular benchmark, the way a passage is written measurably changes the odds of it being reused. That’s encouraging news for a smaller site, because reuse depends heavily on how clear the passage is, rather than purely on how powerful the domain hosting it happens to be.
What changed on 22 July 2026?
Google turned on AI Overviews and AI Mode in France on Wednesday 22 July 2026, which means French searchers now see an AI-written answer above the usual results, on mobile and on desktop alike. The rollout is progressive across accounts, devices and queries, so you might not have it on your own screen yet while a customer already has it on theirs.
France was the last big European market to flip, because of the neighbouring rights dispute between Google and French press publishers that had already cost the company 500 million euros in fines in 2021 and another 250 million in 2024. Switzerland didn’t wait: AI Overviews have been live there since 26 March 2025, alongside Belgium, Germany, Italy, Spain, Austria, Ireland, Poland and Portugal. If you sell on both sides of that border, as many of my clients do, you already had one foot in the new world without necessarily noticing.
AI Mode deserves its own mention, because it does more than add a box above the results. It’s a full conversational tab where your question gets fanned out into sub-questions handled in parallel, and where the searcher keeps refining without ever going back to a list of links. Inside that mode, the only way your business exists at all is by being named in the text of the answer.
Pew Research Center measured what this does to behaviour, and the numbers are useful for calibrating your expectations. Across 68,879 Google searches run by 900 US adults in March 2025, of which 12,593 produced an AI summary, people clicked a traditional result on 8% of the visits where a summary appeared, against 15% of the visits without one. Links inside the summary itself were clicked on just 1% of visits, and the browsing session ended right there on 26% of pages with a summary, compared with 16% of pages showing only traditional results.
Read those figures as a shift in shape rather than a straight collapse, because the searches that still produce a click are the ones where the summary wasn’t enough. Somebody comparing two suppliers, checking whether you work in their canton, or trying to price a project won’t settle for four generated sentences, and that’s precisely the visitor who was going to contact you anyway. What the numbers do kill is the old habit of judging a page by raw sessions alone, since the same page can now be doing more work for your business while sending you fewer people than it did last year.
What are the real differences between GEO and SEO?
The core difference fits in one sentence: SEO goes after a position in a list, GEO goes after a citation inside a text. Everything else follows from that, including how you write your pages and how you judge whether the work paid off.
| Classic SEO | GEO | |
|---|---|---|
| What you’re after | A position in the list of links | A citation inside a written answer |
| What brings the visit | A click on your title | A mention of your name, sometimes with no click |
| The unit that counts | The whole page | A self-contained passage of a few sentences |
| How success is measured | Position, clicks, click-through rate | Impressions in AI features, brand mentions |
| Where authority comes from | The links pointing at you | The links plus what’s said about you elsewhere |
The most concrete consequence concerns the unit of work. In classic SEO you optimise a page for a keyword and hope to watch it climb, which still works and which I still do for every client. In GEO the machine doesn’t take your page, it takes a piece of your page, so a paragraph that fully answers one precise question and holds up once lifted out of its context is worth far more than a beautiful page whose answer is diluted across six screens.
The second consequence hits measurement, and it’s the one that unsettles most of the owners I work with. A company named in an AI Overview without receiving a click has still gained something, because its name appeared at the exact moment somebody was looking for what it sells, yet nothing in a classic dashboard tells that story. So you have to accept tracking two things side by side, the traffic on one hand and the presence inside answers on the other, without confusing one for the other.
What does Google actually recommend?
Google publishes its own guidance, and it’s a lot less dramatic than what the market is selling. Its AI optimization guide, last updated on 10 July 2026, repeats that SEO best practices remain relevant because its generative AI features on Search are rooted in its core ranking and quality systems.
The first condition is purely technical and it already rules out plenty of sites. To be shown as a supporting link in AI Overviews or AI Mode, a page must be indexed and eligible to appear in Google Search with a snippet, so if your site blocks crawling, loads badly or serves pages Google declines to index, no amount of clever writing will make up for it. That’s exactly the kind of blocker a proper website audit surfaces, before anyone starts talking about content.
The rest of the guidance rests on something Google says without hedging: creating content that people find original and genuinely useful will likely shape your presence in generative AI search more than any other suggestion on the list. After that come page experience, mobile readability, loading speed and whether a visitor can easily tell your main content apart from everything around it. For a local business or a shop, Google adds Business Profiles and product feeds, which overlaps with the local SEO work you may already be doing.
Put differently, the best preparation for AI answers is a clean, fast, honest site, which happens to be the best preparation for everything else too. If your fundamentals are shaky, the groundwork for ranking your website on Google is still the starting point, and GEO gets built on top of it.
Which GEO tactics are a waste of time?
Three heavily marketed tactics do nothing for Google Search, and Google says so itself in the documentation quoted above. The first is llms.txt, the text file meant to explain your site to language models, which Google’s Search team has been saying since 2025 that Search doesn’t use and isn’t planning to.
The official wording leaves little room for interpretation, since it states that you don’t need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search including its generative AI capabilities, because Google Search itself doesn’t use them. The file is harmless if you already have one, but it’ll never replace an hour spent making a real page better.
The second wasted tactic is chopping your content into artificial micro-blocks, usually pitched as essential for the AI to understand you. Google notes there’s no requirement to break your content into tiny pieces for AI to better understand it. The third is bolting on special structured data meant to flag your page to generative engines, when Google says structured data isn’t required for generative AI search and there’s no special schema.org markup you need to add. Structured data still earns its keep for classic rich results, which is reason enough to keep yours tidy, but it isn’t the master key you’re being sold.
Then there’s a genuinely risky one, which is buying brand mentions across sites and forums to artificially populate the sources these models read. Google explicitly files inauthentic mentions under the tactics you can ignore, and twenty years of watching search shortcuts play out suggests this sort of thing eventually turns on whoever bought it.
How do you measure visibility inside AI answers?
Since June 2026 Search Console has included a performance report dedicated to generative AI, which shows how many times links to your site were shown in a generative AI feature on Google Search. The report sits apart from the classic performance report, which saves you from mixing two different worlds in the same chart.
You can follow how those impressions move over time, see which pages earn the most and the fewest of them, and understand where they come from by country and by device. Google notes that it’s still working with site owners to figure out which data would help them most, and that it may add further metrics over time, so treat this report as a first window rather than a finished dashboard. A young or quiet site will see an almost empty report, simply because it hasn’t picked up enough impressions in those features yet.
For engines other than Google there’s no official equivalent, and I haven’t found an honest shortcut worth recommending. The method that works in my own week is to write down the twenty questions your customers genuinely ask before buying, put them one at a time into ChatGPT and Perplexity, then record who gets cited and on which phrasing. It’s manual, it costs an hour, and it teaches you more about where you actually stand than any tool sold on a monthly subscription.
Where do you start with a site that’s already live?
Start with the pages that already earn impressions, because a page Google knows and displays has far better odds of being reused than a brand new one nobody has heard of. The job is then to make each of those pages easier to quote, which comes down to four moves I repeat on every project.
- Put a complete two to four sentence answer at the top of the page, understandable without the rest.
- Phrase your subheadings as real questions, and answer each one in the sentence right below it.
- Replace every vague claim with a number, a date or a checkable example, and name the source.
- Add a frequently asked questions section using the exact wording your customers use.
What I like about those four moves is that they improve the page for a rushed reader exactly as much as for a machine, so you’re not betting on a trend. A page that answers in three lines and then expands is more pleasant to read, it keeps its visitors longer, and it lets a generative engine slice it cleanly.
The order matters more than people expect, so resist the urge to rewrite everything at once. Take your five best-performing pages, apply the four moves to those alone, and leave the rest of the site untouched for a full quarter so you can actually tell whether the change did anything. Rewriting forty pages in a fortnight feels productive and teaches you nothing, because you’ll have no way of separating the effect of your edits from the ordinary movement of a search engine that reshuffles itself every week anyway.
The rest of the work is what you already know, meaning titles and snippets written properly for classic results, a clear website structure that groups related pages together, internal linking that connects them so a machine can tell which ones carry authority on which subject, and regular refreshes of the content that ages. None of that is new, and that’s precisely the reassuring part.
What I tell my clients right now
My advice is to leave the strategy alone and change the thermometer instead. The businesses panicking since 22 July are hunting for some unheard-of technique that would put them back in front, when the only thing that genuinely moved is how their visibility shows up, with a growing share of appearances that produce no click and that none of their old metrics capture. The sensible move is to open the generative AI report in Search Console, write down where you stand this week, and watch it over three months before drawing any conclusion at all.
On content my position hasn’t shifted in years, and Google’s guidance keeps confirming it. A page that brings something the others don’t, written by someone who genuinely knows the subject and is willing to commit to specific numbers, gets cited by machines because it’s useful to humans. A page written to flatter an algorithm gets ignored by both, and it’ll get ignored faster as the models keep improving.
If your current site doesn’t meet those basics, meaning it’s slow, awkward to read on a phone or built on foundations nobody really controls, no layer of GEO will rescue it. That’s usually the real subject hiding behind the question of AI search, and it’s where I start when a client gets in touch. If you’d like to talk it through for your own site, getting in touch takes five minutes and I’ll tell you straight whether the problem is your content or your foundations.
