How does Google search work in 2026 with AI Overviews and AI search?
AI Overviews and AI Mode have been active in France since July 22, 2026: how they work, the role of SEO, citation criteria, measurement, and an action plan.
On July 22, 2026, Google activated AI Overviews and AI Mode in France, two years after their launch in the United States. For some searches, a response written by Gemini now appears above the links, and a new tab allows users to query Google like an assistant. Companies that previously focused primarily on their rankings must now also monitor whether their pages are being featured in these responses. We are working on this topic in our GEO agency, alongside SEO.
Here is what has changed, how Google selects the pages it uses, and what we recommend to our clients.
Key takeaways
- AI Overviews and AI Mode have been available in France since July 22, 2026.
- They rely on Google's index: a page that is poorly indexed or poorly ranked is unlikely to appear there.
- Google states that no specific technical optimization is required to appear in an AI Overview.
- According to Semrush, AI Overviews cite branded sources in 87% of cases, but only name a brand in 20.7% of responses.
- Our priorities: updating existing content, answering questions clearly, publishing clean data, and obtaining mentions on other sites.
What has changed on Google in France in 2026?
Google launched both features on the same day. They were already running in over 200 countries. In France, the rollout had been blocked by the dispute over neighboring rights with press publishers, for which Google was fined 500 million euros in 2021, and then 250 million in 2024.
AI Overviews
TheAI Overviews are summaries generated by Gemini, displayed at the top of certain search results pages. They answer the user's question and link to a few sources. They do not appear for all searches, but rather for those requiring an explanation or a comparison.
AI Mode
TheAI Mode is located in a separate tab, next to the standard results. Users ask their questions via text, voice, or photo, and then follow up with additional questions in the same conversation. The responses are longer than those in an AI Overview and contain more links.
How does Google choose the pages included in its AI responses?
Based on its index
AI Overviews and AI Mode draw from theGoogle index, which is the same one used to rank standard search results. If a page is not indexed, or does not rank for any queries related to the question, it is very unlikely to be used.
No dedicated markup
According to Google's documentation, there are no additional technical requirements or specific optimizations needed to appear in an AI Overview. If a service provider tries to sell you "AI Overviews markup," ask them what it is based on: no code can guarantee a citation.
By prioritizing clear answers
In the tests used to write our GEO Playbook, Google's models more frequently pulled from pages that provided a direct answer direct answer to the question, with short sections and information that is easy to extract. A page that takes three paragraphs to get to the point will lose out to a clearer one.
By citing sources without always naming brands
Semrush analyzed responses generated by several engines. AI Overviews cited brand sources in 87% of cases, but only named a brand in 20.7% of the responses. Your page can serve as a source for an AI Overview without your name appearing in the text the user reads.
Is traditional SEO still useful in 2026?
Yes. Gemini, AI Overviews, and AI Mode rely on Google's index, and therefore on traditional SEO : indexing, speed, and content that ranks.
Expectations for a page are evolving, however. It must still entice users to click, but it must also be easy to summarize and cite, and come from a brand that search engines clearly identify. Both approaches are presented on our GEO agency and SEO agency.
What should you do to be visible on Google in 2026?
Revise your existing pages before publishing new ones
Our primary advice is to work on your existing content. It is less expensive than publishing dozens of new pages, and the results appear faster. For every important page, check three things: it answers a specific question within the first few lines, each section covers only one point, and all figures are dated.
Organize each page so it is easy to summarize
- section headings written as your customers' questions;
- a two- or three-sentence answer under each heading;
- lists and tables for comparisons and steps;
- a FAQ at the end of the page, with short answers.
Add structured data
Google does not require them for AI Overviews. The structured data nevertheless tell search engines what the page contains: organization, service, article, FAQ, reviews. We have listed the tags to prioritize in our article on structured data to prioritize for AI.
Publish figures that only you have
Models need figures to provide answers, and they often lack them. A study based on your proprietary data (pricing, lead times, client results, trends observed in your sector) has a good chance of being cited, with your brand mentioned as the source.
Get mentions on other sites
An AI is more likely to recommend a brand that other sites talk about. Specialized press articles, rankings, comparisons, professional directories: these third-party sources count in the answers, even when they do not contain a link to your site.
Control the technical side
Fast pages that are readable without complex JavaScript and accessible to bots. These rules already applied to SEO. They carry more weight with models that quickly abandon a page that displays poorly.
How to measure your visibility in Google's AI answers?
Rankings and organic traffic no longer tell the whole story. A page might lose clicks because an AI Overview already answers the question, yet still be cited within that same Overview.
For our clients, we track a fixed list of questions that are critical to their business:
- theappearance rate in AI Overviews and AI Mode;
- citations (the page as a source) and mentions (the brand name in the text);
- the tone of the responses when the brand is cited;
- the diversity of the sites that mention it;
- its presence compared to that of its competitors.
Semrush AI, Omnia, or Promptwatch cover some of these indicators. To link this visibility to traffic and conversions, we describe our method in the article on proving that your content works on AI search engines.
What we do at Growth Room
We first tested GEO on our own site, then compiled the method into our GEO Playbook, which over 4,000 people have requested. We update it based on the tests we conduct every week, as models change their sources regularly.
On our site, we have:
- revised our existing articles to answer our prospects' questions more directly;
- published data from our campaigns, such as our budget and fee ranges for an Ads test, or the results of our study on 4 B2B accounts tracked over 10 months;
- tagged our pages and FAQs;
- tracked our presence on third-party sites, such as growth marketing agency rankings on Impli and Zeliq, or our profiles with client reviews on La Fabrique du Net and Sortlist.
We then measure our appearance rate across a panel of fixed questions on Google, ChatGPT, and Perplexity.
