What are AI Overviews?
When you ask Google a question, it no longer just shows ten blue links. For a growing number of queries, it writes its own answer in a few paragraphs, generated by its AI, with links to the sources it used. That is an AI Overview: Google answers in place of the websites, while relying on them.
The rollout is gradual, country by country and query type by query type. Informational queries ("how", "why", "what is the difference between") are served first; local and transactional queries keep more of the classic format. But the direction is clear and it will not reverse: the results page is becoming an answers page.
Who loses traffic, and who gains?
When the answer is displayed directly, a share of searchers no longer clicks. The first casualties are "quick answer" content: definitions, generic tutorials, articles written to catch a query without adding anything the others do not. That traffic evaporates.
On the other hand, two profiles do well:
- The cited sources. Being one of the two or three links shown in the AI answer means capturing a highly qualified click: the person has read the summary and wants to go deeper with you.
- Identifiable brands. When the AI recommends "hiring a specialist", the businesses the models know and can place (clear activity, geographic area, proof) are the ones that get named.
For a small business, the right reflex is therefore not to mourn the traffic lost on generic queries, but to win the source spot on the queries that concern its trade and its area.
How do you become the cited source?
AI engines pick their sources based on measurable criteria. Four levers, in order of impact:
- Extractable passages. The AI cites short, self-contained blocks: a 2-3 sentence answer under a clear heading, a list, a table. A page that buries the information in 20-line paragraphs is not citable. Structure every page around real questions, with the answer first.
- Facts, numbers, dates. "We respond within 48 hours in the Var area, starting at €90" is citable. "A responsive team at your service" is not. Models prefer to quote verifiable statements.
- A clear entity. Who you are, what you do, where: JSON-LD Organization structured data, a consistent about page, an up-to-date llms.txt. If AI tools cannot place you, they cannot recommend you.
- Crawler access. Many sites block Google-Extended or GPTBot without knowing it, often because of a plugin setting. Check your robots.txt: a block takes you out of the game, silently.
What our audit checks
The "AI Optimization (GEO)" category of our audit measures precisely these conditions: presence and validity of JSON-LD, llms.txt, extractable passages and lists, citable factual figures, AI crawler directives in the robots.txt, content depth. It is the category where even well-built sites hit a ceiling, because nobody ever told them what to do. The free diagnostic gives you your score in 30 seconds.
What comes next?
AI search does not replace SEO: it extends it. The fundamentals (a healthy site, solid content, authority) remain the foundation, and the citability criteria stack on top. The businesses that start now build a durable lead: models construct their knowledge of the web over months, not hours. The best time to become citable was a year ago. The second best time is this week.