The search bar is becoming an answer bar. You type a question, and an AI sums up three sites before you've clicked anything. At Google I/O last month, Google confirmed the direction: "AI Mode" and generated summaries move to the front. The question my clients ask has shifted. Before: "how do I rank on Google?" Now: "is my site even going to matter?"
Here's how I answer, concretely.
What actually changed
The click is no longer a given. When an AI summary sits at the top of the results, people click the links below it far less. The Pew Research Center measured this on tens of thousands of real queries: with an AI summary present, users click a link about 8% of the time, versus 15% without one. Ranking first no longer means getting visited.
The flip side is that the new currency isn't position — it's the citation. Showing up as a source inside the AI answer. And the visitors who do click through arrive with more context, so often with stronger intent.
That's scary when you live off traffic. But it changes very little about how I build a site. If anything, the opposite.
What I don't change
The fundamentals that make a site "citable" by an AI are exactly the ones that already made it good for Google — and for a human.
A fast site. An AI, like a crawler, reads a page that responds quickly more easily and more often. On the sites I build, interaction response time stays low because there's little JavaScript to execute — that's the whole point of going static.
A clean structure. One main heading per page, clear subheadings, paragraphs that answer a specific question. That's what an AI extracts most easily.
First-hand content. What Google calls E-E-A-T — experience, expertise, authoritativeness, trust. A generic AI can rephrase generalities all day. What it can't invent is a real account of a real project. That's exactly this blog's editorial line: no hollow listicles, lived experience.
What I do change
Where I adjust is machine readability. Three things, already in place on this blog:
- Structured data. Every article emits JSON-LD (
BlogPosting: author, date, language). It's the format crawlers and AIs read to understand who says what, and when. - Clean bilingual setup.
hreflangtags link the French and English version of each page. An AI serves the right language to the right reader without conflating the two. - An open feed. One RSS feed per language, stable and readable URLs. The easier a page is to index and follow, the better its odds of being picked up.
None of this is an "AI hack." These are web standards, laid down properly.
The trap to avoid
At I/O, Google also warned: manipulating or buying citations to appear in AI answers is the wrong road. With every shift in search, an industry of shortcuts pops up promising to trick the algorithm. It always ends the same way.
I prefer the other bet, slower and safer: a fast, honest, well-structured site that says true things. What an AI likes in 2026 is what a good reader liked in 2016.
In short
AI changes the storefront, not the foundations. If your site is slow, confusing or hollow, AI will just make it invisible faster. If it's fast, clear and grounded in something real, it stays a source — for a human and a machine alike.
Got a site to (re)build in that spirit? Let's talk.