Google Confirms: AI Users Are No Longer Searching by Keywords – Your 2025 Stratégies Are Obsolete

I watch 15 sites a week. They all have the same problem.

I watch 15 sites a week. E-commerce, SaaS, SMBs. I’ve seen the same issue for 8 months now. SEO teams are still optimizing pages for « lightweight running shoes » or « noise-canceling headphones ». They spend hours on volume/difficulty matrices. Traffic stalls. Sometimes it drops. Why? The user has changed how they search. Google just confirmed it.

Last week, Google published a report on AI Mode usage in the US over one year. Shivani Mohan, VP Data Science at Google Search, revealed a key figure: the average query from AI Mode users is three times longer than a standard search. Three times. We’re moving from 3–4 words to 10, 15, sometimes 20 words. Goodbye to the keyword « best mattress ». Hello to a real question: « I suffer from lower back pain and my partner moves a lot at night—what mattress would suit us both without waking me? ». That changes the game.

The report also says conversational follow-up queries jumped over 40% on average every month since launch. Users don’t settle for one answer. They converse. They dig deeper. More than one search in six is multimodal—image, voice, video. You no longer type a keyword. You describe a context. So is your content ready for these long, conversational questions?

The numbers that bury keyword performance

Let’s look at the raw data. Shivani Mohan notes that the top 5 keywords typed in AI Mode are: Information, Identify, Find, Explain, Summarize. The first 5 words in queries are « what », « how », « I », « is », « can ». « I » lands in 3rd place. Your customers are no longer searching for a product. They’re sharing a personal situation: « I have flat feet and my knees hurt, can you help me find a running shoe that will not make it worse? »

For an e-commerce site, a product sheet that just lists specs no longer meets the demand. The typical query becomes « I have flat feet, I weigh 90 kg, my knees hurt after 3 km—what running shoe should I choose? ». If your page doesn’t contain this context, Google will skip it. Image searches have grown over 40% monthly since launch. People photograph their old worn-out shoe on the outer edge and ask « what type of gait do I have? ». Your content must speak this language. It’s the end of single-keyword optimization. It’s the beginning of entity-driven architecture.

And you might think: « I’ll just expand my long-tail keyword list. » Wrong move. Because the combinatorial explosion of contexts is infinite. Instead, you must provide a structure that Google can map to conversational questions on the fly—semantic clusters rich in entities. That’s what I did for a client, and the results speak for themselves.

Pour illustrer le passage du mot-clé à l’entité, voici comment nous avons structuré le cluster sémantique du site running évoqué dans l’étude de cas. Le pilier central couvre la question utilisateur, et les pages satellites détaillent chaque sous-problème.

+290% sessions in 7 months: inside a running site’s success

A running e-commerce client. 800 SKUs, 4,000 organic sessions per month. Their SEO was built on a matrix of 1,200 keywords. Each keyword had its category page or optimized article. The team published 4 articles weekly. Result: a year later, still 4,000 sessions. Frustrating.

When I dug into the site, I stopped content production. No new articles for 6 weeks. I rebuilt the architecture with semantic clusters around pain contexts and body types. We broke down the real intent: « I run, I hurt, why, what do I do ». We listed the entities: sensitive knees, IT band syndrome, overpronation, drop, cushioning, runner weight. Then we structured a cluster around the pillar « choosing running shoes for sensitive knees ».

7 months later: 15,600 organic sessions—that’s +290%. Traffic wasn’t coming through the old keywords anymore. It came from hundreds of long conversational queries like « running shoes for fragile knees after meniscus surgery » or « what shoe when you weigh 95 kg and have flat feet ». The site started appearing in AI Overviews. No ads. Just by speaking the same language as the user and Google AI Mode.

Le framework DOSE enseigné par Guillaume Attias à la BMO Academy structure concrètement la transition vers une stratégie centrée sur l’entité. Le schéma ci-dessous détaille chaque étape et ses livrables.

The DOSE framework: why your pages become audible again

This result didn’t fall from nowhere. I apply the DOSE framework, taught by Guillaume Attias at BMO Academy. DOSE stands for:

• Decompose the intent. Start from the human question, not the keyword. List all real scénarios (example: « I run twice a week, outer knee pain »).
• Organize the entities. Map out the concepts, symptoms, objects: outer knee, ITBS, stabilization, insoles, brands.
• Structure the clusters. Create a pillar page that covers the topic, linked to satellite pages that detail each sub-problem. Everything properly interlinked.
• Evaluate the signals. Stop tracking keyword rankings. Measure engagement on conversational pages, citations in AI Overviews, growth in long-tail queries.

For this client, the page « Running shoes for fragile knees » was the pillar. We wrote it answering 12 complete questions, with headers like « My inner knee hurts when I run—what shoe should I choose? ». Each answer integrated related entities (synovial plica syndrome, varus, 8 mm drop, Adidas Solarboost). Result: within 3 months, this page attracted 47% of the site’s conversational traffic.

DOSE is a concrete method. It makes product pages capable of dialoguing with Google’s AI.

Content that no longer speaks to keywords but to contexts

A concrete example. Before, the « running shoes for flat feet » sheet listed reviewer scores, colors, sizes. Today, the same page starts with: « If you have flat feet and your knees hurt after 5 km, you’re in the right place. We’ll first explain why your body type demands specific support, then show you the 3 models that will keep you from suffering. » The tone is empathetic, informative. The subheadings that follow answer « why do my knees hurt? », « what cushioning for flat feet? », « how do I test stability? ». Only then do the products appear.

We also created satellite articles: « Inner knee pain when I run: causes and solutions », « Overpronation: how to spot it in 3 minutes », « Test: are your running shoes right for your weight? ». Each article points to the pillar and vice versa. The whole thing is marked up with FAQ schema, HowTo, and Product. Google now understands this site is a goldmine for anything related to running pain. Citations in AI Mode have become routine.

The metric you’ve tracked for 10 years is now worthless

Many refuse to admit it: tracking keyword rankings barely matters anymore. For a running client, 80% of organic traffic came from queries they’d never tracked. Their legacy keywords, where they ranked first, were losing volume. People no longer type « running shoes pronation ». Yet their need was met by a conversational page, and sessions climbed. Focusing on rankings for 50 keywords means missing the real action.

I now measure three signals of conversational engagement: the number of clicks from queries over 8 words (in Search Console, once AI Mode filtering becomes visible); the number of pages cited in AI Overviews; time spent on contextualized pillar pages. These metrics tell you everything. Not your rank tracker.

Another example—a bedding client. Traffic stalled for 3 months while maintaining rankings. Analyzing long queries (« mattress for couples with a heavy sleeper and a light sleeper »), we discovered they were generating 12,000 sessions from long-tail search… but not measuring it. We adapted pages into DOSE clusters. Overall traffic jumped +65% in 4 months. The legacy keywords barely moved. The lever was invisible… until we looked at where Google was actually hitting.

What you can do this week

5 concrete actions, no fluff.

1. Identify your conversational questions. Google Search Console, filter for queries over 8 words that generated impressions in 2026. You’ll find a goldmine of ignored contexts.
2. List user scénarios. Take 3 personas and write their full questions, with their pain points. « I have X, I’m looking for Y for Z, under these conditions. »
3. Create one pillar page per need, not per keyword. One page for « help me choose a mattress when my back hurts and my partner moves » rather than 10 sheets on « mattress back pain », « firm mattress », etc. Structure it as a cluster.
4. Use entities. Link your content through proper schema.org markup (FAQPage, HowTo, Product) and think in entities: explicitly mention symptoms, contexts, solutions so Google associates them.
5. Measure conversational impact. Track clicks from AI Overviews (UTM parameters if possible), time spent on your new pillar pages, number of citations in AI responses. Relegate keyword rankings to the background.

I’m not selling you a method. I’m showing you the pages. They speak the same language as the AI Mode user. Today, they’re capturing thousands of sessions on queries you weren’t even tracking. That’s where it’s happening.

Stéphane Jambu

SEO & AI Engineer

I build growth systems / AI / Neuroscience | 650+ clients · 80 LinkedIn testimonials · 30 years of expertise · 15 years of systems running without me.

Follow on LinkedIn