Google: use AI « in the best way possible » for AI Search – here’s how
Summarize this article with AI
A client calls me on a Tuesday morning. He invested 15,000 € in AI content.
4,500 catalog references. 2,200 organic sessions per month. A content team producing 12 articles per week with AI. And yet, traffic is stagnant. Worse, it’s declining.
The director has this heavy statement: « Stéphane, we bet everything on AI and Google ignores us. »
I check Search Console. 947 pages indexed. Three-quarters receive zero visits in 90 days. The semantic footprint is full of holes. No internal linking. No semantic cocoon backbone.
15,000 € gone.
Then, a few days later, I come across an official statement that reframes everything. Nikola Todorovic, Director of Software Engineering at Google Search, responds to Martin Splitt about adapting sites to new AI Search features. His key phrase, reported by Roger Montti of Search Engine Journal: « Clearly, this is something we can advise.«
He doesn’t say « produce at scale. » He says: use AI to analyze data, study competition, and improve your ability to create value.
It’s my client’s Tuesday. The following week, we stop everything. We redesign the architecture. We keep AI, but we change its role.
« There is no magic wand »: what Google actually expects from AI
In the exchange between Splitt and Todorovic, the opening question is: how can the ecosystem thrive with AI features in search?
The answer contains neither roadmap nor checklist. Todorovic admits there is « no magic wand« . But he insists on one point: site owners must continue ensuring their content delivers real value. And to achieve that, AI, used wisely, can amplify that value.
« My guiding principle is that site owners need to continue making sure their products, their sites are providing value. AI can help with data analysis and competition research, and clearly, this is something we can advise. »
I observe with my clients that the word « value » is often misused. People think layout, originality, tone. But for Google, in the context of AI Search, value is first and foremost semantic.
A page has value if it answers the search intent exactly, without dilution. And this is worked out upstream through search space mapping, not downstream through flashy writing.
Key takeaway: Google isn’t asking you to write with AI. It’s suggesting you use AI to understand where and why to write.
The mistake I see 3 times a week: producing without mapping
Let’s recap. 12 articles per week. 4 writers + 2 AI tools. The flow is steady. But pages land on isolated islands, no linking, no cocoon. Thematic coverage is as thin as onion skin.
The real cost isn’t the 15,000 €. It’s the dilution effect. Each new shallow page reduces the average semantic density of the site. Google receives a fuzzy signal. The ranking of strong pages suffers.
What my successful clients do:
- They never write before they have an intent map.
- They use AI for analysis tasks: keyword clustering, named entity extraction from SERPs, identifying gaps against competitors.
- They build semantic cocoons before any production. Each future page already has its place and mission.
In the case of my client with a 4,500-product catalog, we first extracted the 3,200 relevant long-tail queries with AI. Then we grouped them into 127 thematic silos. Only then did we order content.
Immediate result: we deleted 214 pointless pages. Crawl budget breathed. First improvements came 6 weeks after deployment.
My DOSE framework: how I use AI to map value
Since 2016, I’ve applied the DOSE framework taught by Guillaume Attias within BMO Academy. It structures my cocoon approach. And today, AI fits into each step.
Diagnostics: AI scans Search Console, server logs, and SERPs to identify zombie pages, semantic duplicates, and coverage gaps. When I audit a site, I no longer wait 3 days for an inventory. I get it in 4 hours.
Orientation: AI crosses competitive data (cocoon depth, pages per silo, internal linking) with the site’s historical performance. This gives us direction: strengthen this silo, prune that one, create a bridgehead on a blank thematic area.
Structure: This is the heart of the cocoon. AI generates hierarchies from semantic clusters. It proposes inbound and outbound links between parent and child pages. I verify human relevance, but the machine does 80% of the modeling work.
Execution: Here, AI can help with writing, but only within ultra-precise briefs. Each brief states the exact intent (informational, transactional, navigational) and entities to cover. AI drafts a framework. The writer enriches, fact-checks, densifies.
The result isn’t a content factory. It’s a forge for useful pages.
Why this works: Our brain searches for patterns. So does Google. A semantic cocoon creates a relevance pattern that guides robot exploration and topic comprehension.
47,000 sessions, 11 months, zero increase in ad spend
I’ll walk through the trajectory of the client mentioned earlier. The audit dates to September 2023. The site is an e-commerce specializing in professional tools.
| Period | Organic sessions (avg. monthly) | Actions taken |
|---|---|---|
| Sept 2023 | 2,200 | DOSE audit, identified 214 pages to remove |
| Oct – Dec 2023 | 2,500 | Internal linking restructuring |
| Jan – Mar 2024 | 5,800 | Deployment of 47 initial cocoons |
| Apr – Jun 2024 | 13,400 | Enriched content in priority silos |
| Jul – Aug 2024 | 47,000 | 1,300 active cocoons, refined linking |
Ad budget didn’t move. Content production cost dropped 37% because we only produced what had a place in the cocoon. No more orphaned content.
Another striking figure: the number of pages generating at least 10 clicks per month went from 89 to 412. Traffic density spread across more entry points. Google started trusting the structure.
And AI in all this? It mapped the 3,200 queries, modeled the 127 silos, and produced the briefs. Not a single page published without being integrated into the architecture. That’s exactly what Todorovic describes: AI serving value.
What you can apply tomorrow, without breaking everything
No need to overhaul everything. I recommend 4 immediate actions to integrate AI « in the best way possible » into your AI Search strategy.
1. Run an express semantic audit with AI. Export your Search Console queries over 16 months. Use an AI tool to group them into intent clusters. You’ll see the gaps, overcoverage, and oddities. With one of my clients, a single pass revealed that 40% of their content competed with their own product pages.
2. Build a minimal viable cocoon. Pick a silo with 15 to 20 queries. Map a parent page and 5 to 7 child pages. Write them following AI-generated briefs. Link pages together with varied anchors. Publish. Measure after 6 weeks.
3. Train your team on analysis AI, not just writing AI. Writing AI without briefs is a trap. Analysis AI (clustering, entity extraction, NLP) is a compass. My clients who train their writers on analysis rather than production gain an average of 27% more pages ranking in the top 10.
4. Schedule a monthly « AI x SEO » check-in. Analyze deployed cocoon performance with AI tools that detect anomalies. Adjust briefs. It’s a virtuous cycle, not a one-time sprint.
« But Stéphane, wouldn’t Google risk penalizing AI? »
No. And that’s the whole subtlety. Google never said AI was forbidden. It said automatically generated content without added value violated its guidelines. Nuance.
I’ve checked with 17 clients. In every case, when AI is used to structure rather than to replace editorial thinking, performance climbs. The algorithm doesn’t detect AI. It detects uselessness.
Google’s advice through Todorovic clarifies this gray zone. AI is a lever to invest where value is strongest. This aligns perfectly with what my cocoons create: zones of semantic density where authority accumulates.
So yes, use AI. But like an architect uses 3D modeling software, not like a brick-making machine.
A 45-minute audit to map your next cocoon
I take your site and show you live the semantic gaps and pages diluting your authority. At the end of the call, you leave with the structure of a priority cocoon — no commitment, no sales pitch.
Book a strategic call — 45 minFrequently Asked Questions
Does Google really recommend using AI to produce content?
No. The software engineering director proposes using AI to analyze data, study competition, and improve the ability to create value. Pure content production is only a consequence when the framework is in place.
Can AI penalize my site if I use it wrong?
Massive uncontrolled use often results in content without added value, which Google may treat as spam. The issue isn’t AI, but semantic density of your pages. With a well-built cocoon, you have nothing to fear.
What should I start with to apply AI in SEO without making mistakes?
Start with an audit of your semantic coverage. Use an AI tool to cluster your Search Console queries. Identify a high-potential silo. Build a mini-cocoon with AI-generated briefs. Measure results in 6 weeks.
How long does it take to see the effects of cocoon-based architecture?
First effects on restructured pages are visible between 4 and 8 weeks. A complete transformation takes 6 to 11 months, as described in the case study. Patience is rewarded with steady growth and no ad dependency.
Should I delete old content when switching to cocoons?
Not necessarily. First eliminate pages with zero traffic and no inbound links, then redirect useful poorly-ranked pages. AI helps identify these pages so you keep only what serves semantic linking.
