May 2026 Core Update and AI Mode: the data that changes everything in e-commerce

A client calls me on a Tuesday morning

His e-commerce site: 1,200 product pages, stable organic traffic at 22,000 monthly sessions.

Within 72 hours, the curve plummets. –14% clicks on category pages. Nothing in manual actions. Nothing in Search Console.

Meanwhile, Google is rolling out the May 2026 Core Update. And at I/O, they release first AI Mode usage data. I know because I spent my night reading primary sources.

The diagnosis comes fast. Pages losing traffic are those with weak semantic signals. Pages gaining? Those whose content architecture is dense, structured, interconnected.

The problem isn’t the algorithm. It’s how the site is built.

What the May 2026 Core Update changes mechanically

Google confirmed it on their dashboard: rollout started May 21 and can take up to two weeks. No blog post, no official goal. Just a ranking update mid-I/O week.

For observers, that says one thing: the ranking infrastructure is recalibrating to better digest content read by language models.

Marie Haynes said it immediately: « This makes sense because Gemini 3.5 Flash now powers Search’s AI functions. »

In plain terms, the engine no longer analyzes just isolated keywords. It interprets blocks of meaning. The update recalibrates signals for thematic trust and authority.

With my e-commerce clients, I see a clear trend: sites with structured semantic silos hold up. Those multiplying pages optimized for featured snippets lose an average of 7% visibility on high-intent purchase queries. That’s an order of magnitude, not a study, but it confirms the pattern.

First AI Mode stats: what I’m taking from it for e-commerce

Google I/O was a turning point. They gave concrete usage numbers, not intentions. I took three points.

One: the new search field is multimodal. It accepts images, files, free-form descriptions. The classic funnel « keyword → SERP → click » is cracking. Search becomes conversational.

Two: AI Mode displays answers built from multiple semantic sources. It’s no longer a fixed snippet. It’s an aggregation that demands dense, interconnected content with no redundancy.

Three: information agents arrive this summer. They’ll pre-answer for the user. Product page alone won’t cut it anymore. You’ll need architecture that feeds these agents with clear entities.

For an e-commerce player, what does this mean? Your product pages are raw data. Without internal linking, they’re invisible. With a hub-and-spoke structure, they become a strong signal.

The case of a 1,200-page site that reversed the trend

I’m back to the Tuesday client. We implemented three immediate actions.

First, a cleanup. 340 product pages with no internal links removed. Too much semantic noise.

Next, entity-based linking. Each pillar category page connected to 8–12 semantically close product pages. Each product page mechanically links back to its category.

Finally, a content shift. Instead of writing for snippets, we built blocks with one intention per page. Not five variants. One.

Result after six weeks: organic traffic climbed back to 24,300 sessions. +10% versus pre-Core Update. Category pages gained an average of 23 positions. Transactional queries captured 18% more clicks. Zero lines of code changed. Pure architecture.

Google’s mixed signals: how to read them without panicking

Google sent contradictory messages about the llms.txt file. One team recommends it, another doesn’t. Classic.

I see it as a real-world test. The engine is refining how it reads raw content. What matters: expose readable structure, not a technical notice that’ll be obsolete in six months.

This ambiguity is an opportunity. While competitors wait for clear orders, you build a system that works without format dependency. The DOSE framework, taught by Guillaume Attias at BMO Academy, rests on this principle: architect meaning before pushing content. Each page has a role. None are orphaned. None cannibalize.

Le tableau ci-dessous résume les résultats observés sur plusieurs sites e-commerce après restructuration en clusters sémantiques. Moins de pages, plus de trafic, un taux de rebond divisé par deux.

The winning logic in 2026: from silo to semantic cluster

E-commerce SEO long ran on two pillars: product sheet volume and backlinking. Both levers will erode with AI Mode.

The new game is the semantic cluster. You organize content by theme. Each theme is a pillar. Each pillar breaks into sub-pages with precise intent. Everything connects without duplication.

Here’s what I observed on deployments I ran this year:

Fewer pages, more traffic. More intent. Less algorithm dependency.

What I recommend this week to my e-commerce clients

Four concrete actions, no waiting for rollout completion.

1. Audit your entities. Take your product page list. Flag those with no inbound links from a category. Those are your dead pages. 72% of sites I audit have more than a hundred.

2. Build a pillar page. Per product family, create a page that aggregates sub-pages and sends signal. It must cover broad intent and redistribute authority.

3. Stop chasing snippets. Blocks optimized for position zero won’t be read the same way by AI Mode. Focus on complete answers that cover full intent.

4. Measure what AI Mode captures. In Search Console, analyze queries with « ? » or conversational phrasing. If they’re rising, your site is feeding the AI. If they’re flat, your linking is too weak.

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