Google-Agent: The Pivot That Buries 20 Years of SEO
Summarize this article with AI
What exactly did Google announce?
March 25, 2026. Marie Haynes publishes. I open her newsletter. I rate her analysis 10/10.
Her title: Why Google’s New Google-Agent is the Biggest Mindset Shift in SEO History. The announcement barely made a ripple in France. Yet it’s redrawing the commerce web.
Google rolled out a new user-agent. Technical name: Google-Agent. When an agent using Google infrastructure (Project Mariner, for example) visits your site, it identifiés itself with this tag.
It’s not just another crawler.
It’s a machine that comes to act. Fill out a quote form. Add a product to cart. Compare your prices with a competitor. Negotiate with your back-office.
At the same time, Google is pushing five protocols worth knowing:
- MCP (Model Context Protocol): secure agent access to your back-end data. Already adopted by Anthropic and OpenAI.
- A2A (Agent-to-Agent): two bots talking directly to each other. My agent talks to your agent. No human in between.
- UCP (Universal Commerce Protocol): the machine buys your product straight from the SERP. Without ever visiting your site.
- A2UI (Agent-to-User Interface): the agent composes its own interface. Your CSS might not figure in it anymore.
- AG-UI (Agent User Interaction): the middleware layer distributing AI responses in real-time.
Liz Reid, Google’s Search director: « There’s probably a world in which many agents are talking to each other. »
Nick Fox, her right hand: « Search becomes AI Search, and the Gemini app becomes your personal assistant. »
The e-commerce executive who thought ChatGPT was his biggest threat is behind the curve. The real shift is called Google-Agent. It arrived this month—not in two years.
That speed is what makes the subject urgent for anyone who still wants to influence purchase decisions in 2027.
Why it’s more radical than Panda, Penguin and <a href= »https://www.hi-commerce.fr/glossaire/#bert » class= »hc-gloss-link » title= »Definition: BERT »>BERT</a> combined
I started in SEO in 2007.
Since then I’ve seen Caffeine (2010). Panda (2011). Penguin (2012). Hummingbird (2013). RankBrain (2015). BERT (2019). MUM (2021). SGE (2023). AI Overviews (2024).
Nine updates. One logic.
Each time, Google kept serving pages to humans. It just got smarter at ranking the ones matching intent.
Google-Agent breaks that contract.
Google is no longer a traffic hub. Google is becoming an actor that executes.
The distinction sounds like nothing. It changes everything.
Concrete example, observed with a luxury client three weeks ago. A user asks Gemini: « Find me a rose gold bracelet for my wife, €600 budget, delivery before Sunday, engraving possible. »
Old world: Gemini returns ten links. User clicks. Compares. Hesitates. Ends up with a known competitor.
Google-Agent world: Gemini visits ten sites in parallel via WebMCP. Fills out quote forms. Fetches delivery times live via stock API. Filters on engraving. Returns three ready-to-buy products in one tap.
The click vanishes. The decision flows through the agent.
Marie Haynes puts it bluntly: « From 1998 to today, we created content. We gave it to Google to train the AI. In return, we got human traffic and ad revenue. That partnership, in its traditional form, no longer exists. »
What that means for an e-commerce executive:
- The product page optimized for clicks becomes a narrative API consumed by an agent.
- schema.org (Product, Offer, FAQPage) shifts from SEO bonus to commercial survival condition.
- Your back-office must be able to talk to an agent: real-time stock, dynamic pricing, slot availability.
- Social proof, testimonials, third-party reviews: the agent reads and weighs them in its verdict.
Those who built their SEO on « top 10 best X » listicles stuffed with affiliate links have 18 months, maximum, before the agent routes around them. The window shrinks with each announcement.
For 20 years, SEO rewarded expertise in ranking. 2026 SEO rewards your ability to make your site readable by a machine that acts.
It’s not the same discipline anymore.
Tactical consequences for e-commerce: the what, the how
Let’s get concrete. Across the 650+ e-commerce clients I’ve supported since 2016, five consequences stand out. I’m already prioritizing them with those calling me in April 2026.
1. The end of « top 10 best X » pages
Listicles stuffed with affiliate links lived on clicks. An agent has zero reason to click a list. It wants the answer, not the detour.
Comparison pages that survive: those delivering an argued recommendation, owned, with explicit criteria and structured data. Pages piling on affiliate content without expertise will vanish to the agent.
2. The product page becomes machine-readable
An agent consumes JSON-LD before reading your HTML.
Product, Offer, AggregateRating, Review, FAQPage: anything not marked up doesn’t exist for it.
What I’m seeing with my clients this year: sheets with complete schema are cited 3 to 5 times more often in Gemini and ChatGPT than sheets default-Yoasted. The order of magnitude I measure with Meteoria.
3. Back-office must speak MCP—or get bypassed
If your CMS can’t expose an MCP-compatible API (stock, price, slots, order status), an agent will visit your competitor who can.
WooCommerce, Shopify, PrestaShop are already working on connectors. In-house or closed back-offices have a timing problem.
For a tech team, an MCP connector prototypes in 10 to 15 days on a standard back-office. For others, start with agent-ready content while editors catch up.
4. UCP: the transaction happens in the SERP, not on your site
Universal Commerce Protocol, plainly: Google lets a machine buy your product straight from the SERP.
Your margin for personalization, up-sell, cross-sell, email capture? It flows through the agent.
So you must capture value upstream, on the product page itself. In the schema. In the enriched description. In structured social proof.
Funnel optimization logic gives way to self-sufficient product sheet logic.
5. Citable content becomes your new PageRank
An agent doesn’t rank. It cites. It weights. It crosses sources.
Content it cites shares three traits:
- A clear thesis in the first 150 words.
- Precise, sourced numbers.
- Structure that enables extraction (H2 as questions, lists, tables).
Exactly what I apply to this article. That’s why it will get cited more readily than an 800-word slab without markup.
One more thing. Most executives tell me « we’ll see in 6 months. »
This week, Google launched the user-agent. This week, e-commerce editors are publishing their first MCP connectors. This week, agents are already crawling your competitors.
Six months is what you need to deploy 30 to 50 semantic silos making a site agent-friendly. If you start today, you’re in the wagon. If you wait until October, you’re chasing.
DOSE and neuroscience: where does user dopamine go?
Since 2020, I’ve applied the DOSE framework to every site I support—Dopamine, Oxytocin, Serotonin, Endorphins. That’s Guillaume Attias (BMO Academy) who taught me. It’s the lens I use to audit any conversion funnel or page title.
The Google-Agent pivot doesn’t change the hormones. It shifts their trigger point.
Dopamine—anticipation
In classic search, it fires when the user types a query and scans the 10 results. The click = the hit.
In the Google-Agent world, the answer arrives instantly. Dopamine no longer runs through the click. It runs through the quality of the generated response.
The user no longer feels the anticipation of choosing. They feel the immediate satisfaction of a recommendation. Whoever appears in that recommendation captures all the attention.
Oxytocin—trust
An agent recommending a product transfers its own authority to the brand it cites.
If Gemini says « my recommendation: model X from brand Y, » the user grants Y a trust level no traditional ad reaches.
The agent becomes your best sales rep. Or your highest barrier to entry—depending which side you’re on.
Serotonin—status
Brands appearing in AI responses gain perceived status.
I’m already seeing this with my clients: a product cited in ChatGPT sees its on-site conversion rate rise—even when traffic comes from another channel.
The AI citation reconfigures brand perception. Like a TV mention 15 years ago.
Endorphins—relief
Maybe the deepest shift.
The mental effort of comparing 10 sites, 3 reviews, 5 specs—vanishes. The user feels relief at having an answer. And associates that relief with the agent giving it.
The brand fades to the background. Unless it made itself indispensable in the corpus the agent consults.
What this means for your content
- Write to trigger the instant response, not to hold the user for 4 minutes.
- Put the thesis, the number, the recommendation in the first 150 words. The agent extracts that zone first.
- Build sections answering a clear question. One H2 = one intent. One paragraph = one answer.
- Add tangible proof: exact numbers, named testimonials, measured cases. The agent weighs authority, not volume.
The good news: this matches what already converts humans. Writing for the agent is writing better for the human who remains.
No other marketing discipline can claim such alignment between both audiences.
