Agentic web: the 5 shifts Marie Haynes is observing with her clients

A client calls me on Tuesday. His traffic has been dropping for 8 months.

He’s done everything right. Produced content. Improved internal links. Optimized tags. Zero penalties. Zero Search Console alerts.

The problem isn’t him.

It’s the nature of search that’s changing. Marie Haynes has been observing this with her clients for months—so have I. She just published a field report that lines up exactly with what I’m seeing in my deployments. Five structural mutations. Not projections for 2027. Active changes in April 2026.

I’m going to break them down here with what I’m observing on my end.

Shift 1: Search is no longer a search engine. It’s an agent manager.

According to Marie Haynes, the fundamental change comes down to one sentence: « Search is changing from a place where we passively look for information to an ‘agentic manager’ where we actually get things done. »

Sundar Pichai said it outright in an interview: « A lot of what are just information-seeking queries will be agentic in Search. »

Concrete translation: most informational queries no longer generate clicks. They’re resolved in AI Overview or AI Mode. On mobile, when Marie Haynes opens an AI Overview, traditional search disappears. It switches to a conversation with the AI. Web sites remain available. If needed.

Scale I’m observing with my e-commerce clients: between November 2025 and March 2026, category pages that were capturing informational traffic lost between 12% and 38% of organic sessions. Not because content degraded. Because Google answers directly.

Traffic doesn’t migrate to a competitor. It doesn’t migrate anywhere. It stays in Google’s interface.

I review 15 sites a week. All of them pulling traffic on « how to X » or « what is Y » queries are seeing the same decline. Transactional pages (purchase, booking, signup) hold up better. They require an action that AI can’t complete on its own.

Yet.

Shift 2: Gemini in Chrome transforms daily workflow

Marie Haynes writes: « I finally got access to Gemini directly within Chrome, and I cannot believe how much I’m using it to converse across my open tabs and Google Docs. It’s hard to imagine how I did work before without it! »

She can open multiple tabs. Call Gemini in the sidebar. Ask it to analyze Search Console AND the page simultaneously. Example prompt she gives: « Look at my GSC data for this page and also my page itself and tell me what questions my audience is asking about that I don’t cover well. »

Result: Gemini cross-references click data, actual queries, page content, and suggests what to add.

I’ve been testing this for three weeks. On one client site (professional training sector, 450 indexed pages), I applied this workflow to 12 strategic pages. Gemini identified 47 uncovered questions that were showing up in Search Console impressions. We integrated them. +18% impressions on those pages in 21 days. +9% clicks.

The gain isn’t spectacular. But the time invested is ridiculous: 35 minutes to handle 12 pages.

Google just added Skills in Chrome. You can save a prompt like Marie’s. Turn it into a reusable skill. Apply it to any page on your site. Marie projects that eventually, these Skills will allow not just suggesting edits, but applying them directly in WordPress via agentic capabilities (WebMCP or equivalent).

We’re almost there.

Google also launched a Windows shortcut: Alt+Space opens a Search bar that queries your local files, Google Drive, your apps. Results display in AI Mode. On Mac, it’s Option+Space. Marie can now launch a Gemini conversation on top of any task, continue it in Chrome, without breaking flow.

Consequence for us: people won’t necessarily open a site to find an answer. They’ll ask Gemini. The AI will search. Synthesize. Propose. The site stays a source. But it’s no longer the entry point.

Shift 3: Antigravity beats <a href= »https://www.hi-commerce.fr/glossaire/#claude » class= »hc-gloss-link » title= »Definition: Claude »>Claude</a> Code for complex agentic tasks

Marie Haynes tested Claude Code. ChatGPT Codex. Claude Cowork. She came back to Google’s Antigravity.

Direct quote: « In my hands, Antigravity is better. […] Antigravity is not a coding tool. I mean it is. But its power comes in using the Agent Manager. »

She spent months trying to build a plant monitoring app with Claude Code. Never finished. One Sunday, she opened Antigravity. She described what she wanted. Then she used this prompt: « Ask me questions one at a time so you understand what I want to build. »

Antigravity asked questions. Clarified the need. Built the app.

I don’t code plant apps. But I’ve used Antigravity since February to automate SEO audit tasks that took me 90 minutes per site. Example: cross Search Console data, server logs, page depth, and identify indexed URLs that never crawled by actual users. Before, I scripted that in Python. It worked. But each new client needed tweaks.

With Antigravity, I described the need once. The Agent Manager built the workflow. Now I give it files. It outputs the sheet. 8 minutes instead of 90.

Marie sees the same thing with her clients. Those adopting agentic tools (Antigravity, but also other orchestrators like LangChain for the more technical) automate entire SEO stack sections. Not to replace humans. To free time for strategic decisions.

One client of mine (B2B marketplace, 8,000 product sheets) deployed an agent that monitors logs, detects high-traffic pages with low conversion, and surfaces weekly alerts with suggested fixes. The agent doesn’t make edits. It says: « Page X has 1,200 sessions, 0.4% conversion, generic title tag, missing CTA above the fold. » The team validates. The agent generates code. The team deploys.

Processing time per page: 4 minutes. Before, it was 25 minutes plus prioritization meeting.

+12% conversions on treated pages in 6 weeks.

Shift 4: Informational clicks are collapsing—it’s not your SEO breaking

Marie Haynes says it plainly: « If you’ve seen your clicks from Search dropping over the last couple of years, it’s not likely that you’re doing things wrong in terms of SEO. Search is heading in a different direction. »

AI Overviews and AI Mode answer informational queries directly. On mobile, when Marie opens an AI Overview, traditional search disappears. The user switches to a conversation. Sites remain accessible. But they become optional.

Scale I’m observing with my clients between September 2025 and March 2026:

• Media/blog sites: -22% to -41% organic clicks on « guide », « tutorial », « definition » type articles
• E-commerce sites: -8% to -19% on category pages with editorial content at the top
• SaaS sites: -14% to -28% on « resources » / « learning center » pages

Transactional pages (product sheet, contact form, booking page) lose much less. Between -2% and -7% depending on sector. Why? Because AI can’t finalize the purchase. It can recommend. It can compare. But it can’t pull out a credit card.

Yet.

One of my clients (health sector, medical articles in plain language) saw informational traffic drop 34% between October 2025 and February 2026. We stopped producing purely informational content. We pivoted toward high-intent pages: « Book an appointment », « Compare treatments », « Download the full guide ». We cut production volume by 60%. We invested the budget in redesigning 18 transactional pages.

Result in 10 weeks: -29% total organic sessions. But +17% conversions. Cost per conversion dropped 23%.

Traffic is no longer the goal. Action is.

Marie sees the same shift with her clients. Those clinging to vanity metrics (sessions, page views) panic. Those focused on conversions adjust their architecture. They strengthen pages that generate action. They let AI handle the informational side.

Shift 5: Google opens links in a side panel—you stay in the ecosystem

Google just announced a new feature: when you click a link in AI Mode, the site opens in a panel next to the conversation. You never leave Google’s interface.

Marie Haynes notes: « Google has just announced that now when you click on a link in AI Mode it will open the website in a panel next to your AI Mode conversation. »

Direct consequence: the user can keep asking Gemini while browsing your site. They can ask: « Summarize this page for me. » « Compare this product with the one you showed me before. » « Find the price on this page. »

You no longer control the journey. Gemini becomes the copilot. You become the source.

I’ve been testing this for 12 days on desktop. I searched « best ergonomic office chair ». AI Mode gave me a synthesis. I clicked a link. The site opened on the right. Gemini stayed on the left. I asked: « Is this model suited for someone 6’1″? » Gemini scanned the page. It answered, citing the specs.

I never scrolled the product sheet.

For us, what does that mean? Two things:

1. Information architecture must be AI-readable. Not just readable by Google. By Gemini, which will extract, synthesize, rephrase. If your product specs are hidden in an image or poorly tagged table, Gemini won’t see them.
2. Your CTA must be visible immediately. If the user doesn’t scroll, your « Add to cart » button at page bottom doesn’t exist. It needs to be above the fold. Always.

Marie Haynes anticipates this will accelerate. Users will increasingly « visit » a site less. They’ll consult it via a panel while conversing with an agent. The site becomes a reference. Not a destination.

That doesn’t mean the site dies. It means it must be designed to be cited, not browsed.

Pages that work in this paradigm:

• Structured info (perfect schema.org tags)
• CTA visible without interaction
• Direct answer to one precise question (no 300-word intro filler)
• Actionable content (price, availability, specs, purchase link)

Pages that struggle:

• Long SEO intros stuffed with keywords
• Content spread across multiple sections
• CTA buried at page bottom
• Critical info in untagged images

I’m rebuilding architecture on 80% of sites I audit for this reason. We compress. We structure. We lift CTAs. We shift everything possible to schema.org.

Clients applying this see conversion rates climb even as traffic falls. Because visitors who arrive are more qualified. They come to act, not read.

What this changes for you right now

Marie Haynes closes her field report on a note that struck me: « The web is changing so quickly, and many of the things we’ve spent years building might soon be obliterated as AI advances. But there is also so much opportunity if we can figure out where to focus. »

Translation: stop producing informational content to rank. It doesn’t work anymore. Or it works less and less. You’ll get impressions. Maybe a few clicks. But AI will answer instead of your page 7 times out of 10.

Focus on pages that trigger action.

Scale observed with my clients who pivoted between November 2025 and March 2026:

• Content volume produced: -40% to -65%
• Budget reallocated toward transactional pages: +120% to +200%
• Organic conversions: +8% to +19% depending on sector
• Cost per conversion: -11% to -27%

One client (B2B SaaS, 140 employees) stopped publishing blog posts in January 2026. They’d been producing 12 articles per month since 2019. Cost them 4,500€/month (writers + editing + images). They cut it all. They invested the 4,500€ in redesigning 8 pillar pages: « Features », « Pricing », « Demo », « Competitor comparison X », « Case studies », « Integrations », « Security », « Support ».

Result after 11 weeks:

• Blog traffic: -51% (logical, no new content)
• Pillar page traffic: +23%
• Demo requests: +34%
• Trial conversions: +18%

The CEO told me: « We should’ve done this two years ago. »

Another adjustment I recommend systematically since January 2026: restructure e-commerce category pages. Before, we put 400 words of editorial content at the top. « Discover our selection of ergonomic office chairs to improve your work comfort… » It ranked. It doesn’t rank anymore. AI answers instead.

New model:

• Clear H1 (« Ergonomic office chairs »)
• Filters visible immediately (price, material, height, weight capacity)
• Top 3 products featured with direct CTA
• Editorial content at bottom, after products, only to feed the AI (not the user)

Test with one client (professional furniture, 320 SKUs): we restructured 14 category pages this way. Traffic: -9% on those pages. Conversions: +16%. Average cart value: +8%.

The traffic that remains is more qualified. People who click aren’t coming to read. They’re coming to buy.

Is your site still designed to be visited—or to be cited?

Marie Haynes observes five shifts with her clients. I see the same. Maybe you do too.

The agentic web doesn’t kill sites. It changes their function. You’re no longer a destination. You’re a source. A source that AI cites, synthesizes, rephrases, recommends.

If your architecture is still built to generate informational traffic, you’re going to suffer. If you pivot toward transactional pages, structured, with visible CTAs and data exploitable by agents, you’re going to win.

My clients who applied these adjustments between November 2025 and March 2026 all lost traffic. Between -8% and -34% depending on sector. But they all gained in conversions. Between +8% and +19%.

Volume drops. Quality rises.

The question is no longer « How do I rank for this query? » It’s « What action does this page trigger? »

If you can’t answer, the AI will. And the user will never come.

You have the five shifts. You have the scale. You have real cases. Now open your Search Console. Filter pages losing traffic. See if they’re informational or transactional.

If they’re informational, stop optimizing them. Invest elsewhere.

If they’re transactional, you have an architecture problem. Not a content one.

Which is it for you?

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