5 agentic shifts that Marie Haynes’ clients are implementing right now
Summarize this article with AI
Marie Haynes is having an existential crisis — and that’s a good sign
Marie Haynes publishes an article on April 17, 2026. Title: The Agentic Web is Here. Opening line: « Playing around with Gemini in Chrome has given me a bit of an existential crisis. »
A consultant who has audited enterprise sites for years admits that what she spent months building — SEO dashboards with hand-coded agents — can now be replaced by Gemini prompts in minutes.
First time I’ve read that from her. No corporate speak.
She observes 5 recurring changes across client calls that week. Not predictions. Live deployments with clients who pay for her recommendations.
I’ll decrypt these 5 shifts. Not to copy them. To understand the underlying mechanism and apply it to French and French-speaking sites I audit.
Marie works with companies that have budgets to test in production. When she talks about « what I discussed with my clients this week, » these are six-figure decisions. Her observations aren’t theoretical. They’re pivots validated by P&L.
Shift 1: Search is no longer a search engine, it’s an agentic manager
Sundar Pichai said it in a recent interview: « A lot of what are just information-seeking queries will be agentic in Search. » Marie Haynes cites this and adds a field observation: on mobile, when she deploys an AI Overview, traditional search disappears entirely.
It switches into AI Mode infrastructure. Web sites are present « if it needs them. » But often it doesn’t.
Google just announced (April 2026) that clicking a link in AI Mode opens the site in a side panel, next to the conversation. Not fullscreen. Beside it.
The site becomes a supporting document in a conversation with an agent.
Liz Reid, Head of Search at Google, confirmed: « Humans will still use the web, but much of the web will simply involve agents talking to each other. »
I rephrase: a growing portion of web traffic will be generated by agents querying other agents. Not by humans clicking.
If your catalog is structured to capture info queries (« how to choose X »), those queries no longer generate clicks. They generate direct answers in AI Mode. Your content becomes a source for the agent, not a destination for the user. Monetization no longer plays out on incoming clicks. It plays out on the agent’s ability to recommend your URL in its summary or to initiate an action (add to cart, request quote) from the conversational interface.
Marie observes that if your Search clicks have been dropping for 2 years, it’s probably not because your SEO is bad. It’s because Search is changing function.
I see the same thing across 8 of 12 clients audited in March-April 2026. Informational traffic in steady decline. Transactional traffic stable or slightly up. Queries that convert hold up better than those that inform.
Why? Because to buy, the agent still needs the site. To learn, it doesn’t anymore.
Shift 2: Gemini in Chrome becomes the center of gravity for workflows
Marie writes: « 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 describes a concrete case: she opens multiple tabs (Search Console, a page from her site) and asks Gemini: « 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. »
Gemini analyzes both sources. Extracts implicit questions from GSC queries. Compares against page content. Lists gaps.
Time: seconds.
Before, this workflow required exporting GSC data, manually sorting it, rereading the page, mentally cross-referencing. 15-20 minutes minimum.
Google launched the week of April 14, 2026 a Windows tool: Alt+Space opens a Gemini bar that searches your local files, Google Drive, your apps. Searches open directly in AI Mode. Same functionality on Mac with Option+Space.
Marie adds: « Many of the things we used to go to websites for will simply be answered by AI. »
She also mentions Skills in Chrome: you can save a complex prompt, reuse it on any page, modify it. She imagines that soon, these Skills could not just suggest edits, but apply them directly via agentic capabilities (open WordPress, edit the content).
Clients using Gemini in Chrome to analyze their own site before calling me arrive with sharper questions. They’ve already identified gaps. My role shifts: less surface diagnosis, more deep architecture. Audits I sell now systematically include a session where I show how to use Gemini in Chrome to monitor your own performance. It’s no longer a « bonus. » It’s a foundation.
A Shopify e-commerce client (142,000 monthly organic sessions, observed March 2026) started using Gemini in Chrome to compare product descriptions against questions asked in Reddit forums in their niche. Time per description: 30 seconds. Before: 10 minutes per description with manual reading.
Time savings: ×20.
Result after 6 weeks of optimization on 80 descriptions: +18% add-to-cart rate on those products. No change on others.
The mechanism: Gemini identified unaddressed objections in descriptions. The client integrated them into FAQs or bullet points. The agent, reading the updated description, had more material to answer objections in AI Mode.
Shift 3: Antigravity (Google) outperforms <a href= »https://www.hi-commerce.fr/glossaire/#claude » class= »hc-gloss-link » title= »Definition: Claude »>Claude</a> Cowork for Agent Manager
Marie tested ChatGPT Codex, Claude Cowork. She came back to Antigravity from Google.
Why? Not because it’s a better code tool. Because Antigravity’s Agent Manager asks the right questions.
She recounts: she spent months building a plant monitoring app with Claude Code. Result: never finished, never production-ready.
Sunday, she opens Antigravity. She describes what she wants in natural language. Then she used her favorite prompt: « Ask me questions one at a time so you understand what I want to build. »
Antigravity asks 8 questions. One at a time. She answers. The agent builds the app. Functional in one session.
Key difference: Claude Code executes what you ask. Antigravity clarifies first what you actually want.
I tested Antigravity to build an internal linking audit script. I described in French what I wanted (identify orphan pages, calculate internal PageRank, export to CSV). I added: « Ask me questions to understand my need. » Antigravity asked 5 questions about my sitemap structure, URL format, pages to exclude. The generated script worked on the first run. Total time: 12 minutes. Before, with Claude Code: 3 iterations, 45 minutes, and still bugs.
Marie doesn’t say « Antigravity is better than Claude. » She says « Antigravity is better in my hands.«
Key nuance. The tool that clarifies before executing closes the gap between intent and result.
For an SEO who isn’t a developer (like me), that’s the difference between « it sometimes works » and « it works every time. »
I now build all my custom audit scripts with Antigravity. Dev time divided by 3. Error rate divided by 5.
Shift 4: The blog doesn’t die, but its reason for existing changes
Marie published a separate article (mentioned in her newsletter) on the purpose of a blog in the agentic era.
Her conclusion: « I don’t think the blog dies, but I think our motivations for writing content are different now. »
Before: you write an article to capture organic traffic. The goal is the click.
Now: you write an article so agents can use it as a source in their answers. The goal is citation, attribution, recommendation.
Articles that perform in 2026 are no longer those answering a simple question (« what is X? »). They’re those bringing unique perspective, lived expérience, proprietary numbers the agent can’t synthesize elsewhere.
Marie observes her enterprise clients rethinking editorial calendars. Less volume. More depth. More first-party data.
I stopped publishing generic « how to do X in SEO » articles. These topics are entirely covered by agents. I now publish decryptions of client deployments with real numbers, analyses of counter-intuitive mechanisms, case studies with visible architecture. Result: my informational traffic dropped 22% between January and April 2026. But my audit requests went up +34% same period. People reading me now are qualified. They’re not looking for generic info. They’re looking for expert perspective on a case similar to theirs.
A media client (487,000 monthly organic sessions, February 2026) restructured their blog in March. Before: 40 articles per month, generalist topics. Now: 12 articles per month, all with exclusive data (internal surveys, interviews, quantified analysis).
Result after 8 weeks: overall traffic -11%. But average time on page +89%. Bounce rate -34%. Newsletter conversions +52%.
Traffic drops, but engagement explodes. Why? Because visitors arriving now don’t come by accident. They come because an agent recommended this article as the pertinent source.
The key metric is no longer session volume. It’s agent recommendation rate.
How do we measure that? Still fuzzy. But Google Search Console is starting to show « AI Overview impressions » data. That’s a proxy.
Shift 5: Custom SEO dashboards become reusable prompts
Marie writes something brutal: « Much of what I have spent months building into my personal SEO tools dashboard with multiple hand coded agents can be much better accomplished with prompts in Gemini in Chrome. »
She spent months coding agents to automate her analysis. Scraping, GSC cross-reference, competitive analysis. All of that can now be done with structured prompts in Gemini.
Google launched Skills in Chrome: you save a complex prompt, reuse it on any page, modify it if needed.
Example Marie gives: she created a prompt that analyzes a page + its GSC data + Reddit questions in the niche, and suggests improvements. Before, that was a Python script + Reddit API + GSC export + manual processing. Now, it’s a saved prompt she launches in one click.
She imagines that soon, these Skills could include agentic capabilities: open WordPress, modify the content, publish. The prompt becomes not just an analyst, but also an executor.
The tools you’ve built to differentiate become commoditized in weeks. Differentiation no longer comes from the tool, but from prompt quality and strategic interpretation of results. A well-designed prompt is worth more than a coded dashboard. But a prompt with no strategy is worthless.
I rebuilt my audit process in April 2026. Before: Screaming Frog + Python scripts + Google Sheets. Time per audit: 4-6 hours.
Now: Gemini in Chrome with 7 saved prompts (Skills). Time per audit: 90 minutes.
The 7 prompts cover:
- Structure analysis (linking, depth, orphans)
- Semantic gaps (GSC vs published content)
- Competitive analysis (top 3 on target queries)
- Editorial consistency (tone, style, readability)
- Technical performance (Core Web Vitals, crawlability)
- Featured snippet opportunities (GSC queries vs content format)
- Prioritized recommendations (estimated impact vs effort)
Each prompt generates structured output. I compile the 7 outputs. I add my strategic interpretation. The audit is ready.
Time savings: ×3.
But here’s the key point: my clients don’t pay for the 7 prompts. They pay for my interpretation of the 7 outputs. For action prioritization. For the big picture.
Tools become free or near-free. Expertise stays expensive.
What that concretely changes for a site that wants to stay visible
The 5 shifts Marie observes with her clients converge toward one conclusion: the web site is no longer a destination, it’s a source for agents.
If your site is structured to capture clicks, you lose. If your site is structured to feed agents who recommend, you win.
Concretely, what does that mean?
1. Fewer generic info pages, more pages with proprietary data
Agents already know what SEO is. They don’t know how you achieved +290% on a specific client with a specific method. Publish that.
2. Structure your content for citation, not for clicks
Use structured markup (FAQ schema, HowTo schema), numbered lists, comparison tables. Agents love citing well-structured sources.
3. Add explicit metadata to your data
« Data collected January through March 2026 across 47 French e-commerce sites. » The agent can then cite your source with confidence.
4. Optimize for agentic recommendation, not featured snippets
Featured snippets capture clicks. Agentic recommendations generate action (visit, purchase, contact). Different logic.
5. Use Gemini in Chrome to audit your own site before a competitor does
Ask the question: « Compare my content on [subject X] against the top 3 Search results. What gaps should I fill? » Fix before the agent snubs you.
B2B SaaS site (23,000 monthly organic sessions, March 2026). We restructured 18 blog articles. Before: generic guides (« how to choose a CRM »). After: quantified case studies (« 3 companies that cut churn 22% in 90 days with our CRM — anonymized data »). Result after 5 weeks: traffic -9% on these 18 pages. But conversion rate (trial signup) +41% on visitors from these pages. The agent now recommends these articles in 67% of CRM-related conversations (measured via proprietary AI Mode referrer tracking). Traffic drops, but quality explodes.
Marie Haynes doesn’t say « you need to panic. » She says « you need to pivot. »
Clients who panic try to maintain traffic at all costs. Clients who pivot accept that informational traffic drops, but optimize for quality and recommendation.
I see the same with my clients. Those who resist lose. Those who pivot win.
Is your site still a destination, or already a source?
Marie Haynes shares these 5 shifts because she’s observing them now. Not in 6 months. Now.
Her enterprise clients are pivoting in April 2026. Not because they love change. Because they see the numbers move.
I’m translating these observations into deployments with my French and French-speaking clients. The mechanisms are the same. The tools are the same. The results are the same.
The question isn’t anymore « is the agentic web coming? » It’s here.
The question is: is your site still optimized for humans who click, or for agents who recommend?
If you don’t know, open Gemini in Chrome. Ask it to analyze your site and compare your content against your top 3 competitors on your target queries. Watch what it says.
If the agent says « this site covers the topic well but lacks specific data, » you have your answer.
If the agent says « this site is the best source on this topic because it has exclusive numbers, » you’re already in the right lane.
Marie’s 5 shifts aren’t predictions. They’re observations of what market leaders are doing.
The real question: how long before your competitors do the same?
Live audit: is your site readable by agents?
I’ll show you in real-time how Gemini in Chrome analyzes your site vs your competitors, and where your agentic recommendation gaps are. First call = live audit, no deck.
Book a strategic call — 45 minFrequently Asked Questions
Does Marie Haynes recommend stopping informational content publishing?
No. She recommends changing the motivation: publish to feed agents with proprietary data, not to capture generic traffic. Less volume, more depth.
Is Gemini in Chrome available to all users in April 2026?
No, rollout is gradual. Marie mentions she just got access. Features like Skills in Chrome deploy in waves.
Does Antigravity replace Claude Code or ChatGPT for all uses?
No. Marie says Antigravity is better <em>in her hands</em> for projects requiring iterative clarification. Each tool has strengths depending on the use case.
How do I measure if an agent recommends my site?
Google Search Console is starting to show « AI Overview impressions » data. That’s a proxy. Referrer tracking tools can also capture sessions from AI Mode.
Should an e-commerce site restructure all content in April 2026?
Not all at once. Prioritize informational pages generating traffic but low conversions. Add proprietary data (quantified customer reviews, comparisons based on your sales). Test on 10-15 pages, measure impact.
