AI Overviews: Lost clicks aren’t lower quality—here’s what changes for e-commerce
Summarize this article with AI
39.8%. And nobody saw what really changes
39.8%. That’s the organic click decline measured in a real-world test when Google displays AI Overviews.
The number was published at the end of March. Updated in April. Then expanded. It cycles through newsletters and LinkedIn feeds. People talk about it like a catastrophe.
Except the real problem isn’t the number—it’s the rush to interpret it.
A client calls me on a Tuesday morning. He runs an e-commerce site with 12,000 pages. Auto parts catalogs, 3,500 SKUs. Organic traffic hovering around 6,300 sessions per month.
He shows me the graph. The curve drops. 6,300 sessions. Then 5,100. Then 4,200.
« Stéphane, we’re losing our best visitors. The ones coming through precise product queries. Google keeps them with its summaries. The clicks left over—that’s just passing traffic, not qualified. I’m sure of it. »
I take time to analyze his Search Console data. I cross-check with GA4. I pull the segments. And I see something different.
Average order value on the remaining sessions is stable. Conversion rate even climbed from 1.4% to 1.8%. Less traffic, sure—but quality traffic.
So I tell him:
« Your problem isn’t click quality. Your semantic architecture is capturing fewer qualified clicks. »
And I show him the study that just came out.
The study that breaks intuition: lost clicks weren’t « bad »
The study is by Saharsh Agarwal and Ananya Sen. Academic research. Randomized. Real-world. Published, updated, with additional analysis. Search Engine Journal covered it.
The setup? One group sees AI Overviews, the other doesn’t. You compare the behavior of visitors who click.
The metrics are simple:
- how many visitors quickly return to the search results (back to search);
- how many leave the site in under 10 seconds without interaction;
- how long they spend on the page after clicking.
Result: no statistical difference.
When AI Overviews are absent, about 4 clicks out of 10 go back to the SERP. When they’re present, same proportion.
18% of visits end in under 10 seconds in both conditions. Time on site? Indistinguishable.
In short: the clicks « saved » by removing AI summaries aren’t better. They’re the same.
Liz Reid, VP of Search at Google, claimed AI Overviews reduce « bounce clicks, » low-value visits. She didn’t publish data. The study approaches the question from the other angle. If summaries were mostly absorbing low-value visits, extra clicks without them should pull metrics down. That’s not what we observe.
The study shows a reduction of 39.8% in organic clicks when AI Overviews are present. In both groups, 40% of clicks return to search. Visits under 10 seconds are 18% in both groups, and time on site shows no statistical difference.
For an e-commerce marketer, this conclusion is a gut punch.
Because it says: you’re not losing bad clicks. You’re losing clicks, period.
What « click quality » means when you’re selling online
A typical agency might’ve told you: « See, AI Overviews clean up your traffic. Focus on quality. »
Wrong. The study proves AI summaries don’t filter quality. They reduce traffic volume uniformly.
That’s a nuance to keep in mind.
On an e-commerce site, one fewer click isn’t just a dent in Analytics. It’s a product page unvisited. A category page unread. A navigation path that doesn’t happen. And a lost sale.
When I audit a catalog, I look at three things:
1. Pages still getting clicks despite AI Overviews.
2. Their current conversion rate.
3. Queries that used to drive traffic and no longer do.
Let me give you an example. Another client, outdoor gear retailer. 1,400 product pages. 320 commercial queries tracked. Before AI Overviews: 210 daily clicks on these queries. After: 128 clicks.
But here’s the thing: the pages still receiving clicks converted at 2.7%, up from 1.2% before. The remaining traffic was just as profitable, sometimes more.
The study doesn’t say remaining clicks are better. It says they’re not worse. That’s massive.
It means the priority isn’t chasing lost volume. It’s wringing every ounce of revenue from each visitor who still arrives.
Less noise. More signal.
In e-commerce SEO, I’ve always said: « We don’t hunt traffic. We hunt traffic that buys. »
AI Overviews don’t change that rule. They enforce it.
An auto parts site, 4 months, +42% traffic with zero ad spend
Back to the auto parts site I mentioned at the start.
Traffic was down. But the remaining visits converted well. The diagnosis took an hour.
The problem? Too many pages optimized for generic queries like « part [brand] ». Precisely the ones where AI Overviews give the answer directly. The AI summary shows a parts list, a photo, sometimes a comparison table. The user doesn’t need to click.
But on very specific queries— »Honda HRX 537 lawnmower head gasket »—the AI summary doesn’t fully answer. It doesn’t show price, availability, customer reviews. A click is still needed.
So we pivoted the strategy.
We mapped transactional queries not covered by AI Overviews (using semantic cocoons). We strengthened conversion signals on those pages: price schema, availability, customer reviews, technical FAQ.
We also stripped filler content from product pages—those long intro texts nobody reads—to concentrate semantic weight on differentiating attributes.
Results after 4 months:
• Organic traffic on target queries: +42%.
• Average conversion rate of restructured pages: 1.4% to 2.7%.
• Average order value stable, but order volume up 31%.
Zero extra ad spend. Just a semantic overhaul.
What worked? Stop shouting into the void on queries where Google already answers. Invest effort where AI can’t say it all.
Semantic cocoon: the structure that makes your pages essential despite AI
In e-commerce work, I start with one principle: a page should exist only if it answers an intent AI doesn’t fully cover.
Sounds extreme. But that’s what a semantic cocoon is.
A cocoon is pages linked by thematic logic. Each covers one precise angle. Each points to the next. Together, they form a mesh that signals to Google: « Here, we answer completely, even what AI can’t summarize. »
Example: on a motorcycle gear site, I built a cocoon around « modular helmets ». It includes:
- a category page with technical comparison,
- product pages with advanced schema,
- buyer question pages (safety standards, noise, maintenance),
- a buying guide based on real use cases.
Each page answers a sub-intent that AI Overviews can’t condense into a paragraph: noise thresholds, certification standards by country, structured customer reviews.
Result? The site captures traffic on peripheral queries, hyper-qualified. Where competitors wait for a click on « modular helmet reviews, » we capture « quiet modular helmet for plus sizes. »
AI Overviews don’t diminish SEO value. They shift the playing field. Winners will hunt clicks at the finest granularity.
That’s what I call building systems that run without me.
A well-architected cocoon doesn’t depend on one page. It creates semantic authority across an entire topic. Even if AI steals the click on the main query, you capture longer, more motivated clicks.
The $8,000 mistake: why chasing volume became a trap
Another client invested $8,000 in a content plan targeting « high-volume long-tail queries. »
The agency promised 50 articles monthly. 200 pages in 4 months. Target: +50% traffic.
Result? Traffic rose 23%. Conversion rate dropped from 2.1% to 0.9%.
Why? Those pages attracted visitors asking questions, not looking to buy. When Google rolled out AI Overviews, many of those informational pieces lost clicks. The few remaining visits were bounces.
The problem was « volume first » approach, not just AI.
In e-commerce, volume doesn’t pay anymore. What pays is transactional intent captured at the right moment.
Don’t abandon informational content—structure it to lead to purchase. Each page needs an exit ramp to a product, a category, a listing. And each page must answer a question AI doesn’t fully cover (or covers poorly).
The real $8,000 mistake? Creating pages that can’t survive one paragraph of AI summary.
Now I evaluate every content project with one rule:
« If Google’s AI summary answers 80% of the intent in 3 lines, this page isn’t worth building. »
It changes everything. You produce less. You produce better. Each page built has real odds of capturing a click that converts.
AI Overviews are a noise filter for e-commerce SEO, not a threat. Good news for those who play for precision.
What you must do now: 4 levers protecting your traffic
You manage an e-commerce site. You see traffic eroding. You don’t know where to start.
Here’s what I do with my clients.
1. Identify queries still generating clicks.
In Search Console, filter queries with impressions > 100 and click-through rate above 2% despite suspected AI Overviews presence (verify manually on SERPs). These queries are your foundation.
2. Cross-check with GA4.
Export these queries and match them to landing pages. Look at conversion rate, time spent, actions. Which pages convert? Reinforce them. Which pages survive without converting? Revisit their value proposition.
3. Build pages for non-summarizable transactional sub-intents.
AI Overviews struggle with detailed comparisons, compatibility guides, hyper-specific usage questions. That’s where you build your cocoon.
4. Test AI impact on your key pages.
Take your 20 most profitable pages. Search the query on Google. Check if an AI Overview appears and what it says. If it fully answers, your page risks losing clicks. Adjust it to deliver non-summarizable value (testimonials, visual comparisons, real-time availability).
These steps don’t need a huge budget. They need method.
And crucially, they rest on one fact: remaining clicks are worth just as much as clicks before.
Your mission isn’t mourning the past. It’s ensuring every visitor through the door leaves with a cart.
Volume may drop further. But if your conversion rate doubles, you win.
And if you don’t know which page to start with, ask yourself:
Do I know what fraction of my clicks land on a page that doesn’t match the real visitor intent because an AI Overview already said the main thing?
Audit the real quality of your traffic against AI Overviews
I examine your most exposed pages to reveal which remaining clicks are profitable and how to multiply them. No agency fluff. Just field diagnosis with your data.
Book a strategic call — 45 minFrequently Asked Questions
What are AI Overviews?
Google shows AI summaries at the top of results. They answer your question directly. The AI pulls info from web pages. No need to click.
Why does the study say lost clicks aren’t lower quality?
Performance metrics—bounce, time spent, back-to-search—stay the same whether the visitor saw an AI Overview or not. Clicks lost are worth as much as clicks kept.
How do I measure AI Overviews’ impact on my e-commerce site?
I compare clicks and click-through rates on my pages before and after AI summaries appear. I use Search Console and GA4. I manually test SERPs for key queries.
Should I optimize pages to appear in AI Overviews?
Yes, but not if it steals the click. I optimize for queries where AI doesn’t cover the full purchase intent, by adding details it can’t summarize.
What content resists AI Overviews best?
Detailed comparisons, compatibility guides, interactive pages, highly specific buyer questions. Anything that doesn’t fit in one paragraph.
