Shocking data: AI Overviews expose negative reviews without user intent — what to do in e-commerce?

The silent nightmare of AI Overviews

A client calls me on a Monday morning.
He sells high-end running equipment.
His organic traffic is stable. His Google rankings haven’t moved.
But his conversion rate has collapsed.

We check everything: site speed, checkout flow, ad campaigns.
Nothing.

So we dig.

I launch ChatGPT with a generic query: « What’s the best treadmill for heavy-duty use? »

And there, shock.

The AI throws out a detailed comparison. It cites three competing brands, then… mentions his.
Not for its qualities.
For a negative review from 2023 found on Reddit: « Delivery took 11 days and customer service never replied. »

The worst part? The user hadn’t searched for the brand. He was just looking for a product.
The AI made the connection all by itself.
And that connection repeated on Google AI Overviews, on Bing Copilot, on Perplexity.

That’s the new face of e-reputation.

Data compiled by Search Engine Journal in Q1 2026 is unambiguous: AI Overviews extract and expose negative reviews even in the absence of brand-focused search intent. According to the study, four signals govern this automatic surfacing: recency and volume of complaints, their specificity (naming a precise feature), the authority of the platform hosting the review (Reddit, Trustpilot, major forums), and the recurrence of the same grievance across multiple sources.

When a review checks all four boxes, it « rises » in AI responses, sometimes for a query about a competitor.
True reputation sniper.

I now systematically scan this phenomenon across my e-commerce clients. And I find it in 8 out of 10 cases.
That’s why I built a method to stop it.

Why your brand appears in a response that has nothing to do with it

Conversational search engines no longer rank pages.
They compare reputation signals the way we once compared backlinks.

The Search Engine Journal study details the mechanism, and it’s brutally logical.

First signal: recency and volume. A negative review posted 2 years ago, but regularly reposted and commented on, stays active. A single inflammatory Reddit comment can generate 47 reactions in 24 hours. The AI sees this engagement spike as a fresh marker. The content rises.

Second signal: specificity. « Bad service » triggers nothing. « The seat padding collapses after 3 weeks » is caviar for AI. This level of detail names a component, a feature, a timeframe. The statistical model sees it as a verifiable fact, therefore relevant to include in a comparison.

Third signal: platform authority. Reddit, Trustpilot, Amazon Reviews, specialized forums are training sources for LLMs. Their discussion thread structure, their voting systems, their longevity carry weight. A complaint on a DIY forum from 2004 can resurface if it cites a recurring defect.

Fourth signal: cross-source recurrence. If the same grievance — say « battery doesn’t last 2 hours » — appears on Reddit, on a Google review, and on Trustpilot, the AI consolidates it into consensus. And negative consensus becomes a paragraph in the AI response.

I verified this mechanism on a camera equipment store. Three identical complaints about lens mount quality, expressed across three different platforms within 5 weeks. Result: Google’s AI Overview for « hybrid reflex for video » displayed in position zero a statement: « Some users report a fragile mount on the XYZ model. »

No link. No star rating. Just that sentence, isolated, at the top of the SERP.

You see why this is hot.

These signals aren’t a bug. They’re intrinsic to how generative AI works. And that changes everything for your SEO.

What it costs an e-commerce business

Let’s talk numbers.

The treadmill client lost €1,800 in monthly revenue in 3 weeks.
His conversion rate dropped from 3.2% to 2.75%. A 14% decline.
Without a single page losing rank on standard Google.

Why? Because the searcher types « best treadmill », reads the AI Overview at the top of the page, sees the criticism, and doesn’t even click the brand’s link.
He bails.

On an average cart of €67, 14% fewer conversions means 27 lost sales each month.

And this site isn’t an isolated case. I observe the same phenomenon on a nutrition supplement brand: a 2024 Reddit thread mentioning « metallic taste and stomach issues » was cited by ChatGPT, Gemini, and Copilot within 6 days.
Result: -9% click-through rate from SERPs, and slow steady erosion of organic revenue.

The impact is more insidious than a Google penalty. With a penalty, you see it in Search Console. Here, everything looks normal.
But your qualified visitors are reading things you don’t know about.

In 2025, a Fast Company study cited by SEJ already reported cases of AI misrepresenting brand statements.
In 2026, the phenomenon generalizes.
The worst part? 63% of online shoppers trust AI summaries (according to a recent SparkToro poll), often without checking the source.

This silent cost is the new urgency in e-commerce.
The question is no longer « Does my SEO work? » but « What are users seeing in the AI response before they even click? »

The anti-negative-signal audit: map before you act

Don’t shoot blind.

The first step is auditing your negative footprint in AI ecosystems.
I proceed in three phases.

Phase 1: Run 15 to 20 comparison queries in your sector, never typing your brand name.
Example: « best 40L hiking pack », « comfortable hiking pack for weak back », « hiking pack with hydration bladder and breathable back panel ».
Use ChatGPT, Gemini, Copilot, and Google AI Overviews (via US VPN if needed).
Note every mention of your brand, plus tone and source platform cited.

Phase 2: Identify the trigger signals.
For each negative review cited, check all four boxes: recency + volume? Specificity? Platform authority? Cross-source recurrence?
You’ll get a clear matrix.

Phase 3: Prioritize.
A review checking all 4 boxes is absolute emergency. 3 boxes, high priority. 2 boxes, watch. 1 box, tackle if you have resources.

On the running client, I found 7 distinct negative mentions.
5 came from Reddit, 1 from Trustpilot, 1 from a sports forum.
3 checked all 4 boxes.
I flagged them as P0.

This audit takes two hours.
But without it, you’re flying blind.
It’s like optimizing a product page without looking at keywords: counterproductive.

I’ve delivered over 1,300 semantic cocoons since 2016.
And I can tell you that AI reputation audit is now as critical as internal linking audit.
The two complément each other anyway: a good cocoon amplifies your strengths, AI audit plugs your leaks.

Delete, respond, bury: the 3-level strategy

Once you have the map, you act.
The method boils down to three verbs.

1. Delete what’s illegitimate.
A fake review, a defamatory or obsolete comment can be flagged on the originating platform. Reddit, Trustpilot, Google Reviews all have procedures.
Be surgical: identify the CGU violation.
Never ask for deletion just because a review is negative. You’d lose all credibility.

I helped a childcare client get a Trustpilot review removed that accused a product of « causing chemical burn ». The review was clearly false (no support tickets, no photos). Removed in 18 days.
Meanwhile, the AI Overview took 9 more days to stop citing it. But it vanished.

2. Respond to legitimate reviews.
A calm, factual, dated response changes everything. The AI scans the discussion thread. A constructive response shows you fix problems.
« Hello X, we identified the delivery issue and modified our process as of March 12. Here’s a €15 voucher. »
This conversation becomes a positive signal the AI can cite too.

On the camera client, a forum review denounced mount fragility. The manufacturer replied describing the new reinforced version and offering a swap. Result: the AI added « The manufacturer has since corrected the defect and offers replacement. »
You retake control.

3. Bury persistent negative signals under a layer of positive content.
This is the most powerful step. Because it makes you master of the corpus where the AI draws from.
I forge entire semantic architectures around the brand: usage guides, sourced comparisons, transcribed video testimonials, structured FAQs, technical articles, etc.
The more precise and specific your content, the more the AI privileges it for answers.

A negative review buried among 27 positive and detailed pieces of content loses most of its weight.
Over time, it doesn’t even get pulled.
The signal dilutes.

This triptych — delete, respond, bury — works every time.
But it must be activated simultaneously on all four signals identified by the SEJ study to turn off the tap.

Build a content layer that AIs love to cite

The « bury » phase deserves its own chapter.
Because that’s where SEO reclaims its true nobility.

I don’t sell content for Google. I build information systems that run without me.
That’s what I do with semantic cocoons.
A cocoon is a set of pages linked together by a tight thematic structure, each page targeting a precise query and reinforcing the linking of the others.

For e-reputation, I adapt the principle: I create a reputation cocoon that surrounds the brand with trustworthy content.

Concretely:

• Complete guides around each product family, with hard data, diagrams, step-by-step tutorials.
• Objective comparisons where your product is shown alongside competitors, citing lab tests, verified reviews, measurements.
• Structured FAQ pages with QAPage schema markup, answering all technical questions detected via AlsoAsked and Google suggestions.
• Customer testimonials transcribed as text, accompanied by precise data (« I shaved 12 minutes off my half-marathon in 3 months »).

This type of content is ultra-specific, sourced, dated, regularly updated.
It checks every quality box that AIs seek.

I deployed this cocoon for a sports nutrition brand in January 2026.
In 7 weeks, across 10 test queries like « best whey for muscle gain », the AI cited 0 negative reviews and 8 pieces of the brand’s content (testimonial, blog post, internal study).
Before, they were at 4 negative citations and 0 positive.

Organic traffic didn’t explode. But conversion rate climbed from 2.3% to 3.8%.
Because people were reading positives before clicking.

This result isn’t magic. It’s mechanical: if you feed the corpus with structured, precise, redundant (never duplicate) content, the AI captures it.
And your own pages become the preferred source for the answer.

I use the DOSE framework, which I learned at BMO Academy with Guillaume Attias, to design these architectures. But the key is that each content block is built for human AND machine. No filler, no paraphrasing.
Real substance.

And that’s what no low-cost agency gives you.

Control the narrative in 2026: the new e-commerce SEO

SEO has changed nature.

We’re no longer just fighting for a blue link.
We’re fighting for the sentence the AI writes about you before the searcher even sees your site.

SEJ data confirms it: AI Overviews have become a new reputation channel. Free channel. Automated channel. Uncontrolled channel.
Except for those who get ahead of it.

I still hear e-commerce owners say: « I do SEO to rank first on Google, everything else is not my problem. »

Wrong.

Today, ranking first isn’t enough if the AI snippet right above it says your customer service is slow.
Companies ignoring this shift will watch their revenue erode without understanding why.

The good news? This battlefield is still 90% virgin.
Few brands understand the four-signal mechanics.
Even fewer have deployed a reputation cocoon.

You have a 12- to 18-month window to gain advantage.
After that, it’ll be too late. The corpus will be saturated.

So my advice: tonight, run 15 ChatGPT queries with your market keywords.
See what’s being said about your brand.
And if you freak out a little, that’s good. It means you’re lucid.

Next, audit the 4 signals. Delete the indefensible. Respond to the legitimate. Build the content layer that’ll bury everything.

I’m not selling you the method.
I’m showing you the results.

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