Google alerts on commodity content in AI Search — 3 paths for e-commerce
Summarize this article with AI
The phone call that triggered everything
A client calls me on a Tuesday morning. €8,000 invested in SEO content. 6 months of writing. Result in Search Console: 3 pages ranked, zero clicks on generative snippets. He sells fitness accessories—a market drowning in identical descriptions, the same comparison tables, the same « best value for money » angles. His pages are clones. And AI agents don’t cite clones.
I asked him one thing: show me a product sheet that only you can write. Silence. The breakthrough came when we dissected the guide published by Google on May 15, 2026 and covered by Search Engine Land. This document, soberly titled « Optimizing for generative AI features, » makes a diagnosis I’ve been seeing in the field for months: commodity content doesn’t survive AI Overview or agents.
For this client, 89% of product pages held zero proprietary data. No internal measurements. No exclusive user feedback. No comparisons from the warehouse. Nothing screaming « this brand knows something others don’t. » It was interchangeable content. Commodity content.
Google’s official guide: an alarm nobody saw coming
The document makes no mystery of it: « Commodity content is information that offers no unique perspective or value — it’s what every site in your industry already says. » A simple definition. But for e-commerce, it’s a shock. Because 70% of product sheets I review every week repeat the exact manufacturer specs, the same « fast delivery » hooks, and the same generic reviews.
The guide distinguishes SEO, AEO (Answer Engine Optimization), and GEO (Generative Engine Optimization). It’s not another abstract paper: Google explains how its models select sources for generative snippets. And the verdict lands: if your page contains nothing more than the site next door, it won’t be cited. No citation, no visibility in AI Overview. No visibility, no traffic.
What struck me was the explicit mention of « AI agents. » Google isn’t just talking about its own AI module. It’s talking about an ecosystem where autonomous agents crawl the web to synthesize answers. In this world, a page’s value is no longer measured in direct clicks, but in citations. E-commerce merchants still churning out « checklist » content are locking themselves out.
The good news? That same guide opens very concrete paths. It calls for « differentiating » content, « proprietary data, » perspectives rooted in real expérience. That’s exactly what I apply with the DOSE framework, which Guillaume Attias teaches at BMO Academy. Define unique angles. Optimize around proprietary data foundations. Systematize differentiation. Then externalize production without losing the DNA. Google’s guide validates this approach.
Why 92% of e-commerce pages are invisible to LLMs
Of the 15 audits I ran in Q1 2026, one number chilled me: on average, 92% of product pages contained zero distinctive signal for an LLM. No measurement. No case study. No exclusive comparative test. The content was just there to tick the « page indexed » box.
Generative models work by attention. They scan billions of documents, spot patterns, and assign confidence scores to sources that inject novelty. When one product sheet says « ergonomic grip » and 147 other pages in the sector say the same thing, the algorithm has no reason to favor any one of them. Result: they all fall through the floor.
I have a client in camping gear. 4,000 organic sessions a month. A catalog of 800 references. Zero semantic architecture. Each sheet repeated the official tent description, weight, dimensions, waterproofing. We had a graveyard of identical pages. None contained the data this client actually possessed: feedback from their field testers, wind-resistance measurements taken in their warehouse, photos of assembly timed down to the minute. That material was sleeping in an Excel file. We stopped generic content production. We rebuilt each sheet around these proprietary data points.
The mechanism is simple: an LLM spots a source providing an original element (a precise measurement, a test protocol, a real comparison). It flags it as an authority node for that query. If you deposit no exclusive data, you disappear from the training base.
Path 1: Forge differentiated cluster architectures
The first answer to commodity content is architecture. Not one more word. Not a blog post « 10 tips. » A structure of semantic clusters that partitions your catalog into islands of exclusive meaning.
I’ve been forging these clusters since 2016. A cluster is a set of linked pages answering the same intent, with a pillar page aggregating unique data and satellite pages detailing it. The effect on LLMs is immediate: they see a coherent, dense cluster where each page adds a layer of proprietary information. It’s no longer a field of clone pages. It’s a verifiable knowledge well.
With the fitness client, we built 3 clusters: « elastic resistance measurements, » « calories burned by exercise with our internal tests, » « durability comparison over 6 months of use. » Each cluster is watered with data this client measured physically. Result: 14 months later, they go from 4,000 to 37,000 organic sessions. +820% traffic on satellite pages. And regular citations in Google AI Overview for 47 queries tied to band resistance.
The DOSE framework, as Guillaume Attias teaches at BMO Academy, serves as a compass here. You Define proprietary angles. You Optimize the cluster structure so each page signals something distinct. You Systematize variant creation with templates injecting raw data. And when the system runs, you Externalize production without losing exclusivity. It’s an architectural investment that pays every month without new ad spend.
Path 2: Inject proprietary data into every product sheet
Differentiation doesn’t come from creative genius. It comes from unlocking data you already own and nobody else has. A warehouse, customer support logs, a test bench, a customer panel. Turning raw data into citable content—that’s path number 2.
Take a hi-fi equipment seller I work with. 2,200 pages. 11 employees. Zero AI traffic. The owner opened his workshop doors to me. On a USB drive, hundreds of measurements of harmonic distortion, frequency response, listening tests run by his team. Years of proprietary data never published online. We created for each product an enriched technical sheet with these exclusive measurements, paired with a comparison table of 3 competitors tested under identical conditions. Six weeks of work.
Six months later, 14 of his pages became the primary source for Google AI Overview on queries like « tube amp harmonic distortion 0.1%. » Organic traffic jumped +470%. Agents cite his tables. Why? Because real comparative data was nowhere else to be found.
What Google’s guide hints at is that e-commerce’s future runs through publishing what I call « brand data »: numbers from your internal operations, test protocols, customer surveys treated statistically. It’s not marketing content. It’s documentary content. Citable for an LLM. Unreplicable for a competitor.
Path 3: Build a citable corpus for agents
AI agents are not web browsers. They don’t flip through pages. They extract assertions, data, citations. Their mechanics are those of an archivist who only retains sentences attributable to a reliable source. The third path is to turn your pages into a reservoir of exploitable citations.
An online wine merchant came to me. 6,000 references. Traditional « tasting notes » content copied from the vineyard. Invisible to AI. We launched a program of 47 interviews with sommeliers, each delivering a sharp opinion and a scored number after blind tasting. Each quote was inserted into the matching product page, with the expert’s name, affiliation, and tasting date. This instantly creates sourced assertions. Agents grab them. Today, 23% of this site’s traffic comes from citations in external generative responses.
The lesson: an LLM needs attributable sentences. A generic opinion « excellent value for money » is worthless. A sentence like « According to our sommelier Marc L., tested March 12, 2026 at 12°C, this syrah develops chalky tannins measured at 3.2 g/L of polyphenols » becomes data. It’s sourced. It’s unique. It’s dated. It’s citable.
I’m not saying produce thousands of interviews. I’m saying choose 10 to 15 assertions per cluster, present them in structured format, and have them appear in marked-up schema (ClaimReview type, though not the only option) that Google and other agents know how to read. The official guide stresses « content features that are easy for AI to parse. » That’s exactly it: authority sentences, ready to use.
What I observe from Southeast Asia
Since April 2025, I’ve been based in Southeast Asia. The time difference gives me a clear view of markets waking while Europe sleeps. One thing strikes me: the most agile e-commerce merchants aren’t the ones producing the most content. They’re the ones who own data nobody else has.
I watched a diving equipment seller with just 300 pages dominate AI citations on 120 queries across Asia. His secret? Test sheets for sealing integrity run in 3 partner aquatic centers, with time-stamped photos and measurements. Another in electronics turned his support logs into a machine generating failure statistics by model. Every published data point became an anchor for agents. It’s not a budget question. It’s an angle question.
Google’s guide buries the idea that « good enough » content cuts it. In 2026, interchangeable content is wasted time. The 3 paths I’ve shared (differentiated clusters, brand data, citable corpus) don’t require rebuilding your site from scratch. They ask you to take assets you already own and give them a shape machines can exploit.
What will your next product sheet have that the average doesn’t?
Your live audit in 15 minutes
I run 3 strategic pages from your site in front of you. We identify together where your content is interchangeable—and how to turn it into an asset citable by AIs. No jargon, no pitch.
Book a strategic call — 45 minFrequently Asked Questions
What is commodity content according to Google’s guide?
It’s content with no unique perspective, repeating what the entire industry already says. A product sheet identical to the supplier’s, a comparison containing only public information, an article paraphrasing other pages. Google is clear: this type of content won’t be picked by generative snippets and AI agents.
Why is e-commerce particularly exposed?
Because 70 to 90% of product pages share the same supplier specs. LLMs scan thousands of near-identical pages and find no reason to favor any one. Without exclusive data (internal tests, proprietary measurements, structured field feedback), an e-commerce site becomes invisible in AI Search.
Should I abandon classic product descriptions?
Not necessarily. Enrich them with signals only your business can produce: a comparison table from your own tests, a technical measurement you’ve made, a sourced interview excerpt. The base description stays useful for traditional SEO, but the differentiating element must jump out to an agent.
How do I know if my current content is commodity content?
Take 10 product pages, strip the brand name, and ask yourself if a competitor could publish them unchanged. If yes, it’s commodity content. A simple test: search for your most distinctive phrase in quotes on Google; if it shows up hundreds of times, you have a problem.
Are semantic clusters enough to get cited by AI?
They create the structure. But you must inject proprietary data into each cluster. A well-forged cluster with unique angles (measurements, tests, real comparisons) becomes a mine for LLMs. Architectural work is the foundation; exclusive material is the fuel. Together, they generate regular citations.
