Your product pages are no longer read by Google Search alone. They now feed AI-generated recommendations, and the main channel isn’t your site: it’s your Merchant Center feed.
The shift is measurable. In November 2024, AI Overviews appeared on 2.1% of transactional queries. By early 2026, an analysis of 20.9 million shopping keywords put them at 14% of shopping queries. Ecommerce used to be spared. It no longer is.
I’m Florian Zorgnotti, an SEO and GEO consultant. What follows covers what actually changes in your product data, which fields carry weight, and why a French merchant can act today without being able to measure natively.
Does ChatGPT recommend your brand?
Measure your presence and find out which brands get cited instead of you. No credit card.
Key takeaways
- Volume has shifted: from 2.1% of transactional queries in late 2024 to 14% of shopping queries in early 2026.
- The feed beats the page: Merchant Center serves as the canonical source for product recommendations across AI Mode, AI Overviews and the Gemini app.
- You can act but not measure: Conversational Attributes is rolling out globally, while the AI performance report stays limited to the United States.
- No clicks in Google’s report: share of voice is measured in impressions, with no clicks or conversions. It’s a leading indicator.
Do AI Overviews really affect shopping queries?
Yes, and the change is recent. Ahrefs measured 2.1% of transactional queries triggering an AI Overview in November 2024. A Visibility Labs analysis covering 20.9 million shopping keywords puts that rate at 14% in early 026, close to a sixfold rise in a matter of months.
The mix of affected queries moved the same way. According to a Semrush analysis of more than 10 million keywords, commercial queries grew from 8% to 18% of AI Overview appearances, and transactional queries from 2% to 14%.
One methodological caveat belongs here. These measurements are American, and the French rollout of AI Overviews only dates from 22 July 2026. The magnitudes give you a trajectory, not a snapshot of the French market. Nobody holds French data on this yet, and claiming otherwise would be dishonest.
What matters for you isn’t the exact figure but the direction. Queries like “best [product] for [use case]”, “alternative to [brand]” or “cheap [product]” are progressively shifting toward a synthesised answer where a few brands get named and the rest are absent.
Why your Merchant Center feed matters more than your product page
Because generative engines prefer verified, structured product data over a page crawl. Merchant Center supplies exactly that: a pipeline where the merchant submits price, availability, identifiers and images directly, in a form the machine doesn’t need to interpret.
That pipeline reaches beyond Google. Industry analyses in 2026 describe Merchant Center as the canonical product data source for AI Mode and AI Overviews, with indirect use by other generative shopping surfaces. Your feed no longer serves Google Shopping alone.
In my view, this is the reversal French merchants absorb most slowly. We keep treating the feed as a technical byproduct of the catalogue, handled by a developer or a plugin, while polishing product pages for classic SEO. The priority order has flipped: on shopping queries captured by an AI, an incomplete feed makes you invisible even with excellent product pages.
The fields that carry weight
- Product title, which carries the semantic match with a question phrased in natural language.
- GTIN and identifiers, enabling reconciliation across sources and trust in the data.
- Availability and price, whose accuracy determines whether an engine will recommend a product it can’t verify as in stock.
- Structured attributes such as colour, material or style, which Google now flags when missing.
Conversational Attributes: the field that changes things
Google announced a feature called Conversational Attributes at Google Marketing Live in May 2026. It lets you add attributes and descriptions designed for conversational language directly in Merchant Center, which Google’s AI systems use to match products to questions asked in natural language.
The logic is easy to grasp. Your feed says “trail running shoe, 8mm drop, 280g”. The shopper asks “which shoe for mountain running when my knees hurt”. Between the two sits a layer of use-case vocabulary that conversational attributes fill in.
One point matters for you: this feature is rolling out globally. A French merchant can use it right now, with nothing to wait for.
Why you can act without being able to measure
A structural asymmetry penalises French merchants, and it deserves naming plainly. Google shipped two building blocks in 2026: a write block and a read block. They don’t share the same geographic coverage.
| Building block | Function | Availability |
|---|---|---|
| Conversational Attributes | Enrich product data for conversational search | Global rollout |
| AI performance insights | Measure product share of voice across AI surfaces | United States pilot only |
The AI performance insights report appeared on 13 July 2026 in a limited set of US accounts, under Analytics then Products. Google mentions an upcoming expansion to Australia, Canada, India and New Zealand, with no firm date. France isn’t named.
Two limits of this report deserve attention before you expect too much from it. It covers organic AI traffic only, excluding paid ads. More importantly, it exposes no click, conversion or revenue metric: share of voice there is calculated in impressions. It’s a leading visibility indicator, not a performance metric you can build an ROI on.
The practical consequence is direct. A French merchant who wants to know whether an AI recommends their products has to run an external check, since native measurement isn’t available to them.
How do you measure product visibility without Google’s report?
Through repeated, aggregated checks on real shopping prompts. The method differs from classic SEO tracking because the unit of analysis is no longer the keyword but the purchase question.
Build a basket of questions the way your customers actually phrase them: “what’s the best [category] for [use case]”, “alternatives to [competitor]”, “[product] under [budget]”. Then record, for each one, which brands get named and which don’t.
Three cautions are worth repeating. First, an isolated check is worthless: generative answers vary from run to run, so only share of voice aggregated across many checks produces a signal. Second, separate citation, mention and recommendation: your site can be the cited source of an answer that recommends a competitor. Third, track each engine separately, since their sources barely overlap.
That’s the logic I built Cockpyt AI on with Laurent Séjourné: weekly measurement across ChatGPT, Gemini and Perplexity, share of voice against competitors, a Google Analytics 4 connection linking visibility to real traffic, and a monthly GEO audit producing a prioritised action plan. Flat price at 29 euros a month billed annually, 14-day trial with no credit card.
The action plan for your product data
Here is the sequence I apply with ecommerce clients, in descending order of return.
- Audit feed completeness. Identifiers, availability, price, images, structured attributes. An incomplete feed caps everything else, whatever the quality of your pages.
- Add conversational attributes. The feature is globally available: it’s the only lever that directly changes the data Google matches against shopping questions.
- Rewrite product titles in use-case vocabulary. Supplement technical nomenclature with the terms your customers actually use.
- Configure your competitor set. The day native measurement reaches France, an empty competitor set will make the benchmark unusable.
- Establish an external baseline now. Record your share of voice on strategic shopping questions, so you hold history when native data arrives.
- Work on reviews and third-party mentions. On comparison queries, third-party sources outweigh your own product messaging.
The mistakes that keep you invisible
- Polishing the page and neglecting the feed. On a shopping query captured by an AI, the feed is what qualifies you.
- Stacking Product markup hoping for citations. Markup helps parsing, but no public study demonstrates a direct effect on AI citation rates.
- Leaving stock-outs unflagged. Inaccurate availability data is grounds for exclusion, not a detail.
- Waiting for Google’s report. It has no announced date for France, and it won’t give you clicks or conversions anyway.
- Measuring once. Without aggregation across many checks, you’re photographing noise.
FAQ
Do AI Overviews appear on product queries in France?
The French rollout dates from 22 July 2026 and remains progressive, so no reliable measurement of the trigger rate on French shopping queries exists yet. American data puts that rate around 14% of shopping queries in early 2026, against 2.1% of transactional queries in late 2024. Treat those figures as a trajectory, not as French data.
Do you need a Merchant Center account to be recommended by an AI?
For product recommendations on Google surfaces, the Merchant Center feed is the canonical data source. A merchant absent from that pipeline forfeits the main channel through which engines retrieve verified price, availability and identifiers. It doesn’t exclude you from editorial comparison answers, which lean more on third-party sources.
What are Conversational Attributes?
It’s a feature Google announced in May 2026 allowing you to add attributes and descriptions suited to conversational language inside Merchant Center. Google’s AI systems use them to match your products to questions asked in natural language. It’s rolling out globally, which makes it usable right now by a French merchant.
Is the Merchant Center AI performance report available in France?
No. It appeared in mid-July 2026 in a limited number of US accounts. Google announces expansion to Australia, Canada, India and New Zealand in the coming months, with no firm date and no mention of France. A French merchant therefore has to rely on external checks to measure product visibility in AI surfaces.
Does that report show clicks and sales generated by AI?
No. The documentation mentions no click, conversion or revenue metric. All four documented metrics rest on impressions, including a share of voice calculated as your AI impressions divided by the total across you and your competitors. It’s a visibility instrument, not an attribution one.
Is Schema.org Product markup enough to get cited?
No. Markup helps machines parse your pages and serves your classic rich results, but no public study demonstrates that it raises your citation rate in AI answers. On product queries, feed completeness and accuracy weigh more than extra tags on the page.
Should you optimise differently for ChatGPT and AI Overviews?
The shared foundation remains the quality and completeness of your product data. The differences sit in the sources: Google surfaces favour the Merchant Center feed, while comparison-style answers lean more on third-party sources, reviews and articles. Track each engine separately, since their source selections overlap weakly.
Sources
- “Google AI Overviews Now Appear on 14% of Shopping Queries”, ALM Corp, drawing on the Visibility Labs analysis of 20.9 million shopping keywords, March 2026 — almcorp.com
- “Merchant Center’s New AI Performance Insights Report, Explained”, Digital Applied, 14 July 2026 — digitalapplied.com
- “Google adds AI shopping insights to Merchant Center”, MarTech, May 2026 — martech.org
- Johan Sellitto, “Google lance officiellement les AI Overviews en France !”, Abondance, 22 July 2026 — abondance.com
