E-commerce and AI: Are your products being cited?

When a shopper asks ChatGPT or Perplexity which product to choose, your product pages appear in the answer under three conditions: clean structured-data markup, factual text that stands on its own outside its original context, and detailed rather than numerous reviews. Swiss e-commerce is not facing a rupture; it is undergoing a precise shift — the product-research phase is partly migrating to conversational environments — and that shift can be worked on, then measured.

Article published in January 2026, fully revised on May 25, 2026.

A third search regime has taken hold

Before 2024, online product research split between two regimes: a Google search pointing to product pages or marketplaces, and browsing within those marketplaces themselves. Merchants wanting to be found optimized their product pages for Google and cultivated their marketplace presence.

Since 2024, a third regime has joined them: conversational search. A shopper torn between several technical solutions, comparing brands on precise criteria, or wanting contextualized advice without browsing twenty product pages, formulates their question in a generative environment. Perplexity has rolled out a dedicated Shopping feature; ChatGPT and other environments offer search experiences including product comparisons; Google has built AI Mode and AI Overviews directly into its interface, in line with the doctrine it published on May 15, 2026^[1].

This third regime does not replace the first two. It captures part of the upstream journey — the moment when the shopper has not yet decided — and that is precisely the moment when being absent costs the most.

What do models read in a product page?

When a generative environment synthesizes several sources to answer a shopper, it makes a selection among the product pages it consults. The criteria have long been familiar to e-commerce; what changes in 2026 is their severity.

Structured data in JSON-LD format — the Product, Offer, AggregateRating, and Review schemas from Schema.org — is the language through which a product is correctly understood by machines^[2]. A page without markup can be misread, or ignored in favor of a better-described competing page. The requirement is not new; the penalty for its absence is.

The page text must provide factual elements that hold up outside their original context. "The ideal solution for your needs" offers no citable information. Technical specifications, typical uses, product limitations, return conditions, actual delivery times: this is the material a model can draw on to answer a shopper. The rule applies to the product page just as it does to the rest of the site — it's the same work as building a site designed to be cited, which I describe for service businesses.

Verified reviews count according to their quality, not just their volume. A product with two hundred generic five-star reviews is used less by models than a product with forty detailed reviews describing actual use. This reversal changes the economics of review collection: the chase for volume gives way to an incentive for precision.

Three languages, three visibilities

The Swiss market operates structurally in three languages, with English added for a significant share of B2B transactions and international purchases. A page available only in French is invisible to the German-speaking shopper who phrases their query in German in a generative environment; a poorly translated page sends a degraded brand image on the French-language query.

The trilingual requirement is not a 2026 discovery — serious players in Swiss e-commerce have carried it for a long time. What changes is the penalty: models process each language independently and produce different recommendations depending on the query language. A brand that sacrifices a language sacrifices a share of market it will not see in any of its standard SEO dashboards.

Prices and personal data: the Swiss guardrails

Dynamic pricing powered by predictive models has spread across e-commerce worldwide. In Switzerland, its deployment calls for particular restraint: commercial culture remains attached to price transparency, and the Price Indication Ordinance governs price display with a rigor above the European average^[3]. Moderate, legible adjustments remain compatible with this framework; abrupt, opaque variations erode trust — the asset that is slowest to rebuild in a market of limited size.

AI-driven personalization, meanwhile, falls within the scope of the FADP as soon as it processes personal data: contextual recommendations, chatbots, behavioral profiling. The obligations are precise — clear consumer information, a legal basis, caution on international transfers^[4]. A merchant who automates without building these obligations into the initial scoping exposes itself to a risk disproportionate to the expected conversion gain. Conversely, disciplined, transparent deployment becomes in itself a trust signal that Swiss consumers recognize. When this type of tooling — personalization, catalog automation, internal assistants — goes beyond the question of visibility, it falls under the work of tailored solutions.

Three trade-offs for a Swiss merchant in 2026

The shift described here does not call for a major technical overhaul for most merchants. It calls for three trade-offs.

The first is editorial: revising product pages — and more broadly the site hosting them — so they carry factual information usable outside its context, in every language covered. The work is done category by category, progressively, and benefits all three search regimes at once — classic Google, marketplaces, generative environments.

The second is structural: auditing existing JSON-LD markup, identifying pages missing the Product, Offer, AggregateRating, and Review schemas, and closing the gaps. This is technical integration work, not a lengthy project.

The third is measurement: assessing the citability of the brand and its products in generative environments, on the queries shoppers actually formulate. This measurement did not exist two years ago; it now follows codified protocols, such as the one I published in MCVA Cahier No. 1 — the logic of separating ranking from citation is the same as for service businesses, which I detailed in AI Visibility: Does Your SME Appear in ChatGPT?. To place these trade-offs within an SME's general roadmap for AI, see What Should a Swiss SME Do About AI in 2026?.

This discipline — honest, complete product pages, clean markup, sustained multilingualism, built-in compliance, measurement of the gaps — is not spectacular. Nor is it optional. Swiss e-commerce is not in peril; it is under heightened requirements.

Key takeaways

— Product research is partly migrating to conversational environments: a third regime joining Google and marketplaces. — Models select based on three signals: complete JSON-LD markup, factual text usable outside its context, detailed rather than numerous reviews. — Three trade-offs suffice for most merchants: editorial, markup, citability measurement — in every market language.

FAQ

How do I know if my products appear in AI answers?

By measuring it directly: formulate the queries your shoppers actually ask, run them across several environments — ChatGPT, Claude, Perplexity, Gemini, Mistral — and document where your brand appears, in what context, and with what accuracy. A single test is not enough; it's repetition at a steady cadence that yields insight.

Is Schema.org markup mandatory?

It is not legally mandatory, but it is the language through which machines unambiguously understand a product page: product, price, availability, ratings, reviews. A page without markup can be misread or ignored in favor of a better-described competing page.

Do I need to translate all my product pages into German and Italian?

Neither all nor none: those of your actual market. Models process each language independently; a missing language is an invisible share of market. Priority goes to the highest-stakes categories, in the languages where your shoppers actually formulate their queries.

Is dynamic pricing allowed in Switzerland?

Yes, within the framework of the Price Indication Ordinance, which imposes strict transparency in consumer-facing display. Moderate, legible adjustments are compatible with this framework; abrupt, opaque variations create a regulatory risk and, above all, a cost in trust.

Are your products cited when your shoppers compare? The Score GEO™ pre-audit measures for free where your company appears — and doesn't appear — in the answers of ChatGPT, Claude, Perplexity, Gemini, and Mistral. A quantified answer within a few days. Request a free pre-audit

Sources

[1] Google Search Central, Optimizing your website for generative AI features on Google Search, published May 15, 2026. developers.google.com/search/docs/fundamentals/ai-optimization-guide [↩]

[2] Google Search Central, Product structured data. developers.google.com/search/docs/appearance/structured-data/product [↩]

[3] Price Indication Ordinance (PIO), SR 942.211. www.fedlex.admin.ch/eli/cc/1978/2057_2057_2057/fr [↩]

[4] Federal Act on Data Protection (FADP), revision of September 25, 2020, entered into force September 1, 2023. www.fedlex.admin.ch/eli/cc/2022/491/fr [↩]

Jérôme Deshaie is CEO and founder of MCVA Consulting SA, an augmented agency based in Valais. Fifteen years serving major international brands, now working directly with Swiss SMEs. Background.