How ready are Swiss online shops for AI shopping agents?

Are Swiss online shops ready for AI shopping agents?

Not yet, on the evidence of the 240 shops we audited in Geneva, Vaud, Valais, Fribourg, Neuchâtel and Jura in July 2026. No shop reached an A, and in three quarters of the shops read in full, an agent could not read the product it came to buy: the price, stock and identity were on screen but not declared as data.

The practical reading is that most of these shops are visible to agents and readable as text, and lose them at the point where the agent needs a fact it can verify. That is the cheapest stage to repair, because the facts already exist in the shop’s catalogue.

How was the audit done?

Our audit crawler visited each shop’s public site in July 2026 and scored how far a buyer agent could get, stage by stage. It did not log in or place orders, and the dataset was closed on 12 July 2026.

• Which shops. The 240 are a list we compiled of online shops based in the six cantons that sell to consumers, mostly independent brands and specialist retailers. It is not a random sample of Swiss e-commerce, so the figures describe this set of shops rather than every Swiss shop.
• What the crawler checked. Access for answer-engine crawlers in robots.txt, the sitemap and llms.txt, Product and Offer schema on product pages, the seller’s Organization identity, machine-readable ratings, the cart and checkout path, and reorder or account paths after the sale.
• Browser rendering. Where a plain fetch found no structured data, the crawler rendered pages in a full browser to confirm the gap, which matters for shops that inject their JSON-LD with JavaScript. For 38 of the 211 shops read in full, the render confirmed there was no JSON-LD at all.
• Confidence filter. Each audit was graded high (91 shops), medium (120) or low (29) confidence. Low means the site could not be read deeply enough, usually because the crawl was blocked or too thin, so those 29 count in the grade distribution and the median but never in the stage, price or sector findings.

What does the score mean, and what does it not?

The agent readiness score runs from 0 to 100 and combines the five stage scores. Grades follow 15-point bands, with B from 70 to 84, C from 55 to 69, D from 40 to 54 and F below 40, so an A needs 85 or more, and the best shop in the set scored 84.

The audit stages map onto the five stages this site uses for an agent’s purchase:

Modelled throughput is the audit model’s estimate of the share of agent purchase attempts that could complete on the site, given what failed at each stage. It is an estimate from the crawl and not a measurement of real orders or sales, and a shop with a 4% throughput may sell well through its human visitors. The main leak is the stage where the model loses the largest share of agents.

How are the grades distributed?

Nearly two thirds of the shops landed in C or D, a quarter reached B, and none reached A.

The low-confidence audits weigh on the F grade: 9 of the 28 F grades came from sites the crawler could not read in full, which is a failure for an agent too, but a less certain one.

At which stage do AI agents drop out?

At Understand, for 157 of the 211 shops read in full. These shops let crawlers in and have pages an assistant can summarise, but the product itself is not declared as data, so an agent cannot confirm the price, the stock or that the item is the one it was asked for.

The price finding shows the mechanism. On 117 of the 211 shops, the only price the crawler found was in the page text, with no Offer.price to confirm it. Google asks site owners to make sure their structured data matches the visible text (updated 10 December 2025), and a price that exists only as text gives an agent nothing to match. Discovery is rarely the problem: only 1 shop of 211 lost most agents there, so blocking crawlers is not what holds these shops back.

Which sectors are furthest behind?

Food and drink, the largest group, had the lowest median score at 56 and the highest share of text-only prices, while watches and jewellery led at 68. The sector figures use the 211 shops read in full.

Food and drink covers chocolatiers, wine producers, coffee roasters, breweries and fine food, and its fields go beyond price, as set out for agentic commerce in food, wine and chocolate shops. In fashion, sizes and variants add a second layer once the price is fixed, covered under agentic commerce for fashion. The sector groups of 20 to 31 shops are small, so a difference of a few points between them is not meaningful.

What should a shop fix first?

The product data, because it is where three quarters of these shops lose agents and because the facts already sit in the catalogue. The order below follows how many shops lose agents at each stage.

1. Product and Offer markup on every product page, generated from the catalogue, with price, priceCurrency in CHF, availability and a GTIN where one exists, present in the HTML before JavaScript runs and matching the visible price.
2. Seller identity and ratings, an Organization block and machine-readable reviews, which is the main leak for 40 of the 211 shops.
3. A purchase path an agent can follow, with price, stock and the add-to-cart URL exposed together on one offer.
4. Reorder and order-status paths, which matter most for consumables such as coffee, wine and supplements.

The chocolatier audit shows the first step on one shop from this set: named in 58% of the Google AI Overviews we tested, with no structured price on any of 26 rendered product pages.

Where do these findings stop holding?

The audit covers French-speaking Switzerland only, and one point in time, July 2026. Of the 240 sites, 174 were read in French and 66 in English, and the shops in German- and Italian-speaking Switzerland were not part of it. Shops change their markup, so a shop’s grade today may differ from its grade in July.

The score measures what an agent can read and do on the public site. It does not measure AI visibility, meaning how often assistants name a shop, which depends far more on the guides and reviews they cite. Modelled throughput is an estimate, not a sales figure, and agent checkout itself launched in the United States first, so these shops are preparing ahead of a rollout rather than losing Swiss agent orders today. The guide to agentic commerce in Switzerland covers what that rollout depends on, from TWINT to CHF prices with VAT.

How do I check my own shop?

Run one product page through the free readiness check, which reads it the way the audit crawler does and shows which fields are missing. For the whole shop, the agentic commerce readiness check repeats this audit on your catalogue, adds the AI visibility measurement from the chocolatier case, and hands your developers the fixes in order. If you are new to the topic, what agentic commerce is explains the five stages in plain terms first.

The audit used public pages only, and no shop in it is named here.