What does agentic commerce mean for food and drink shops?
Food and drink is chosen on origin, ingredients and delivery date, and a gift adds a deadline. Our July 2026 audit found a chocolatier near Geneva named in 58% of the Google AI Overviews we tested, while none of its 26 rendered product pages declared a price an agent could read, so the recommendation ended at the product page.
What the customer sees
Chocolate
Milk chocolate bar with cocoa nibs, 100 g
CHF 7.90
What their agent sees
"@type": "Product",
"name": "Milk chocolate bar with cocoa nibs, 100 g",
"category": "Chocolate",
"countryOfOrigin": "CH",
"additionalProperty": { "name": "Allergens",
"value": "Milk, may contain nuts" },
"offers": {
"price": 7.90,
"priceCurrency": "CHF",
"availability": "InStock",
"deliveryTime": { "maxValue": 2 }
}Price, stock and delivery can be compared
Who this page is for
- Chocolatiers, coffee roasters and confectioners selling online
- Wine merchants and speciality drink producers
- Delicatessen and regional food shops shipping across Switzerland
Usual starting point: Machine Readability Sprint
What does a food and drink shopper ask an assistant?
A typical request reads like this one: "A box of dark chocolate pralines from a Swiss chocolatier, under CHF 60, delivered to Zürich by Friday as a gift." The agent needs the price, the delivery date to that postcode, gift options and allergens, and it needs to confirm the seller is the chocolatier itself.
A box of dark chocolate pralines from a Swiss chocolatier, under CHF 60, delivered to Zürich by Friday as a gift.
Which product fields decide the purchase in food and drink?
In food and drink, five fields decide whether an agent can shortlist a product: price and pack size, ingredients and allergens, origin, delivery date, seller identity. The table shows why each one matters to the shopper's request here and, for your developers, how it is declared so an agent can read it.
| Field | Why the agent needs it | How it is declared |
|---|---|---|
| Price and pack size | Gifts are compared on price per box or per bottle | Offer.price, priceCurrency, weight |
| Ingredients and allergens | A follow-up question in almost every food conversation | Visible text near the top and additionalProperty |
| Origin | Swiss origin, cocoa origin or appellation is often in the prompt | countryOfOrigin, additionalProperty |
| Delivery date | Gifts are bought against an occasion | shippingDetails with deliveryTime |
| Seller identity | The agent checks that it is buying from the maker | Organization linked as seller |
What does the data say about AI shopping for food and drink?
In our July 2026 audit of 73 food and drink shops in Romandie, 58 lost the most agents at product data and 47 showed their price only as page text. Food is 10% of Swiss online sales, and TWINT holds two thirds of mobile distance-selling transactions, which shapes how an agent will eventually pay.
| What was measured | Source, date and scope |
|---|---|
| Among 73 food and drink shops (chocolatiers, wine producers, coffee roasters, breweries and fine food), 58 lost the most agents at product data, and 47 showed their price only as page text. | Autobasket audit of 240 Romandie online shops, July 2026. Switzerland, Romandie, 73 food and drink shops |
| A chocolatier near Geneva was named in 58% of Google AI Overviews and 33% of Perplexity answers we tested, yet none of its 26 rendered product pages declared a price an agent could read. | Autobasket audit, July 2026. Switzerland, one shop, 24 prompts per engine |
| Home electronics made up 28% of Swiss online sales in 2025, fashion 16%, home and living 15% and food 10%. | HANDELSVERBAND.swiss with NIQ/GfK and Swiss Post, 12 March 2026. Switzerland, share by retailers’ main assortment |
| Mobile payment accounts for 74.4% of Swiss distance-selling transactions, and TWINT holds 67.0% of those mobile transactions. | Swiss Payment Monitor 1/2026, University of St.Gallen and ZHAW, February 2026. Switzerland, domestic distance selling, by number of transactions |
Where does an agent purchase break for food and drink?
We audit food and drink stage by stage, because each stage fails for a different reason. The rows below show the failure we look for first at each stage and what changes once it is fixed, in the order a shopper's request meets them, from the first question to the reorder.
-
Discover
Where it breaks
The shop is named in guides and press, but the product pages do not answer gifting questions
What changes once it is fixed
Gift, shipping and allergen answers on the pages assistants retrieve
-
Understand
Where it breaks
Prices appear only in body text, with no
ProductorOffermarkupWhat changes once it is fixed
A price and availability the agent can read on every product, generated from the catalogue as
ProductandOffermarkup -
Choose
Where it breaks
No ratings or seller identity an agent can read
What changes once it is fixed
The agent can confirm it is buying from the maker and read your ratings, through an
Organizationblock and ratings in the markup -
Buy
Where it breaks
A cart exists, but an agent cannot confirm the offer it is buying
What changes once it is fixed
The agent can confirm the exact box and price it is buying, then complete the checkout where the channel allows it, with prices in CHF including VAT
-
Retain
Where it breaks
No reorder or subscription path for coffee or wine
What changes once it is fixed
Coffee and wine reorders the assistant can place on a schedule the customer approved once
Where should food producers start?
We usually start this kind of engagement with the Machine Readability Sprint. In our chocolatier audit the gap was almost entirely structured data, which the Sprint covers. From there, the order of the remaining work follows the stage table above, so each step fixes the next place where agents drop your products.
In our chocolatier audit the gap was almost entirely structured data, which the Sprint covers.
What should food producers read next?
These guides and worked examples take the questions on this page further for food and drink. Each link names the topic it covers and says in one line why it matters for a shop like yours, so you can open the one that answers your next question rather than reading all of them in order.
- Why AI recommends a chocolatier that agents cannot buy from
The case behind the 58% figure: a shop assistants already name, and the product data that stopped agents at checkout.
- Agentic commerce in Switzerland: TWINT, CHF and VAT
How CHF prices, VAT and TWINT look to an agent, which matters for every Swiss food and wine order.
- Are Swiss online shops ready for AI agents? 240 shops audited
The full audit of 240 shops, including the 73 food and drink shops summarised above.
- How to get your products recommended by ChatGPT
How assistants pick the passage they quote, so gifting and allergen answers end up in it.
What do food producers ask about agents?
These are the questions teams in food and drink ask us most often about AI shopping agents, answered in full. Each answer stands on its own, so you can forward one to a colleague, and an assistant asked the same question can quote it as it is.
How do I get my wine shop recommended by AI assistants?
Publish each wine as a product with its price, bottle size, vintage, grape, appellation and stock as fields, and write tasting and pairing notes as text near the top of the page. Assistants retrieve passages that answer the prompt, so "a Chasselas under CHF 25 for a fondue" is answered from pages that state those facts. Age and shipping rules then decide which channels can complete the sale.
How can a food shop handle shipping and freshness questions from AI agents?
Declare the delivery time per region in shippingDetails, and state the conditions that matter for perishables in visible text: chilled shipping, the days you dispatch and the shelf life on arrival. An agent asked for a gift by Friday compares a stated delivery time against that date, and it skips an offer that only promises fast delivery.
Why would an AI assistant recommend a shop it cannot buy from?
Because recommendation and purchase rely on different evidence. Assistants name shops that appear in guides and press, while a shopping agent needs a structured price and availability on the product page. Our chocolatier audit found the first without the second, so the recommendation ended on a page an agent could not use.
How should a food shop declare allergens for AI agents?
In visible text near the top of the product page, so it appears in the passage an assistant retrieves, and as additionalProperty fields in the structured data. Allergen statements must match the packaging, because what an agent repeats about your product remains your responsibility.
Can wine and spirits be sold through AI agents?
Age checks and alcohol shipping rules differ by country and canton, and agent checkout channels handle them unevenly. We map which assistants can complete such a purchase in your market before building anything, and often the right first step is recommendation rather than checkout.
Get your food and drink catalogue ready for agents
Bring one product page from your catalogue to the first conversation, and we will show you which of the fields above an agent can read today and which one it is missing.