How we use artificial intelligence

We sell food and wine from small producers, and the first thing we care about is that you can trust what you see. So we are putting in writing where we use artificial intelligence, where we don't, and the rules we have set for ourselves.

In short

  • The photographs of the products we sell are real photographs. On some shots we use AI to improve how they look, removing reflections and small blemishes: it cleans up, it doesn't invent.
  • The illustrations that accompany our guides and recipes are not photographs: they are generated.
  • Our texts begin with our own ideas and our own information, but a good part of the writing is done by AI. We check them and approve them before they go online.
  • We don't estimate numbers: we calculate them, or we read them off the label.

The product photographs

Every bottle, jar and package you find for sale has been photographed. On those photographs we use automated tools to cut out the background and standardise the framing against white, and on some shots we lean on AI to improve how the picture reads: removing a reflection on the glass, a small blemish, the noise of a shot that didn't come out well. It is the same work that used to be done in the darkroom and later with photo editing, only faster.

What we do not do is alter the product. The colour of the wine, the shape of the bottle, the format, the capsule, the cork: those stay as they are. If a photograph strikes you as better looking than the real thing, that is the light, not us changing what's inside.

One commitment in particular, because it matters most: we never generate the label. Vintage, denomination, alcohol content, ingredients, awards — these are things you read on the real label. No tool rewrites them, and none "reconstructs" them to make them look sharper. If a label comes out illegible in a photograph, we shoot the photograph again.

On some shots the bottle — always photographed — sits in a setting created on a computer: a table, a light, a background, to give the product page a little atmosphere. From August 2026, every image of this kind that we publish carries the disclosure written on the image itself, not only on this page. Older shots do not carry it: they were made when this was not yet our practice.

What we do not do, and never have, is use a generated setting to suggest a specific place. If we show you a producer's vineyard or cellar, it is a photograph of that vineyard and that cellar: our articles about the producers and the territories are illustrated with photographs, many of which come to us from the producers themselves.

The illustrations in guides and recipes

Some images on the site document nothing: they are there to give a text some atmosphere. They are the opening images of our wine and beer guides, and the ones that accompany the steps of a recipe. Those images are generated with AI. From August 2026 we write the disclosure next to the image; for the illustrations published earlier, you'll find it declared here.

Worth saying plainly: the dish you see in one of our recipes is not a photograph of that dish cooked in our kitchen. It is an illustration of how it ought to turn out. The method, the quantities and the timings, on the other hand, are the real ones, and the nutrition figures are calculated — more on that below.

The texts

Here we would rather be explicit, because this is the question that counts. Ours are the starting idea, the direction and the information: which product to tell you about, what there is to say about a wine, which figures are true, what we don't want to write. The writing, in good part, is done by AI: we point the way, the model drafts, and the text you're reading comes out of that step. The versions in the other languages are made the same way.

So yes: many of the words you read were strung together by a model, not by a person. What does not change is where the information comes from and who answers for what it says. Before publication we check and approve every text: if a sentence doesn't convince us, or says something inaccurate, it doesn't go online. If you find an error in a text, it is our mistake, not a program's.

The numbers

Here we have set ourselves a stricter rule than the one we're held to: we don't publish a figure that isn't calculated or read from a source. No number on this site is a plausible estimate.

  • The nutrition figures in our recipes are not estimated. They are calculated from the ingredients: the compositions are added up and divided by the number of servings. When an ingredient is a product from our own catalogue, its composition is read from the nutrition table on its label, so it updates itself as soon as the producer updates it. And if a single ingredient can't be recognised or can't be weighed, the calculation isn't published at all: a total with a piece missing looks complete, and that would be worse than no total.
  • The food and wine pairings we decide ourselves, dish by dish, based on the labels we actually have in the cellar. They are not automated suggestions.
  • Prices, formats, alcohol content and availability come from the catalogue and from the producers.
  • The reviews are from real customers. We don't write them, we don't generate them, and we don't delete them for being negative.

What we don't do

  • We don't generate reviews, comments or testimonials.
  • We don't publish computer-generated images of producers, of people, of cellars or of places and pass them off as photographs.
  • We don't use chatbots or automated assistants: if you write to us, including through WhatsApp support, a person answers.
  • We don't use AI to show you a different price depending on who you are or how you browse.

If you find a mistake

This page only holds up as long as it stays true. If an image, a figure or a description looks wrong to you, write to us: we'll correct it, and if the mistake was ours we'll say so. It's the fastest way to keep us honest.

Last updated: August 2026. This page describes our practices and supplements the information required by Article 50 of Regulation (EU) 2024/1689 on artificial intelligence for content generated or modified with AI. The disclosure on individual generated images stays next to the images themselves: this page explains it, it does not replace it.