The wine industry is caught in AI’s revolving door, but what matters is what a winery itself knows about its wine. Two authors in an Italian trade magazine approach the subject through facts. AI can read scattered material and answer questions in natural language, but only about what can be found online.
Image / Unsplash: Nahrizul Kadri
In the wine industry, AI is still talked about much the way websites were talked about twenty years ago. There are chatbots, image generators, automatic translations, and programs that write, in a few seconds, a text that used to take a marketing department an afternoon. All of this is interesting, but really secondary. The most significant effect of AI on the wine industry is not that it can write a wine description faster. The biggest effect is that, for the first time, the wine industry’s vast, scattered body of information can be brought into a form that can be discussed and worked with.
The idea is best understood through a concrete example. Lorenzo Biscontin, founder of the Italian service CantinaItalia.ai, told the magazine Il Corriere Vinicolo in an interview (no. 31, 5 October 2026) that in five minutes the service produced an interactive, Google Maps-based map of the Sarmassa cru in Barolo. The map showed the neighbouring crus, geographical and geological information, the style the crus lend to their wines, and the area’s producers. According to Biscontin, the service had not been given any instructions for making maps. The core of the example is clear nonetheless. The user does not need to know which database the information comes from or which program builds the map. It is enough to tell the machine what you want to know. In Biscontin’s view, this is a far bigger change than a new way of making a PowerPoint presentation or an Instagram post. You no longer have to speak the computer’s language to the computer.
The wine industry’s vast body of information has so far been scattered
The wine industry is unusual in that an enormous amount of information has accumulated around it, but that information is very rarely in one place. A winery’s history is on its website, the location of its vineyards on another page, the production rules in a consortium document, the geology in a study, vintage data with the winemaker, and consumers’ experiences somewhere else entirely. Anyone who wants to piece together the whole picture has to do real detective work.
For AI, this is not a problem in the same way. It can process large amounts of unstructured information, organise it around a question, and answer in natural language. Biscontin calls this a paradigm shift. Previously, the web gave you at best a list of addresses; now it gives you an answer. According to the title of the interview, the emphasis shifts from quantity to value. According to the magazine, the service holds the contact details of about 3,500 Italian wineries, and the aim is to cover all Italian wineries and to offer as comprehensive information as possible on Italian wine. Its knowledge base is built on, among other things, digitised DOCG, DOC and IGT production rules. According to the magazine, the language model used is Anthropic’s Claude.
This sounds very technical, but for the wine industry everything is significant from a marketing point of view. According to Biscontin’s description, AI does not need a menu, a search field, or a ready-made database structure in the way a traditional web service does. It receives a question as a person puts it and builds an answer using several pieces of information.
The winery website is not dead
According to some forecasts, AI was supposed to make websites unnecessary. Biscontin considers this way of thinking mistaken. According to him, wineries often ask whether it is still worth investing in a website if people look for information from AI. He answers that the website is one of the sources general-purpose chatbots draw on, and the only one the winery fully controls. The winery decides for itself what is said about its vineyards, soils, grape varieties, viticulture, winemaking and history.
The other most important sources are, according to Biscontin, more independent by nature. Wikipedia, Reddit and YouTube are thus platforms that are poorly suited to pure propaganda and better suited to people’s assessments. In an AI answer, a winery’s own site may therefore end up alongside sources the winery does not control, and the winery cannot decide how AI weights the content. For this reason, a winery should ask itself a fairly simple question. If AI is asked tomorrow what makes this winery or this wine special, will there be enough information online to give the right answer?
In many cases, there will not.
“Elegant, mineral and fruity” is no longer enough
Now we come to the hardest part of wine marketing. For decades, wineries have described their wines with roughly the same vocabulary. The wines are elegant, balanced, mineral, fresh, complex and long-lived. They have dark berries, spice, minerality and a long finish. All of this may be true, but the problem is that the same text fits thousands of other wines.
Precisely this observation also emerges in the same issue of the magazine (pp. 12–14), in which Carlo G. Ferretti, founder of Geo Identity Research (GIR), describes his own data from seven Sauvignon Blanc vineyards in the same area. He followed the vineyards from the soil to winemaking, tasting, and marketing texts. This is a series of articles published in a trade magazine, not peer-reviewed research, and Ferretti himself calls Sauvignon the first case in the series. One of his observations was that two retailers described two different wines with the same paragraph, in the same order and with the same punctuation, even though the wines came from the vineyards with the most distant soils in the study.
The observation rests on the GIR data as described by Ferretti, not on an analysis made by AI. It nevertheless shows how little some wine marketing says about the very wine being marketed. If AI answers questions on the basis of texts on the web, the same problem carries over into its answers.
A wine’s real difference arises long before the bottle
What is also interesting in Ferretti’s data is what happens before the wine ends up in the bottle. The seven Sauvignon Blanc vineyards were in the same area, of the same variety and the same vintage, and the wines were made separately. Even so, considerable differences were found between them. Three vineyards, only a few minutes’ walk apart, belong to the same class on the geological map, with the same colour and the same code. When the soil was examined by its minerals, however, the share of foreign material carried by a glacier was less than a fifth in one vineyard, almost as small in another and more than half in the third (15, 18 and 59 per cent). The area’s geological map is therefore not wrong; it simply was not made at that level of detail.
The cellar, for its part, evened out the differences. Without any shared protocol, the winemakers made largely the same choices: malolactic fermentation was blocked in all seven wines, fermentation was almost always taken to dryness, and yeasts were almost always chosen to produce the same aromas. Acidity, pH, and sugar were much closer to one another in the finished wines than in the juices. Part of the soil’s chemical “signature” still made it all the way to the bottle, because it cannot be adjusted in the cellar. In Ferretti’s words, the cellar filters terroir but does not remove it.
In the glass, origin was harder still to tell apart. Twelve trained tasters assessed ten glasses blind and tried to group wines that resembled one another. They did not know that there were only seven wines: of each taster’s ten glasses, three were a second pour of the same wine. This gave a total of 36 pairs of duplicates. The tasters placed the two pours of the same wine in the same group only twice, although by chance this should have happened about seven times. The laboratory, by contrast, clearly distinguished the wines: the concentration of one compound typical of the variety was many times above the detection threshold in one wine, but only one taster out of twelve mentioned it. Ferretti stresses that the result does not mean the wines were alike or the tasters poor. A panel is also not the market.
The result is nonetheless uncomfortable. In wine we talk endlessly about terroir, but according to Ferretti’s data, part of the difference evens out in the cellar, disappears in the glass and after that vanishes once more in words. The difference arises in the soil and shows clearly in the grapes and the juice, but in the story of the wine it can no longer be seen.
AI can look for a difference people do not notice
Perhaps the most interesting use of AI is no longer generative AI but analysis. Ferretti speaks of data science, machine learning and analytical AI, and stresses that he does not mean AI that produces text or images. According to him, this is AI that does not write or invent, but reads. It looks for signals in complex data.
Seven vineyards and hundreds of variables make about 16,000 possible connections, and chance alone can bring more than a hundred seemingly significant ones to the surface. Without a method, it is easy to call mere coincidence a discovery. On the other hand, a real signal may also be hidden in the data so that nobody notices it. According to Ferretti, humans and AI work in both directions. Sometimes the signal is found by a person (in the study, a fossil shell found in a vineyard revealed carbonate where none was expected); sometimes the analysis brings up a relationship that is too weak, complex, or unexpected for the human eye. Then the person’s task is to understand it, check it in the field, and turn it into a decision.
A winery may hold decades of accumulated data on weather, soil, yield, viticulture, fermentation, maturation and the sensory properties of its wines. There may be so much that no single person can grasp it as a whole. AI can help with that.
But here, one should still keep a cool head. AI must not become a quasi-researcher who proclaims the truth. It should be a partner that finds a possible connection and gives the person a reason to look more closely. The human asks, the machine searches, and the human checks. At its best, this produces something neither could do alone.
A winery could get its own digital sommelier
From the consumer’s point of view, the change could be even bigger. At present, buying wine often rests on familiarity, recommendation, a shop’s range, or whatever one happens to see on social media. AI could turn this into a dialogue between the consumer and the retailer or producer. The examples below are hypothetical scenarios, and it has not been shown that consumers actually choose wine on the basis of AI recommendations.
The consumer might no longer ask what a good Barolo is. They might ask: “I like Burgundy Pinot Noir and want a Barolo that is elegant and not too heavy. My budget is 40 euros. What do you recommend?” Or: “I’m going to a dinner where truffle pasta is served. I want an Italian wine with enough structure but not too much alcohol.” There are already a few such services, but their problem is that the answers are generic and imprecise.
If AI knew a winery’s production well enough, it could give a good answer. The winery could get something it has never had: a digital sommelier that knows the entire production and talks with customers around the clock in any language. Biscontin uses as his example CantinaItalia.ai’s consumer-facing assistant, which answers questions about, among other things, winery visits and reception. According to him, it works around the clock and in all languages and forgets nothing, and staff are left with more time for more valuable work, such as targeted, personalised relationships with customers who have already visited. The service provider also claims that it has been trained to answer only wine-related questions and only when it is sure of the answer. This is the service’s own claim, and a wrong answer is always a reputational risk for the winery.
AI can also reach beyond the cellar
An even more interesting idea concerns the management of the business. According to Biscontin, many wineries, even mid-sized ones, do not prepare a budget, and their management software is often geared only to tax obligations. According to him, CantinaItalia.ai’s agents produce a complete budget covering all areas of the business in a few minutes, and in an hour they can develop ten scenarios, refining the starting point by talking to the machine. Where spreadsheets, financial expertise and time used to be needed, a system that works in natural language can produce a first version quickly and then change the assumptions based on the conversation. But calculations must always be checked, because for now, at least, AI makes many mistakes.
“What happens if sales to Germany grow by 20 per cent?” “What if the harvest shrinks by 15 per cent?” “What happens to cash flow if we build a new cellar?” AI does not need to know only the answer. It needs to be able to ask follow-up questions. That is where its real benefit comes from.
The wine industry should now do one thing
There is no point in the wine industry trying to stop AI. There is even less point in using it merely to produce more mediocre content. Instead, wineries should start organising their own data. Not because Google demands it, and not because AI is fashionable, but because at present AI is not being put to proper use.
The data exists, that is: Where is the vineyard located? What is its soil really like? What makes it different from the vineyard 300 metres away? What does the winemaker do differently? What do the analyses say? What have consumers observed over the years? How do the vintages behave?
When the information exists, AI can help make it understandable. But AI cannot conjure real information out of thin air.
In this sense, AI may renew the wine industry by forcing it back to the original question: what makes this wine exactly this wine?
The answer is probably not a generic “elegant, mineral and long”, but may be one particular vineyard, a particular soil, a particular slope, particular old vines, a distinctive way of farming or a single decision made in the cellar. And that is exactly what is worth telling. In Ferretti’s words: a measured story is the only one a competitor cannot copy.
In the age of AI, a winery’s most important marketing tool may not be AI at all, but the knowledge that lies in the winery’s own backyard.
AI only teaches the world to ask about it.
Sources: Fabio Ciarla, “Biscontin: Con l’AI cambia il paradigma”, Il Corriere Vinicolo no. 31, 5 October 2026, p. 11. Carlo G. Ferretti (GIR), “Sette vini diversi. Ma chi se ne accorge?” and related articles, same issue, pp. 12–14.