Last updated: 27 August 2026A bottle recommended in seconds, a fermentation monitored without pause, thousands of plot-level data points analysed before the next rain: artificial intelligence (AI) is already performing, across the wine world, tasks that once demanded time, experience and a practised eye. Does that mean it threatens those whose judgement and sensibility give wine part of its meaning?
The answer comes down to a nuance. AI collects, compares, monitors and predicts faster than anyone, but it does not decide what matters, and that remains the business of the sommelier, the winemaker or the oenologist. It does not abolish these professions; it shifts their boundaries, and expertise moves from accumulated knowledge to the ability to interpret.
This essay explores that shift: what AI makes accessible, what it cannot taste, what it changes from vineyard to winery, and then the risks to be managed, from unequal access and data protection to the homogenisation of taste and dependence.
AI and wine advice: expertise at last within reach
AI makes wine advice accessible to consumers whom professional expertise has never really reached. Someone looking for a bottle to go with grilled salmon, on a budget of no more than €15, for a family dinner, does not necessarily feel the need for a sommelier’s advice. What they need is a recommendation that is quick, clear and reliable enough.
On this ground, AI brings new value. Available at any hour, it asks for no specialist vocabulary and instantly compares styles, regions, grape varieties and price points. Apps can already manage a cellar, record preferences, search for bottles or suggest food and wine pairings.
For many people, then, it is the first genuinely accessible form of wine advice. To present the technology as a rival to the sommelier in every case would be a mistake: very often the choice is not between AI and expertise, but between AI and no advice at all, for any number of reasons.
AI can also ease the intimidation that wine sometimes provokes. It explains the difference between two grape varieties in plain terms, helps identify a style or lets a beginner gain a better sense of their own tastes. It does not remove the need for expertise; it may even make people want to seek it out.
AI and tasting: what the machine cannot taste
AI’s first limit lies in its relationship with taste. It analyses sensory descriptions, compares thousands of tasting notes, studies reviews and spots frequent associations between dishes and wines, but it does not experience wine as a human taster does. The neuroscientist Gordon Shepherd reminds us that the taste of wine is not a property of the bottle that could simply be read: the brain constructs it from many signals (Shepherd, 2016). Antoine Hennion and Geneviève Teil speak in this connection of a ‘sociology of attention’.
In practice, AI does not perceive the texture of a sauce, does not taste how a wine evolves after thirty minutes in the glass and does not directly feel the balance between acidity, bitterness, alcohol, texture and length. Above all, it does not live the situation in which the tasting takes place. Much of this knowledge is, moreover, tacit: as the philosopher Michael Polanyi observed, we know more than we can tell.
A food and wine pairing is therefore never a simple equation between two products. Take a roast sea bass with a sauce vierge, grilled fennel and a citrus-infused oil. Given the dish, the cooking method, the textures, the budget and the diner’s preferences, an AI can propose three relevant options (a classic pairing, a regional interpretation and a more unexpected choice) and explain its reasoning. But it does not know whether, that evening, the sauce was sharper than planned, how the bottle is really developing, or whether the guest is after reassurance or surprise. Lucy Suchman showed as much in her work on human–machine interaction: action is ‘situated’, and no pre-set plan replaces adjustment to context.
A technically correct pairing can therefore fall flat, while a bottle that looks less obvious on paper can turn out to be remarkable in the right setting. The relevance of a recommendation sometimes depends less on the quality of the information than on the quality of its interpretation.
AI and the sommelier: from recommendation to judgement
The sommelier’s profession goes well beyond recommendation, which makes it particularly interesting in the age of AI. Its service is, moreover, a subject of study in its own right: a study published in 2025 on ScienceDirect examined its impact on restaurants’ wine revenue.
A good sommelier is not simply someone who knows more wines than the customer. They read a situation: they ask questions, listen, notice hesitations and understand that the same dish can call for different recommendations depending on the person. One guest wants something familiar, another wants to step outside their habits; one is celebrating a special occasion, another is dining with business partners and prefers discretion. Wine belongs to what Lucien Karpik calls ‘singularities’: goods whose value cannot be read from a technical sheet and for which advice plays a central role. The recommendation becomes a form of dialogue, a relational dimension that is hard to reduce to a database.
Sommeliers nonetheless have no reason to keep AI at arm’s length: it can become a remarkable professional assistant. It analyses a wine list, quickly compares several bottles, researches a producer, prepares teaching materials, rewords a technical description, simulates a conversation with a customer, suggests alternatives when a wine is no longer available, speeds up the preparation of training sessions and helps structure a sales pitch.
The division of roles is clear: AI can prepare the decision; the sommelier keeps the judgement and the responsibility. That responsibility imposes a simple rule: check any automatically generated information before using it professionally. Language models can produce erroneous claims with great assurance, as their own designers acknowledge in the GPT-4 technical report (OpenAI, 2023). Producer, vintage, price, availability, technical details or regulations do not become true because a machine states them with confidence.
AI and viticulture: sensors in the service of the winemaker
In viticulture, AI plays the same assisting role: it refines observation without deciding in the winemaker’s place. Sensors, imaging and analytical tools make it possible to gather far more observations at the level of each plot. Soil moisture, weather, temperature, canopy development, water stress and disease indicators can be tracked with a frequency and precision that human observation alone would struggle to match. A recent review of the literature shows the breadth of these applications, from vineyard sensing to winemaking optimisation (Li et al., 2026). Solutions combining on-board imaging, geolocation and AI, such as the Vitidrone project presented by Inria, even analyse individual vines and follow the progress of plots.
The aim is not to replace the winemaker’s eye, but to help them know where to look first. A system can detect an anomaly; it does not turn it into an agronomic decision. The winemaker must still understand the context: the plot’s history, the vine’s stage of development, the weather, the soil, production goals and the estate’s overall strategy. Technology thus strengthens the capacity to perceive without removing the need to interpret.
AI and winemaking: from alert to diagnosis
In winemaking, digital monitoring brings a considerable gain, but an alert is not a diagnosis. Tools can track temperature, density, pH and dissolved oxygen continuously, compare several tanks, spot deviations and trigger an alert before a problem becomes harder to correct.
The signal still has to be interpreted. A change may be normal, temporary, worrying, or a sign of a situation demanding immediate intervention. The oenologist’s expertise keeps its full value here: working out what the change means and what should be done about it. The move from data to decision remains a professional act.
The cost of AI in wine: the risk of unequal access
The benefits of AI will not be evenly spread across the industry. Large estates, major négociant houses, distribution groups and well-resourced restaurants will invest more easily in connected sensors, imaging tools, cellar management software, recommendation systems or predictive analytics, and will be able to recruit or draw on specialist skills to make use of the data.
A small winemaker, an independent wine merchant or a mid-sized restaurant may, by contrast, lack the means to acquire these tools, build them into daily practice or train their teams. The cost goes beyond buying a sensor or a subscription: it includes installation, maintenance, connectivity, training, interpretation of results and sometimes changes to working methods. A gap therefore risks opening between organisations able to turn information into competitive advantage and those left on the sidelines.
Several levers will be essential to stop AI from worsening inequalities:
simple, affordable solutions suited to small estates;
shared tools, run by cooperatives, trade bodies, producer groups or specialist providers;
accessible training, designed for the wine trade rather than for data specialists alone;
clear interfaces that can turn technical data into information useful for a concrete decision;
human support, so that technology strengthens local expertise rather than replacing it.
The challenge is therefore as much economic, organisational and social as it is technological.
AI, wine and data: protecting trust
Data powers AI, and in wine it is never neutral. On an estate, it covers yields by plot, the health of the vines, vineyard practices, analytical results, volumes produced or commercial performance. In a restaurant or wine shop, it can reveal preferences, purchase history, budgets, drinking habits or events organised. This information has strategic value and can be sensitive.
Everyone must therefore be able to protect it: the estate its practices, yields, winemaking choices and commercial data; the wine merchant or restaurateur their customers’ preferences and history. Consumers, for their part, must know what information is collected, why, how long it is kept and who it may be shared with. Trust becomes a condition of adoption: a tool that improves advice but handles data opaquely risks destroying the very relationship it claims to strengthen.
Professionals will need to observe several principles:
obtain clear consent when personal data is collected;
limit collection to the information that is genuinely necessary;
specify the purposes of use: recommendation, stock management, customer service, commercial analysis and so on;
avoid the resale or uncontrolled sharing of data with third parties;
set clear rules for the use of software suppliers or AI models;
secure the data, particularly that of estates, customers or commercial transactions;
allow individuals to access their data, correct it or request its deletion where possible.
Data protection is not an administrative burden: it underpins the legitimacy of AI in a sector built on trust, reputation and human relationships.
AI wine recommendation: the risk of homogenised taste
Recommendation systems risk homogenising taste, because wine resists the simplifications they require. Algorithms are effective when preferences can be expressed as measurable, comparable criteria. Yet a consumer who says they dislike powerful wines may fall in love with one. A habitual Bordeaux drinker may want something entirely different for a particular dinner. A technically flawed wine can become memorable in an exceptional setting. Wine culture is made of contradictions, exceptions and unexpected discoveries.
A recommendation system therefore tends to reinforce what it already knows. If the data favour popular wines, models mechanically promote those best represented in their databases, while small producers, unconventional styles and experiences that are hard to categorise become less visible. The problem goes beyond algorithmic bias: it touches on the diversity of taste. Constantly optimising for predictability could impoverish what makes wine so rich, its capacity to surprise.
Dependence on AI: keeping a critical mind
Dependence is another risk: used without a critical eye, a powerful tool ends up replacing its user’s own reasoning. If every recommendation is routinely validated by an algorithm, professionals may lose the habit of forming their own analysis. AI should therefore remain a second opinion, not an automatic authority.
This principle weighs most in trades where responsibility cannot be delegated to software. When an agronomic, oenological or commercial decision has real consequences, someone must be able to explain why it was taken.
Can AI taste wine? Current limits and prospects
AI’s current limits are not necessarily permanent. Multimodal systems are advancing, sensory databases are expanding and analytical technologies are growing more precise. Machines might model certain aspects of sensory experience with a sophistication that is hard to imagine today, for example by linking aromatic profiles, analytical data, textual descriptions and consumer preferences to produce ever finer recommendations.
To say that AI cannot taste today is therefore not to rule out that some technology may one day represent or predict certain dimensions of tasting. It means that it does not yet perceive wine as a human being does: it feels neither pleasure nor disgust, neither memory nor the emotional context of a shared moment.
Even in that scenario, one question would remain: what do we mean when we talk about expertise? If the aim is to find the statistically best-suited wine for a dish, AI could become extraordinarily effective. If it is to create a memorable experience, convey a producer’s identity, educate a customer or capture the atmosphere around a table, the problem widens. Wine is not merely a liquid to be optimised: it is also a cultural experience to be interpreted.
The wine industry and AI: towards augmented expertise
The most credible future for the wine industry is one of augmented expertise, neither disappearance nor a status quo shielded by tradition. In it, AI takes on the tasks where speed, data volume and repetition count; the professional keeps those that call for perception, interpretation, judgement, human connection and responsibility.
This shift calls for new skills. Wine professionals will not need to become developers or AI specialists, but they will need to understand their tools well enough to know how to:
formulate a precise request;
verify the information generated;
interpret data or probabilities;
identify biases or uncertainties;
understand a model’s limits;
protect confidential data;
keep their capacity for judgement in the face of automated recommendations.
At the same time, certain human skills will gain in value: tasting, observation, listening, teaching, adaptability, a sense of context and the ability to decide when information is incomplete. The more information machines produce, the more the ability to give it meaning may be worth.
What is human expertise worth in the age of AI?
AI may force us to define more clearly what it means to be an expert. By automating part of the knowledge that is easily accessible, it exposes what is not. Expertise is not reducible to memory or to recognising aromas, and it is not a nostalgic resistance to technology. It lies, perhaps, in the ability to exercise judgement when data is incomplete, contradictory or insufficient.
A sensor detects, an algorithm compares, a language model recommends. Someone still has to decide what really matters. Technology will take over what can be standardised; human expertise will increasingly be judged by what resists standardisation. The outcome will not necessarily be a battle between AI and the sommelier, the winemaker or the oenologist, but a new division of roles: to technology what it does best, to professionals judgement, interpretation and human connection. AI democratises access to knowledge and extends our capabilities; human expertise interprets, adapts and gives meaning.
After all, wine has never been only about choosing the right bottle. The most interesting question is not always which wine to drink, but why that wine, on that evening, with those people. Answering it may well remain one of the most singular forms of human expertise.
Keep Reading
References:Hennion, A., and Teil, G. (2004). Le goût du vin : pour une sociologie de l’attention. In V. Nahoum-Grappe and O. Vincent (eds), Le goût des belles choses. Paris: Maison des sciences de l’homme.Inria. Vitidrone veut faire décoller l’agriculture durable. https://www.inria.fr/fr/vitidrone-veut-faire-decoller-lagriculture-durableKarpik, L. (2007). L’économie des singularités. Paris: Presses Universitaires de France.Li et al. (2026). Machine Learning-Enabled Intelligent Technologies Across the Grape and Wine Value Chain: From Vineyard Sensing to Winemaking Optimization. Comprehensive Reviews in Food Science and Food Safety.OpenAI (2023). GPT-4 Technical Report. https://doi.org/10.48550/arXiv.2303.08774Polanyi, M. (1966). The Tacit Dimension. New York: Doubleday.Shepherd, G. M. (2016). Neuroenology: How the Brain Creates the Taste of Wine. New York: Columbia University Press. https://cup.columbia.edu/book/neuroenology/9780231542876/Study on the impact of sommelier service on restaurant wine revenue (2025). ScienceDirect. https://www.sciencedirect.com/science/article/pii/S0278431925000623Suchman, L. (1987). Plans and Situated Actions: The Problem of Human-Machine Communication. Cambridge: Cambridge University Press.