Getting recommended by ChatGPT when people look for a wine

A new way to choose a bottle

"Which Bordeaux would go with a leg of lamb, for around 25 euros?" Many wine lovers used to ask their wine shop that question. More and more, they also ask ChatGPT, Gemini or Perplexity, or type it into Google, which now displays an AI-written answer at the top of the page.

For an estate, the stakes are simple: when the answer names three or four wines, is yours one of them? And if it is, is what's said about your cuvée accurate?

No one can guarantee being cited. But you can understand how these answers are built, and give yourself the best possible chance.

How an AI builds a wine recommendation

AI engines don't taste wine. They put together an answer from what they find, or what they have learned, on the web. Depending on the tool, two mechanisms work together:

• what the model learned during training, from a large volume of public text;
• what it looks up live: Perplexity, Google AI Overviews, and ChatGPT or Gemini when they browse, consult web pages at the moment of the question, then summarize them, often with links to their sources.

In both cases, the AI is looking for precise, consistent information. To recommend a cuvée, it needs to know which wine it is (appellation, color, grape varieties, vintage), how it tastes, what to serve it with, how much it costs and where to buy it. If that information can't be found, is vague, or contradicts itself from one source to another, the wine is less likely to appear in the answer.

The sources they rely on

A recommendation rarely draws on a single source. AI engines generally cross-check:

• the estate's website, especially its cuvée and vintage pages;
• wine guides and specialist media: tasting notes, articles, selections;
• the press, national or regional;
• listings from wine shops and online retailers that sell your wines;
• customer reviews published online;
• directories and databases of estates and appellations.

Your website is the only one of these sources you fully control: that's where to start. The others you can influence by providing accurate information.

What an estate can actually do

Complete cuvée and vintage pages

This is the top priority. Many estate websites present a cuvée in three lines and a photo. For an AI, that's not enough. A useful cuvée page includes at least:

• the appellation, color, grape varieties and their proportions;
• the vintage, with what sets it apart (growing conditions, harvest, aging);
• a factual tasting note: nose, palate, aging potential;
• concrete food pairings;
• serving temperature;
• price and availability.

When the vintage changes, its page needs to exist or be updated. An AI that finds a sold-out vintage shown as available, or a price that's three years old, may repeat incorrect information.

Structured markup

Schema.org structured data (notably the Product and Offer types) lets you describe a cuvée, its price and its availability in a format that search engines read unambiguously. Invisible to visitors, this markup can usually be generated automatically from product pages.

Consistent information everywhere

If your website says 60% Merlot, a wine shop says 70% and a guide says "mostly Cabernet," the AI doesn't know whom to believe. Take stock of everywhere your wines are presented (wine shops, online retailers, guides, directories) and check that the basic data matches: exact cuvée name, appellation, blend, available vintages.

The simplest approach: one reference tech sheet per cuvée and per vintage, sent to all your retailers.

An FAQ that answers real questions

The questions people ask AI engines resemble the ones you hear in the tasting room: which wine with which dish, how long to cellar it, what temperature to serve it at, what's the difference between two cuvées. An FAQ on your website, with short, precise answers, gives the engines wording they can use directly.

Regularly updated content

Publishing regularly (a vintage release, the harvest, a profile of a grape variety) multiplies the pages an AI can draw on. Each piece of content should link to the relevant cuvée pages, so information flows through the site.

What France's Évin law allows, and what it prohibits

In France, the Évin law regulates communication about alcoholic beverages, including online. It allows objective information: origin, appellation, composition, production method, taste characteristics, ways of serving. It prohibits encouraging consumption and associating alcohol with social, athletic or sexual success.

The good news: what AI engines are looking for is exactly what the law allows. A precise tech sheet, a factual tasting note and food pairings all count as information. Stay in that register, and include the mandatory health warning.

How to measure your presence in AI answers

The manual method is to ask your customers' questions yourself ("which Graves wine with fish?") and note who gets cited. It's instructive, but slow, and the answers vary from one time to the next.

Our GEO Radar tool automates this test: it asks the main AI engines questions related to your business and shows whether your estate appears in the answers. It's free. Repeating the measurement at regular intervals shows whether the actions you've taken are having an effect.

The Château Cazebonne example

Château Cazebonne is an independent estate in the Graves. Its website sold its wines well, but remained invisible for broad queries (Bordeaux wine, appellations, grape varieties). It had no content plan, no logical linking between pages, and no tool to know what was working in Google.

We started by mapping actual demand: queries about wine, appellations, grape varieties, pairings, terroirs. Then we built an architecture of hubs and satellites:

• pillar pages: a directory of Bordeaux estates, appellation, region and ancient grape variety pages;
• product and range pages that build on these pillars;
• an editorial journal (stories, tastings, behind the scenes at the estate) that brings freshness.

Links between these pages are managed from the back office, with automatic suggestions and an audit of broken links. SEO tags, structured data and the sitemap are handled at the page template level. Search Console data is built into the site's admin, so decisions about what to write and what to rework are based on numbers.

Writing is AI-assisted, with a tone-of-voice framework specific to the estate, structured briefs and systematic human review. The team publishes without depending on a developer.

The full details are in the complete case study.

Where to start

If you were to do only three things this quarter:

1. Complete your cuvée pages, starting with your best-selling wines.
2. Send an up-to-date tech sheet to each of your retailers.
3. Measure your starting point, so you can compare in a few months.

Want to know where your estate stands? We carry out a full audit starting at €1,000 excl. VAT, delivered in 8 to 10 days. You can also test your presence in AI answers for free with GEO Radar. Our work with wine estates is presented on the Wine estates page.

Frequently asked questions

Can you pay to be recommended by ChatGPT?

No. Answers generated by ChatGPT are not advertising placements. They rely on the information available on the web. The only way to increase your chances of being cited is to make that information complete, accurate and consistent.

Do you need a new website to appear in AI answers?

Not necessarily. In many cases, completing your existing cuvée pages and adding structured markup and an FAQ is enough to make progress. A redesign is justified when the site doesn't make it easy to manage vintages, prices and availability.

How long does it take to see an effect?

It depends on the engine. Those that consult the web live can take an updated page into account fairly quickly, once it has been crawled. What models learned during training changes more slowly. Hence the value of measuring regularly.