AIRI-Wine industry standard · 27 July 2026 · 7 min

Rheingau: 56 wineries measured — 45 are never named by AI

The first complete regional measurement: all 56 wineries of the Rheingau. 9 of 16 well-known houses get named — but only 2 of 40 independent family estates.

Photo: Pexels

Until now we measured countries: the 20 largest estates in Switzerland, Germany, Austria, Italy and France. That is a reference point, not a peer group — a twelve-hectare family estate stands there next to Antinori. This measurement does it differently for the first time: one whole region, completely. All 56 wineries of Rheingau, asked the questions wine buyers actually ask about the region.

Of 56 estates, 11 are ever named at all — 45 appear in none of the 36 answers. Weingut Schloss Johannisberg leads on 83%.

The split is the more telling number. We divided the panel in advance into two groups: the 16 houses of the region known beyond it, and the 40 independent family estates. 9 of the 16 well-known houses get named (56%) — but only 2 of the 40 independents (5%). Those already known get recommended; those who are not barely appear at all.

The ranking

The first two columns show the outcome — how often AI names the estate. The last two show the website side — how well AI systems can read it. The two measurements are independent and often diverge; that divergence is the finding.

AIRI rating by score: Excellent from 80, Good 60–79, Moderate 40–59, Poor 20–39, Critical below 20.

What the AIRI dimensions show

Three findings

1. Fame decides, not the website. Weingut Schloss Johannisberg leads on 83% with an AIRI of 68/100. What AI recommends here is what books, press and guides say about the region — and those have been the same names for decades.

2. This measurement asks the hardest question — it is a floor, not a forecast. Weingut H.J. Ernst scores 91/100 on AIRI, better than any named house, and appears in no answer at all. So a good score is not a magic number. Generic questions ("which Rheingau estates should I know?") are a fame contest, and it is won by whoever has been in the books and guides for decades.

Put a real buying intent behind the question and the ranking reshuffles. We also put 6 specific buyer questions to the same three systems (18 answers — a probe, not a second benchmark): a stay with a tasting, dry Riesling under 15 euros, a company party for 20, traditional-method sparkling wine. Weingut Schloss Johannisberg falls from 83% to 22%. Weingut Schloss Vollrads leads instead. And 2 estates surface that appear in none of the 36 generic answers — among them Weingut Trenz, AIRI 83/100. With a single mention out of 18, though. We have since run the same probe in four regions: three such estates in total, each named exactly once. That is a hint, not evidence — what holds up here is the reshuffle among those already visible.

That is where your own website counts: not to become famous, but to be found on the question someone with an intent to buy actually asks. Machine readability is necessary for that and not sufficient — the page also has to answer the question. The region averages 61/100: soundly built, but without the signals an AI draws citable facts from.

3. 45 estates do not exist for AI. That is not a judgement of quality, it is a gap in the data. Someone wanting to buy Rheingau wine who asks ChatGPT gets a handful of names and never learns the other 45 exist.