On 11 September the 67 Pall Mall Global Wine Communicator Awards handed its short-form writing prize, worth £3,000, to a London sommelier called Fikayo Ifaturoti, for a piece titled “Collector Logs, No. 4 | The Stradivarius Violin x 1969 Bollinger VVF”.

Within days the wine writing world was arguing about whether a machine had written it.

We published a piece last week about readers preferring to ask an AI rather than a sommelier. This is the same argument arriving from the opposite direction, and it is more uncomfortable, because the people who write about wine for a living cannot answer the question about their own profession.

What is actually known

The accusation started publicly with the writer Miquel Hudin, who asked what anyone was supposed to make of a major award going to what he called a 100% AI-generated piece. Felicity Carter ran it in Drinks Insider on 18 September under the headline “Did a Chatbot Just Win a £3,000 Writing Prize?”. Alice Feiring, Henry Jeffreys, Tim Atkin and Alder Yarrow all wrote about it inside a week.

The evidence people point at is stylistic. Jeffreys picked out the clipped fragments standing in for paragraphs, the triple constructions like “We’ve scanned it. Measured it. Analysed it.”, and the “Not because we lack…” framing that AI models reach for constantly. He called the overall effect arid and impersonal. Feiring’s version of the same complaint is sharper: the piece had nothing sensory and nothing personal in it, and never told the reader why that Stradivarius analogy was chosen, what the Bollinger tasted like, or what a Stradivarius sounds like.

Then the detectors were run. This is where it falls apart as proof.

Four detectors returned 100% AI. Two returned partial. Two returned human. Pangram, the one most often cited in these rows, came back fully machine every time. Eight tools, three different answers.

Why that is not evidence

We are not going to tell you Fikayo Ifaturoti used a chatbot, because we do not know, and neither does anyone else who has written about this.

AI detectors are not forensic instruments. They are statistical guesses about how predictable a text is, and they are known to misfire on writing that is formal, non-native, heavily edited, or simply tidy. A writer who drafts cleanly and cuts hard produces exactly the low-variance prose these tools score as synthetic.

The most useful thing anyone said in the whole row came from Elaine Chukan Brown, one of the judges, who warned against leaning on detectors at all: the hallmarks we have decided are AI hallmarks exist, she pointed out, because real people write that way. The models learned them from us.

Richard Hemming, who founded the awards, says he has it on good faith that the work was not entirely AI-generated, describes the style as a spare one that younger writers use, and argues the Stradivarius comparison shows a mind at work rather than a model. He has also said the awards are writing an AI policy for next year.

That last part is the admission. An awards body that is confident in this year’s result does not need a policy for next year.

The part nobody is going to enjoy

Here is the thing that makes this more than a spat.

The reason a machine could plausibly win a wine writing prize is that a great deal of wine writing already reads like a machine wrote it. Feiring said it plainly: too much of it is promotional copy dressed up as an article. Tasting notes assembled from the same forty adjectives. Regional profiles that could be swapped between appellations with a find-and-replace. Sentences that sound authoritative and mean nothing.

If your genre’s house style is fluent, confident, symmetrical and empty, you have built the perfect training set for the thing that is now beating you, and you have also destroyed your own ability to spot it. You cannot detect a chatbot by looking for soullessness in a field that rewards it.

This is why we do things the way we do here. Antoine sits in front of a winemaker with a microphone and asks what went wrong in 2003. It is slower, it is messier, and some of the best moments are somebody laughing at a stupid question. None of that is generatable, because none of it existed until two people were in a room.

What we would actually like to see

Not a witch hunt. A named individual is currently carrying the reputational cost of an institutional failure, which is backwards.

  • Judges should be able to ask. Feiring’s real charge is not that the judges were fooled, it is that nobody questioned anything. A simple “tell us how you wrote this” would have settled it in an afternoon.
  • Publish the policy now, not next year. Every entrant in 2026 deserves to know what the rule was.
  • Reward the stuff a model cannot fake. The interview nobody else got. The estate you actually walked. The number you checked and found was wrong.

That last one is not a hypothetical. We ran a piece yesterday about a widely reported Champagne statistic that turned out not to be real. A chatbot would have repeated it, confidently, in three symmetrical sentences.

Where this lands

The awards will survive. Wine writing probably will too.

But the honest summary of September 2026 is that one of the industry’s flagship prizes produced a result its own judges, founder and detection tools cannot agree on, and the only certainty is that nobody can prove anything. That is not a technology problem. That is what happens when a whole field spends thirty years writing in a voice that does not belong to anybody in particular.