The Impact of AI on Wine Writing: Will it Be as Big as We Think?
Image by Rawpixel
Does AI signal the death of wine writing? Felicity Carter doesn’t think so.
After ChatGPT was released late last year, there was a sudden rush of writers testing it out. They published the results, which generally ended with the words: “and this article was written by AI!”
The first article was interesting. The second, not so much. ChatGPT, the artificial intelligence chatbot, has a writing style that’s one part soothing to three parts bland.
But still, its emergence prompted people to ask if AI signalled the death of wine writing.
Such AI products are in their infancy. As they develop, they will no doubt have a sweeping impact on publishing and writing.
But what that impact will be still isn’t clear — and it’s much too early to announce the death of wine writing. As it happens, I don’t believe ChatGPT and its cousins will impact traditional wine writing very much at all.
What is wine writing?
Most of the commentary assumed that ‘wine writing’ means ‘tasting notes’. And, indeed, AI can write tasting notes that are indistinguishable from classic wine notes.
The problem — and I know lots of readers will disagree vigorously — is that for many wine consumers, the tasting note isn’t particularly important. What they want is the score.
I was at an industry conference several years ago that discussed scores, and the feeling in the room was that scores reduce complex artisanal products to commodities. Who would dare to score the Mona Lisa out of 100?
Which is true, except the Mona Lisa is not competing on the open market against thousands of other Da Vinci paintings of enigmatic women, of which there are more every year.
There may be no critic working today as important as Robert Parker in his heyday, but at the extreme ends of the market, scores remains an important way to differentiate between the vast numbers of wines offered for sale each year. At the entry level, scores can help with entry in major export markets.
At the fine wine end, an unusual score for an En Primeur release won’t just affect the wine in question, but will have a knock-on effect on the price of older vintages, moving them up and down.
Plus, many consumers absolutely love scores. Open Vivino and watch as wine enthusiasts enter their own scores by the thousands.
As no AI can taste and rate wine, the major critics are safe for the moment.
If you’re reading that thinking, “ah yes, but one day the AI will be able to taste and rate wines,” think again.
In more than 15 years of wine business reporting, I’ve seen numerous tasting-and-rating systems come and go. Some have been based on pseudoscience, like attempts to recommend wines based on DNA, while others have attempted to judge a wine on its chemistry alone. Futile. A wine may have veritable riches of polyphenols and acids to offer, while still tasting awful.
The best system I reported on, which involved heavily-screened tasters sitting and characterising individual wines, was so expensive that it was commercially useless.
For the foreseeable future, the wine market will still need critics to sort through the mountains of bottles, making recommendations and awarding scores.
Wine writing is so much more than that
In any case, most traditional wine writing is feature writing, involving interviews that are combined with a deep knowledge of regions and styles.
AI could probably interview people by email, but it would never ask the unexpected, interesting question. And although it can scan the internet for information on geology and grape varieties, that doesn’t make it reliable. AI works by predictions; it predicts what the next word is likely to be, based on the information it’s been trained on.
When the AI hits a gap in its knowledge, it does something known as “hallucinating”, otherwise known “making it up”.
Ask ChatGPT to write an essay on a topic you know well, and you’ll be struck by the nonsense it produces. Most people seem to think this is just a glitch, and that the AI will get better, but consider this: it’s trained on what’s on the internet, which is mostly marketing material, Wikipedia and anecdotes about cats. The truly high quality material, from scientific journals to the Financial Times, is locked up tightly behind paywalls that are difficult to breach at any scale.
Publications that rushed to use AI learned this to their cost. They had to stop, because the bots were churning out reputation-wrecking nonsense. The only way to use AI safely is to have experts and editors supervise it, which takes work. For most wine publications, it will be easier and cheaper to get an actual wine writer on the job.
There are unusual barriers in wine
The wine media is also protected from the threat of wine-writing AI because of a peculiarity — it doesn’t function like most other media.
Many of the big wine media companies aren’t solely media companies. They’re also tasting event, wine guide, and wine competition companies. The magazines endow the various enterprises with prestige and access — they’re the locomotives driving a train whose carriages are filled with awards, medals, events, and other publications, which in turn help fund the magazines.
Most people think of the wine world as lacking in dynamism but, in fact, a significant section of the wine media made a series of good commercial decisions that have helped shield them from some of the havoc playing out in the rest of the media. And their staff have an unquenchable love of wine that ensures a commitment to quality and fairness. Their medals and certificates are generally much more valuable than the big commercial competitions for a reason.
Not only that, but part of the prestige rests on the calibre of the experts they can attract. This approach means that wine publishers will have almost no interest in replacing their wine writers with AI.
Which isn’t to say that AI won’t be useful, particularly at the commercial end of wine writing. AI is great at taking existing text and chopping it up into social media posts and press releases. It can generate snappy headlines.
There are also startups working on publishing AI-apps to bring down costs and make life easier. One company is working on an AI app that can go through your photo collection and label everything with location, people and dates. Every editor will raise a glass to the first ompany to release this, thereby freeing them from the need to label the 500 winery photos that arrive with names like DSC_001 and DSC_002.
AI might save us from one wine-writing scourge
Ever wondered why wine magazines keep publishing articles about the correct wines to pair with the Thanksgiving turkey? Didn’t we solve that already? Yes, of course we did.
But the way that Google is structured means that wine outlets are obliged to write new versions of these articles, year on year. It works like this: everybody wants their articles to appear at the top of Google’s rankings. To get there, they need to produce authoritative, in-depth content that answers the questions that consumers are asking Google.
Since lots of consumers ask Google what they should serve with Thanksgiving dinner, wine companies must employ lots of writers to create articles answering this question, hoping that Google will put them at the top.
It’s called Search Engine Optimisation (SEO) copy, and you have to keep it fresh, or you’ll fall down the rankings.
I once scrolled to the bottom of the Champagne page of a big German online retailer, to find a huge article, in teeny-tiny letters, about Champagne. It wasn’t there for a human to read — it was there to prove to Google how knowledgeable their site was. Later I scrolled down the Champagne page of a major American retailer, and found the same thing. It’s an extraordinary waste of time and talent, and writing to algorithm is extremely damaging to emerging wine writers.
AI could potentially change all this. Instead of searches bringing up relevant articles, AI might simply give us the answer we’re looking for, in chatty, accessible language. It will probably trawl Reddit and eavesdrop on all the people discussing what they’re going to serve for Thanksgiving, and will aggregate their answers and give them to you.
Other ways that AI could help wine
There are other ways that AI could help the wine trade, and the fine wine trade in particular. First, it can summarise and translate research papers. Right now, there are very few bridges between important wine and viticultural research and the people who could use it. AI could bridge this gap, to the benefit of everybody.
That’s only the start of the things it might do. Probably there are products on the way that we haven’t even realised we needed.
In the end, however, wine is real and tangible. You need humans to taste it and rate it, and humans to tell its stories. AI can’t do that, and probably won’t be able to in the near future.
Wine will be fine, and so will wine writing.
P.S. I asked ChatGPT to find the logical flaws in this essay and it told me that AI will eventually be able to taste and rate wines just as well as humans do. Let’s see.