The A Eye in Your Hotel Bed

Is that online hotel review written by a human or a soulless dark machine?

Tucked into a quiet Bethesda corner, Pisco y Nazca pulses with a sultry, deliberate rhythm—like a secret whispered between lovers over a ceviche that bites back. It doesn’t shout. It doesn’t need to. You walk in and immediately the air thickens with lime and smoke, with stories told through spice and fish pulled fresh from the sea, thousands of miles away.

The Ceviche Nikkei? A bright, defiant hymn to fusion. The tuna, lacquered with soy, kissed by leche de tigre, slides down like a memory you never knew you had—sharp, tender, unapologetic. And the lomo saltado? God help you if you’ve ever mistaken meat and potatoes for pedestrian fare. Here, it’s a fever dream of umami, flame, and precision. The pisco sours—yes, plural, you’ll need more than one—are as dangerous as they are delicate, frothy with regret and citrus.

There is joy here, but it’s not loud. It’s confident. It’s the kind of place where Fisher would lean in, lips stained with rocoto, murmuring that hunger is best when it’s slightly civilized. Bourdain would clink a glass, smirk, and say, This… this is what we’re living for.

Pisco y Nazca doesn’t just feed you. It seduces you.

Well I don’t know if you think that review is written well or not — I find it self-infatuated and arch — but you won’t hurt my feelings either way. It was written by ChatGPT.

I told it to “Write a 200-word review of Pisco y Nazca in Bethesda, Maryland, in a style that mixes those of Anthony Bourdain and M.F.K. Fisher.” (Weird request, I know. They are both my food writing role models.) The above is what came out. Why the hell it referred in the text to the authors whose styles I told it to emulate I have no idea. ChatGPT can be a real dipshit.

It can also be wildly wrong. The restaurant’s on a busy corner, not a quiet one. The pisco sours are good, but carefully titrated and not “dangerous,” if that’s meant to convey that they pack a big boozy punch. You can watch the barkeeps measuring out the drinks behind the long bar: 1.5 ounces of Pisco is all you’re gettin’ pal.

I think “dangerous” was just an inane AI flourish to make things sound Bourdanian.

AI and I

I’ve spared the kind readers of this newsletter of the curious fact that I’ve become something of an expert in generative AI in the last year.

My final year of my day job — with the federal government (!) — was spent on an AI program at my agency, helping develop a ChatGPT-like thing and getting employees to actually use it. (I took one of those crazy Elon Musk “fork in the road” buyout things. Another story, but a ripe one.)

Anyhow, I’ve lately been reporting on the effect AI is having on the travel and hospitality industry for a couple of publications. Frommer’s, in a lapse of judgment, published my story about how you can tell when an online review of, say, a hotel has been written by AI. The story’s here.

The image for the Frommer’s story, titled “How to Spot Fake Reviews Written by AI, According to an AI Professional”

Sniffing out AI fakes isn’t really that hard.

• They often appear under the profiles of people with suspect names (Norman7573) who have only one or two reviews to their name.

• They often are full of glowing generalities, lacking in specific details, and written with suspiciously good grammar and spelling.

• Often they use dorky 7th-grade essayisms like “in summary,” or “therefore.”

The Five-Star Arms Race

But as the review of Pisco y Nazca above shows, you can direct these tools to write some pretty convincing reviews if you give them pointed directions and know a bit about what you’re asking for.

To be clear, though, none of the fraudsters who churn out fake AI reviews for money — generally offshore companies that produce them for unscrupulous hospitality firms that are failing at the review game played honestly — can afford to create the sort of bespoke oddities I coaxed from the dark machine. They need to manufacture a whole bunch of 5-star customer testimonials quickly and get on to the next client.

But the technology is getting better by the minute.

Companies like Google and Tripadvisor are investing a lot of money — in AI technology, of course — to identify and scrub out fake AI reviews. The fraudsters are quickly learning ways, using AI of course, to overcome those defenses, setting up an arms race of AI technologies that determines whether that boutique hotel in San Luis Obispo gets 3.9 stars or 4.8, and whether its service is “OMG the best ever!” or “I couldn’t get a clean towel for three days.”

This may seem trivial, but billions of dollars are spent based at least partly on online reviews. A consulting firm estimated that each household spends an extra $3,000 per year buying stuff based on fake reviews, when they could have had better or similar stuff for less. That may be a high and self-interested estimate, but it’s not out of line to think that you and I have wasted a good bit of money based on bum steers we’ve gotten from fishy online reviews.

So what are you supposed to do? Use the handy tips in my article! Or if you don’t want to follow that link, just abide by the wisdom of this haiku, which ChatGPT generated when I asked it to summarize my article in the format of that ancient Japanese poetic form:

Reviews lack detail
Profiles bare, language cliché
Trust your instincts, friend

I have no idea whether that’s gold or garbage. That’s the problem with this stuff. It hollows your skull, slowly.

Eat This Tip

If you’re just starting to play around with AI, I recommend starting with Perplexity.ai, which is an AI search engine, so its results can be accurate up to the minute. This is not true for ChatGPT, Claude, and other popular chatbots, whose brains are sealed on an earlier date. Perplexity’s free version is pretty good.