This site uses PostHog and Google Analytics (with IP anonymization) to understand how visitors use the service. No advertising cookies. Privacy Policy

TourismIntel
Join free →
All Officina Turistica posts
Translated from Italian · Officina Turistica

It's not a hotel, it's an experience. (Stop.)

Mirko LalliMay 14, 2026AIDestinazioniHospitalityMarketing

The rhetorical pattern that outs AI in every piece of tourism marketing copy

Open LinkedIn right now. Scroll through three recent posts from Italian DMOs or hotel groups with more than fifty properties. Count these phrases:

"It's not just a destination, it's an experience."

"We don't sell rooms, we sell emotions."

"This is no longer tourism, it's an authentic connection with the territory."

Three for three. Always the same structure. Always the same rhythm. Always the same AI.

It's called negative parallelism. It's a perfectly legitimate rhetorical device, used by Cicero, by Manzoni, by the great copywriters of Madison Avenue. And it's also the number-one pattern that every language model produces automatically, dozens of times a day, in every piece of text we ask it to write.

When you find it in your copy, the diagnosis is simple: the text is (90% of the time) machine-written. And more importantly: it sounds like machine-written text even to readers who know nothing about AI.

The numbers behind a silent epidemic

Barron's actually did the research. They took the AlphaSense database — press releases, SEC filings, earnings call transcripts from Fortune 500 companies — and searched for the formula "it's not just a ___, it's a ___".

The results:

  • 49 documents in 2023
  • 100 in 2024
  • 208 in 2025

Doubled. Then doubled again. In two years.

And we're talking about the most expensive documents on the planet. Documents on which companies pay law firms $1,500 an hour to check every comma, and six-figure in-house communications agencies to polish the tone. Published with the CEO's name underneath.

Cisco, Accenture, McKinsey, Workday, Microsoft. Even this year's shareholder letter signed by Satya Nadella: "When Bill founded Microsoft, he envisioned not just a software company, but a software factory." Right…

Now make the mental leap that concerns us. Think about the copy for a regional Italian tourism consortium. No editorial budget. No in-house editor. A junior who opens ChatGPT and types "Write me the launch post for the autumn-winter campaign for territory X."

What comes out? Exactly what you're imagining. Exactly how you pictured it.

If Microsoft doesn't notice that its CEO is publishing a phrase generated by an LLM's pattern matching, a DMO managing communications for 18 territories doesn't stand a chance.

Why all LLMs write like this

The technical reason is trivial. Negative parallelism is a low-cost rhetorical flourish. It creates the illusion of a revelation: first I tell you what something is NOT (negative frame), then I tell you what it really is (revelation). It sounds like a discovery, like depth, like intelligence explaining the world to you.

It costs almost nothing to produce. The first term can be almost anything — as long as the second is an upgrade. "It's not X, it's Y", where Y is just X framed more aspirationally.

Language models are optimised to produce text that pleases. And statistically, negative parallelism is one of the most "pleasing" constructs in the training corpus — which is full of marketing texts, TED talks, self-help books, and American copywriting from the last twenty years. So the models reproduce it. They reproduce it too much. They reproduce it everywhere, regardless of context.

When the context is Italian tourism — where copy is already naturally loaded with aspiration and emotion — the result is devastating. Every post starts to sound the same. Every boutique hotel newsletter becomes interchangeable with every other boutique hotel newsletter. Every destination's tagline starts to be swappable with that of a destination 2,000 kilometres away.

The 15 forms of negative parallelism

The obvious forms. You can spot these from a distance:

  • "It's not X. It's Y."
  • "Not just X, but also Y."
  • "It's not about X. It's about Y."
  • "Forget X. Think Y."
  • "Less X, more Y."
  • "X? No. Y."
  • "X is dead. Y is the future."
  • "The real question isn't X. It's Y."
  • "You don't need X. You need Y."
  • "Stop thinking about X. Start thinking about Y."
  • "X is overrated. Y is what matters."

The sneaky forms. Same structure, softer dress:

  • "Although X may seem…"
  • "At first glance X…"
  • "Most people think that X…"
  • "While X captures attention, Y…"
  • "Everyone's talking about X. But actually Y…"

And then there are the formulas that in Italian tourism have become an involuntary meme:

  • "It's not a destination, it's a lifestyle."
  • "Not tourists anymore, but conscious travellers."
  • "We don't sell products, we tell stories."
  • "It's not hospitality, it's authentic welcome."
  • "It's not tourism marketing, it's community building."

Do you remember writing one of these? I know. I wrote them too, before 2024. When we wrote them by hand, they felt powerful.

Why in Italian tourism it does more damage

There's a structural reason why this pattern hits us harder in Italy than elsewhere.

Italian tourism copywriting is historically sentimental. Selling Italy has always meant selling emotion, authenticity, slowness, warmth, family, tradition. Aspirational language comes naturally to us. It's our cultural competitive advantage.

Now try to imagine what happens when you layer Italy's natural aspiration on top of a language model that already overproduces aspiration. The result is a bubble of nothing. Texts that talk about emotions without describing anything. Pages that promise "unforgettable experiences" without ever saying what the traveller will actually do, where they'll sleep, what they'll eat, how much they'll pay, how far they'll walk.

The problem with negative parallelism in tourism isn't just aesthetic, it's economic. Italian destination content is converging toward a single AI-written average text, indistinguishable from one territory to the next. Tuscia indistinguishable from the Langhe. Cinque Terre indistinguishable from the Amalfi Coast. Trentino indistinguishable from Val d'Aosta.

For the new conversational tourism discovery systems — the ones we'll all be using in 18 months, when "plan me a 7-day romantic trip to Italy" becomes the baseline query — content that doesn't distinguish your territory from the 27 surrounding ones is invisible. Not distinctive. Not recommendable. Not citable.

The AI generating your copy today is the same AI that will need to recommend your territory tomorrow. If your copy doesn't inform it, it won't see you.

The fix, in practice

The solution (to start) is simple: a style instruction file, loaded directly into whichever AI tool you use every day.

I call mine voce-anti-ai.md (I'm sharing my file below 👇). It's a single Markdown file. It contains four things:

  1. The list of banned patterns (negative parallelism and its 15 variants)
  2. The list of banned vocabulary (the 100 AI-speak words like "valorizzare" (to enhance), "innovativo" (innovative), "all'avanguardia" (cutting-edge), "ecosistema" (ecosystem), "in un panorama sempre più…" (in an increasingly…))
  3. Rhythm rules (short paragraphs, variable length, no em dashes)
  4. An anti-overfitting guide, to stop the tool from overcorrecting in the other direction and producing rigid, dry, telegraphic texts

When I work with Claude (it's my main tool, but the same applies to ChatGPT with Projects or any other model that lets you load context files), I always keep it in the project folder. Before every response, the AI reads that file.

When I want to be sure a text doesn't sound like AI, one line is enough:

Run an audit of this text against voce-anti-ai.md.

It re-reads, identifies suspicious phrases, rewrites them. It's a strange thing to watch — an AI correcting another AI (sometimes itself). It works beautifully.

My version has a dedicated section for AI-generated Italian tourism vocabulary: the toxic words we should ban. "Valorizzare", "raccontare il territorio" (telling the story of the territory), "esperienze immersive" (immersive experiences), "borgo autentico" (authentic village), "tradizione che incontra l'innovazione" (tradition meets innovation), "panorama mozzafiato" (breathtaking view), "atmosfera unica" (unique atmosphere), "patrimonio inestimabile" (priceless heritage). Each one, on its own, is still salvageable. All together, in the same text, they're the AI's signature.

The other eight signals

Negative parallelism is the most visible signal. It's the first of a family. For completeness, here are the other patterns that flag AI in a text:

1. The rule of three. AI lists three things when it doesn't know what to say. "Speed, efficiency and innovation." Use two if two is enough. Or four, if you need four. Or the one that actually matters.

2. Rhetorical inflation. "A crucial moment." "An epochal turning point." "A seismic shift." Say what happened. Let the reader judge the magnitude.

3. The participial trap. AI attaches present participles to fake depth: "highlighting its importance", "underlining its significance", "contributing to a rich mosaic of…". Delete them. If the analysis matters, make it with a full sentence and a real argument.

4. False ranges. "From ancient traditions to the most modern innovations." Sounds impressive. Means nothing. If you can't name what falls in between your X and your Y, the range is fake.

5. Elegant variation. AI renames the same thing four times out of fear of repetition. "Claude" becomes "the assistant", then "the model", then "the chatbot". If it's Claude, write Claude.

6. Fleeing from the verb to be. AI never uses "is". It uses "configures itself as", "represents", "constitutes", "positions itself as", "reveals itself to be". Use "is".

7. Title Case headings. AI writes "Key Considerations For Adoption". Humans write "Key considerations for adoption". Sentence case sounds human.

8. Metronome rhythm. Every sentence of medium length. Every paragraph three sentences. No variation. Real writing breathes irregularly. Short. Then a fragment. Then a thirty-word sentence that earns its length because two short sentences introduced it.

The file lives and changes

AI patterns, just like language, change over time.

In English, for instance: "Unlock" passed in 2024. Today it's an own goal. "Harness" and "leverage" are following the same trajectory. By 2027 there will be a list of words we don't yet see as AI-generated, because we're still using them ourselves.

So the voce-anti-ai.md file should be treated as a living document. Every three months:

  1. Re-read the last ten texts you've published.
  2. Circle the words and structures that, on a second reading, sound like machine output.
  3. Add them to the banned list.
  4. Remove the ones that no longer bother you.

The file is your taste in text form. Taste is saying no to 99% of what AI produces, yes to the 1% that sounds like you. The file captures the no.

📌 You can download my anti-AI file at this link (put it to good use!):
   https://mkl.it/voce-anti-ai

A final note, on responsibility

An important clarification: using AI to write is absolutely fine. I use it every day — though in practice more to correct what I've already written than to generate everything from scratch. The point is responsibility for the final text. What you publish under your name (or under the name of your organisation, your DMO, your hotel, your consortium) is yours. Full stop.

If in 2026 you're still publishing "It's not a destination, it's an experience", the problem is simple: you've delegated your taste. You used AI but you didn't use yourself afterwards. You didn't re-read.

Taste, fortunately, cannot be delegated.

Open the next text you have in draft. Search for "not only", "it's not", "it's not about". Find them. Delete them. Rewrite the thought in the positive on its own, without the crutch of the negated phrase before it.

Already sounds better, doesn't it?

Data source: Barron's via AlphaSense, January 2026 analysis of SEC filings, press releases and Fortune 500 earnings call transcripts.

Originally published in Italian by Mirko Lalli on Officina Turistica. Translation preserves the author's original voice.

Read the original (Italian)