The Social Media Hotelier
← Letters

Beigification: The AI Tell Guests Feel Before They Can Name It

The Social Media Hotelier7 min read

A few weeks back I sat down with Heather Bayer for her podcast, Vacation Rental Success. If her name’s new to you, it’s worth knowing: this was episode 673, and she’s one of the people short-term rental hosts always trusts when it comes to STRs, marketing and bookings. We got talking about AI, and landed on something I haven’t stopped thinking about since.

[Full conversation linked at the bottom.]

You can tell within an instant whether a host wrote their own caption or asked AI to do it.

There’s nothing grammatically wrong with the AI version.

There’s just nothing really in it.

Four years ago, we were handed artificial intelligence: a weapon of mass execution. For a while it felt like cheating. Type a question, get an answer. Type another question, get ten captions. Simples.

Then something unexpected happened. Everything started sounding the same.

Scroll through enough short-term rental accounts right now and you’ll find plenty of it: the tell you feel before you can name it. A property description that could belong to the cottage three doors down. A caption that’s grammatically perfect yet says nothing of substance.

I stood on a stage in Brighton a few weeks back and said the line out loud because, despite being somewhat controversial, it needed broadcasting:

“AI didn’t break marketing.

It just exposed the people who were never really marketing in the first place.”

Publishing posts isn’t marketing. Neither is writing a blog, having a website, or getting written up in a magazine, however nice that last one feels.

Marketing is what moves one specific person from “I might visit” to “I want to stay there.” That’s the whole game, and it was always the game, long before AI showed up.

Everything else is just activity and production wearing a marketing costume, and underneath that costume there’s nothing but an imposter. AI is now remarkably good at putting on that costume… at a speed and volume nobody’s ever had access to before. Which means the busywork that used to disguise a lack of strategy doesn’t disguise anything any more. It just produces more of it, faster.

I want to show you the gap I see every week in my coaching work, between hosts who are using AI well and hosts who only think they are, and hand you the actual structure I use to sit on the right side of that gap. It’s got a military name because I’ve spent the last couple of years training to be an officer, and because it turns out a briefing room and a content calendar have more in common than most people might think.

Beigification Is a Choice, Not a Side Effect

Here’s the thing about AI content: it isn’t bad because the model is bad. It’s bad because it was never told anything specific enough to be good.

An AI is, at its core, a machine for predicting the next most likely word. Left alone, “most likely” always drifts toward the middle, the safest, most averaged version of whatever you asked for.

I’ve started calling this Beigification, because that’s exactly what it looks like: beige. Colour draining out of everything until every account and every caption looks like it was issued from the same beige production line. You can spot it in seconds once you know what you’re looking for. The em dashes. The “it’s not this, it’s that” construction. Everything arriving in triplicate. It’s polished and grammatically spotless, and by the time you finish reading it you couldn’t repeat a single thing it actually said.

None of that is a flaw you can patch. It’s what happens by default when you hand a prediction machine a desire instead of a specific brief.

Using AI less won’t fix this. Doing the part of the job that can’t be delegated, before you ever ask it to write a word, will. Think of the difference between a ping-pong ball and a golf ball. They’re the same shape, roughly the same size, even a similar colour. The only difference is what’s inside: one’s packed with weight, one’s hollow, and that’s the only thing that decides how far either one travels once you actually hit it. A polished turd, as I said to Heather Bayer on her podcast just the other day, is still a turd… even if you roll it in glitter first.

Step 1: Define one person, in far more detail than you think is necessary.

Think of it like a paint-by-numbers kit from a charity shop where half the pots have dried up: leave enough colours out and you can just about tell it’s a tiger, but not whether it’s walking through a jungle or crouched in a cave.

Not “adults 35 to 55, mid-income.” One person: how many cars, where they grocery shop, what colour hair. Every extra dot is what turns a shape into a scene AI can actually write toward. This is exactly what the Perfect Potential Guest Mega-Prompt walks hosts through step by step, because most people stall at “adults 35 to 55” and never get any further on their own.

Step 2: Brief it like a second-in-command, not a wish.

“Write me 10 captions” is a wish. I use SMERLAC, straight from military planning:

  • Situation
  • Mission
  • Execution
  • Resources
  • Limitations
  • Ask questions
  • Check understanding

The last two matter most. Ask questions means exactly that: before getting it to execute on your instructions get the model to ask you clarifying questions until it’s at 95% understanding & not guessing.

Check understanding is the habit that catches drift, because a context window works like a column filling up. Once it’s full, new information starts pushing the oldest details out the top, and the model forgets what it knew first, not what you just told it. Get the brief right, and the model genuinely earns the next eighty percent. That’s not the risky part, once it’s been briefed like this instead of wished at.

Here’s the obvious objection: doing steps one and two properly takes real time.

If the whole promise of AI was “this is faster,” spending twenty minutes writing a proper brief before you’ve typed a single caption can feel like you’ve defeated the point. I’d have agreed with that a couple of years ago. What changed my mind wasn’t a productivity argument. It was watching how much time actually gets spent downstream rewriting beige output that never should have gone out, or worse, publishing it and never quite figuring out why it didn’t land. The 20 minutes doesn’t disappear. It just moves to the front of the job, where it’s cheaper.

Step 3: Feed it your idiolect.

Heather told me that she’d dug out the handwritten manuscript of a book she wrote back in 2005, unedited, and fed the whole thing in. It changed everything the model gave her back afterward, because for the first time it wasn’t guessing at her voice from a brief. It had actually heard it.

My own version of that is equally substantial: I spent the first part of this year mining 350-plus hours of my own coaching calls for the patterns and phrases that are actually mine, filtering out anything that belongs to a client rather than to me, and feeding the model that distilled version instead of a raw transcript.

It’s the actual voice-and-pattern library everything I write pulls from now. You don’t need to go to such lengths yourself. Your idiolect is sitting in your old texts and voice notes right now: the specific phrases only you reach for. Feed it that, and the AI stops writing generic slop and starts writing specifically like you.

Step 4: Run the commander’s inspection before anything goes out.

Hunt the beige tells: em dashes, rule of three, polished sentences that don’t actually say anything. If you find them, it wasn’t briefed well enough. Go back to step one or two. A golf ball doesn’t get made by polishing a ping-pong ball harder. None of these tells arrive with sirens.

As AI would be prone to say, “things always seem to be happening [insert AI’s favourite hushed adverb here],” right up until the moment they’ve already gone wrong. [See what I did there?]

How Your Content Marches Into an Asparagus Field

I was leading a night exercise at Royal Air Force College Cranwell, tired and wet, the better part of a day and a half without sleep, and had to get my team to a defensive position before first light. I made a quick decision, told everyone where we were headed and moved everyone fast and silent… textbook execution.

Then the mist cleared and we were dug into the middle of somebody’s asparagus field. A very confused woman walking her dog didn't stay for the breakfast tea offered by our squad of armed service personnel in full combats.

We hadn’t failed because we moved too fast. We had failed because we started moving out before I gave my troops a solid mission-brief. Ten AI-perfect captions aimed at nobody in particular is the same mistake: a beautifully executed march into somebody’s asparagus field.

Speed was never the problem. Direction was.

The full conversation with Heather, including her own confession about never quite cracking Instagram, is here: VRS673, When the Work Comes First, the Magic Follows

Get Letters in your inbox

Keep reading