Ask any AI tool to describe an apartment and you'll get something like this: "Bright apartment in an unbeatable location, perfect for families, with every amenity just steps away." You've read it many times. So have your customers, which is why they skip straight to the photos.
We don't think the tool is to blame. We think it's what we give it.
Here are a few things that seem true to us about why AI-written text all sounds so alike, and about what makes it sound like the business that signs it.
AI without context writes the average of everything it has read
A language model, when it knows nothing else, picks the most likely thing. And the most likely thing is whatever has been written most often: safe adjectives, tidy sentences, no edges.
Give it a single line – "three-bedroom apartment in Gràcia" – and it fills in the rest with what usually comes next. That's where "bright," "unbeatable" and "don't miss out" come from.
It isn't a bug. It's exactly what it was asked to do.
The problem isn't that an AI wrote it, but that anyone could have
Real estate listings sounded alike long before these tools: "move-in ready," "flooded with light," "ideal for investors." AI didn't invent the cliché. It just made it free.
George Orwell, the writer, described it in 1946 in his essay "Politics and the English Language." Bad modern prose, he said, consists in "gluing together long strips of words which have already been set in order by someone else." There were no machines that wrote, and still it describes a language model without context fairly well.
When a paragraph costs seconds, the generic paragraph stops being worth anything.
Adjectives no one can check are the first thing nobody reads
"Spacious," "modern," "high quality." Nobody knows what they mean, so nobody believes them, and readers learn to skip them.
A number doesn't have that problem. "Spacious" is an opinion. "Ninety square meters" is something you can check at the viewing.
When everyone asks for the same thing the same way, everyone gets the same thing
Hook, three selling points, call to action. That structure works just as well for an apartment, a pair of sneakers or an accounting firm, which is why it says nothing about any of them.
Same with restaurant menus. "Market-fresh cooking" meant something when few places used it. When everyone uses it, it no longer sets anyone apart.
What changes the result isn't a better model, but what you give it first
It's almost always three things. The real product details: the full spec sheet, not a summary. Your earlier writing, especially the pieces that worked. And your tone rules: how you address customers, which words you never use, how long a piece should be.
With that in front of it, the AI stops writing everyone's average and starts writing yours.
A good listing says things a brochure never would
Picture the full sheet for an apartment, with the floor, exposure, remodel, building fees and subway. The text that comes out of it could be this:
"Third floor with an elevator on Carrer de Verdi. Three bedrooms, one with no outside window, currently used as an office. South-facing living room. Kitchen remodeled in 2021; the bathroom wasn't. Fontana subway station a four-minute walk away. Building fees: €85 a month."
It isn't prettier. It's more useful. It tells you what's good and what isn't, and whoever calls already knows what they're coming to see.
Saying what's bad is what makes the good believable.
The boring details are the ones that answer questions
The generic description of a skillet promises "elegant design and professional results." The one built from the product sheet says it's carbon steel, that it works on induction, that it isn't dishwasher safe and that it needs seasoning with oil for the first few uses.
None of those sentences is pretty. Each one answers a question that would otherwise land in your inbox.
A reply to a review gives away right away whether someone who was there wrote it
"We're sorry to hear about your experience and are always working to improve" could sit under any one-star review, for any business. A reply that addresses what actually happened, and names the person who helped, looks nothing like it.
The same goes for a social post or the monthly newsletter. AI can draft them. What it can't do is know what happened, unless someone tells it.
A voice is easier to imitate from examples than from descriptions
If your business has its own way of talking – warmer, a little funny, no jargon – we think the AI picks it up better from your earlier writing than from any explanation you give it.
Ask for "a friendly tone" and you get a friendly tone, anyone's. Show it what you've already written and you get yours.
AI gets things wrong even with the best data
Sometimes it invents a detail that wasn't on the sheet. Sometimes it oversells. Sometimes it says everything right at the wrong moment.
That's why the review isn't optional. It doesn't have to take long either: the heavy lifting – drafting from scratch, adapting to each channel, keeping the tone – is already done. What's left is the part only someone who knows the business can do: check that it's true, change a sentence and decide whether it goes out.
The time you save comes from not writing, not from not reading.
A generic text almost never gets fixed by editing it
If it's missing data, no amount of polishing will give it any.
The fastest fix is usually to start over and give the AI what it didn't have. A generic text is a problem of ingredients, not of the recipe.
None of this is new
Good listings have always had concrete facts, a voice people recognize and someone who read them before they went out. The only thing that has changed is that generic is now free, and what's free is everywhere.
That's how our AI CRM works with product and property listings: it drafts them from your data and your earlier writing, and a person reviews them before they're published. If you want to know whether it makes sense for your business, get in touch. The first call is free, and if you don't need us, we'll tell you.