Spend an hour reading used car listings across any European portal and a certain kind of paragraph starts to jump off the page.
This beautiful vehicle is the perfect companion for your summer adventures! ✨ With its elegant lines and impressive comfort, it offers a driving experience that will delight the whole family. 🚗 Don’t miss this opportunity!
Two years ago that read as effort. Somebody had written something rather than leaving the field blank. Today it reads as the opposite, and the shift has been fast enough that a lot of dealerships have not noticed it happened. Buyers now see this register constantly, across every category they shop in, and they have learned to skip it.
What a buyer takes from that paragraph now is not “this is a nice car”. It is: nobody here spent thirty seconds on this listing. And if nobody spent thirty seconds on the listing, what are the odds they spent time on the car?
What is actually going wrong
It is worth being precise, because the obvious diagnosis is wrong.
The obvious diagnosis is that AI writes badly. It does not. The systems available today can write extremely well - better than most of us, in most registers, on most subjects. If the output is weightless, it is because the input was.
Here is the input that produced the paragraph above: a make, a model, a year, a mileage figure, a list of equipment codes, and the instruction “write an appealing description”. That is all the system had. It knew nothing about who buys this kind of car in your town, what they worry about, what your dealership is good at, what the previous owner did with it, or why this particular example is worth more than the identical one three listings down.
Given that input, weightless prose about summer adventures is not a failure. It is the only honest answer available. There was nothing else to say, so it said nothing, at length.
Which brings us to the second problem.
Adding is not improving
There is a documented bias in how people approach improvement. A 2021 study in Nature found that when asked to make something better, people systematically overlook the option of taking something away - given a structure to strengthen, we add a support rather than remove the weight.
AI is exceptionally good at adding. Ask for a description and you will get one; ask for a better description and you will get a longer one. Nothing in the interaction pushes toward removal, and the result is listings that grow steadily more elaborate and steadily less useful.
Meanwhile the buyer is scanning twenty listings on a phone at 22:00 and is looking for four things: is this car what I think it is, is there anything wrong with it, is the price sensible, and can I trust these people. Every sentence that does not serve one of those four is a sentence that pushes the answers further down the page.
Four honest sentences beat two elaborate paragraphs, every time, and the elaborate paragraphs cost you more to produce.
| WHAT THE BUYER WANTS TO KNOW | WHAT GENERIC AI COPY SAYS |
|---|---|
| Is this car exactly what the title claims? | "Elegant lines and impressive comfort" |
| Is there anything wrong with it? | Nothing |
| Why is it priced where it is? | Nothing |
| Can I trust this dealership? | It has answered the first three questions with adjectives |
The version that works
None of this is an argument against using AI for listings. It is an argument against using it lazily, which happens to be the only way most people have tried.
The version that works costs about an hour, once. After that it runs on every car you list.
Give it your own best work. Find twenty or thirty descriptions you wrote yourself and were happy with. Not your average ones - your good ones. Give it all of them and ask what they have in common: structure, length, tone, what you always mention, what you never mention. It will find patterns in your own writing that you could not have articulated.
Let it interview you about your shop. Ask it to question you on what actually sells here. Who buys from you. What they ask on the phone before they visit. What objection comes up most. What you say about a car when you genuinely like it. This is the step everyone skips and it is the one that produces everything worth reading in the finished copy.
Give it the knowledge you carry in your head. This is where the real value is, and it is entirely specific to you. You know that on an imported electric car from southern Europe, the first thing a Northern European buyer wants to know is whether it has a heat pump - because plenty of southern-market cars do not, and it changes winter range materially. That single sentence in a listing is worth more than four paragraphs of adjectives, and no general-purpose system will produce it unless someone tells it.
Multiply that across your range. The gearbox variant that matters on one model. The service interval that scares people on another. The trim level that is worth a genuine premium and is routinely confused with the one below it. All of that lives in the heads of your sales team and almost none of it is written down anywhere.
Save it. All of the above should become a reusable profile that applies to every description you generate from now on, not a conversation you have once and lose. Every serious AI system now supports some version of this. Set it up once, refine it when something annoys you.
Then edit. Somebody reads it before it publishes. This takes thirty seconds per car and it is the difference between a listing that sounds like your dealership and one that sounds like everybody’s.
The line this sits on
There is a broader principle underneath this, and listings are just the most visible instance of it.
AI is unambiguously good anywhere its output lands on your own desk - analysis, research, working out where your marketing money should go, understanding your own data. Nobody outside your business is affected if it gets something wrong, and you will catch it before it matters.
The moment its output becomes the customer’s experience of your business, the standard changes completely. That is not a reason to keep it away from customers. Handled properly, customer-facing AI is the strongest use case in this industry, because the alternative is frequently no response at all. But “handled properly” means grounded in your dealership’s real facts rather than plausible generalities, checked before it goes live, and - in the EU, since 2 August 2026 - disclosed as AI where a person is interacting with it directly.
A listing description sits right on that line. It is the first thing a buyer reads about a car and about you. It is not the place to find out what a system produces when you give it nothing to work with.
Frequently asked questions
Should car dealers use AI to write vehicle descriptions? Yes, but not by pasting specifications into a chat window. Set up a reusable profile built from your own best previous descriptions, your knowledge of who buys what in your market, and the specification facts that genuinely affect a buying decision. Then have a person read each one before it publishes.
Can buyers really tell when a listing was written by AI? Increasingly, yes, because they now meet the same register everywhere they shop. The tells are consistent: emotive adjectives without specifics, emoji, a length disproportionate to the information conveyed, and no mention of anything a buyer would actually want to check. What it signals is not that AI was used - it is that nobody looked.
What should a good used car description contain? Whether the car is exactly what the title claims, including trim level and options. Anything a buyer would want to know before travelling. Why the price is where it is. And something specific enough that it could not have been written about any other example of the same model.
How long should a listing description be? Shorter than most AI-generated ones. Buyers scan on phones. Four honest, specific sentences outperform two paragraphs of atmosphere.
Do I have to disclose that a listing was written with AI? Article 50 of the EU AI Act, applicable since 2 August 2026, covers both direct AI interaction with a person and the marking of AI-generated content, and how it applies to listing copy is worth checking with your own advisers rather than assuming. What is not in doubt is the conversational case: where a buyer is corresponding with an AI system, they must be told so clearly at the start.
Sources
- Systematic neglect of subtractive change: Adams, Converse, Hales & Klotz, “People systematically overlook subtractive changes,” Nature, 2021.
- EU AI Act Article 50 transparency obligations, applicable from 2 August 2026: European Commission.
