Once around the car: a listing from a walk-around

The listing form offers “Start with a walk-around”. It has three steps: take the shots, tell us about the car, check the result. What can be seen is recognised automatically from the shots; the statements you explicitly made are taken from your text. What comes out is a draft, not a finished listing.

The advantage is not that an AI knows better what car you have. It is that you do the work in the order it happens at the car anyway: walk round once, take pictures, and say what you know about the car as you go. The form then fills itself from what you confirm.

Two things the walk-around deliberately does not do. It does not suggest a price; you set that yourself. And it applies nothing without your tick. If the AI feature is unavailable or has reached its daily limit, everything still works without it, through the fields as usual.

The walk-around takes you through eight angles. On a phone, tapping an angle opens the camera straight away; you can also pick photos you already have. How to get light and background right is covered in the photo guide for listings. Here the point is what each image should do for the draft and for the buyer.

If an angle is missing after the analysis, the result says “still missing — take a quick photo?”. That is a suggestion, not an obligation. But a listing without the cockpit or the rear leaves open exactly the questions the walk-around is meant to settle.

Instead of individual photos you can pick a short video in which you walk once around the car. The video stays on your device: the browser itself picks out the usable stills, discards dark and blurred ones, and only those stills are used. The only images uploaded later are the ones you take over as listing photos.

The video is more convenient; the photos are better. A still taken from movement is rarely as sharp as a photo you stood still for, and the odometer is hardly legible in passing. For the draft a video is often enough; for the images a buyer will later zoom into, photos are worth it. You can mix the two: the video for the overview, then individual photos of the cockpit, wheels and engine bay.

If you film, walk slowly, pause briefly at each corner and film in daylight. If your browser cannot open the video or finds no sharp, bright stills in it, the page says so and offers photos instead. Only a limited number of shots is analysed anyway, the first in the list, so several images of the same angle add little to the draft.

In the second step you talk about the car, typed or, on a phone, dictated through the keyboard's microphone. There is deliberately no audio recording; what counts is what stands there as text. The field asks for what no photo shows: previous owners, servicing, accidents, the next roadworthiness test, equipment.

Good sentences are short and carry one statement each: “One owner before me, I bought the car in 2019.” “Always serviced at the same workshop, invoices available.” “Next test due May 2027.” “Tow bar and auxiliary heater.” Sentences like these can be quoted one by one, and that is exactly what matters in the next step.

Leave out whatever you don't know. The hint next to the field says so too, and it matters more than it sounds: “probably accident-free” or “the cambelt should have been done” is harmless in conversation, but in a listing it is a statement you have to stand behind. E-mail addresses and phone numbers do not belong in there either. They are removed before the analysis as far as they can be recognised, but it is simpler not to write them at all.

The shots and the text go to an external AI service for analysis; Carvexia only stores what you apply to the listing. From the images comes what is visible: make and model, a variant only if a badge shows it, body type, colour, visible equipment and a few factual sentences of description. The year often appears only as the span of the model generation; you then choose the exact year yourself. The draft only takes the mileage from a legible odometer in kilometres, never from your text.

Your text yields a different kind of statement: accident-free or not, the number of previous owners, the service history, the next roadworthiness test, remaining warranty and the equipment you mention. Each of these only appears together with the phrase it came from — “You said: …” — and that phrase must appear word for word in your text. If there is no verbatim source, the statement is dropped.

The reason goes beyond the technology: what the listing says are your statements. An analysis that turned “usually serviced at the brand dealer” into “full service history” would have put a promise in your mouth you never made. It is meant to add nothing and infer nothing, and because every statement comes with its quote, you can see at once whether the sentence really says what was made of it.

In the result, every item has a tick box. Whatever was recognised automatically from the images is preselected and labelled as such; the statements from your text are not, and you confirm each of them yourself. Only ticked items are applied. Compare every automatically recognised line with the car and the registration document: a variant or a colour can be off, and “recognised automatically” does not mean “correct”.

Two fields deserve particular attention: accident-free and service history. In the form below the draft they stay on “Not stated” until you confirm a statement from your text or choose “Yes” or “No” yourself — check them there, whatever the walk-around produced. They are the two statements most often argued over after a sale, and only the person who drove and maintained the car knows the answer.

If you wish, the images of the matched angles are added to the listing photos in that order; if there are none yet, the first becomes the cover. Whatever is still missing, such as price, gearbox or a fuller text, you add as usual. Before you publish, read the listing once the way a buyer reads it. If it says anything you would not say yourself, take it out.