Coches.net, Milanuncios, Wallapop and AutoScout24 all carry Spanish car listings. For data work the question is not which has the most cars — it is how much structure each one publishes, because structure you do not have to infer is the only structure you can trust.
The criteria that actually matter
- Is the data server-rendered? Decides whether you need a browser, which decides your cost per row.
- How structured are the fields? A numeric mileage field beats mileage inside a free-text description by more than it sounds.
- Is there a published valuation? Rare, and it removes a whole modelling step.
- What is the seller mix? Determines how much personal data you are handling whether you wanted to or not.
- How specific are the crawl rules? Specific rules are an asset: they tell you what is allowed instead of leaving you to guess.
Inventory size is sixth at best, and usually irrelevant — you rarely need all of it.
Structure: what the site publishes for machines
Specialist automotive marketplaces model a car as a car: make, model, year, mileage, power, fuel type and badge are separate typed fields because the site's own search facets need them to be. That is why coches.net can deliver 55 documented fields from a search response.
General classifieds model a car as a listing with a title, a price and a photo. The vehicle attributes are in the description, which means every field you want is an extraction with an error rate — and errors in inferred fields are invisible until they are load-bearing.
Valuation fields
Coches.net publishes its own market-average estimate and a rank against it on 77.6% of listings. That is unusual, and it changes what is possible without a model of your own — including where the field's limits are.
If a source does not publish one, you are building a valuation model before you can answer the first interesting question. Sometimes that is the right investment. It is always a cost.
Seller mix and the personal-data consequence
This is the criterion people skip and later regret. A marketplace weighted toward private sellers is a marketplace where a raw scrape is full of individuals' personal data — names, mobile numbers, sometimes locations precise enough to identify a household.
That is a design problem, not a footnote. On coches.net, about 40% of listings are private, which is why the Actor never returns their phone numbers — Article 14. On a general classifieds platform the proportion is higher still, and the same decision has to be made before the first row is stored rather than after.
Crawl rules as a feature
Coches.net's robots.txt enumerates disallowed pagination values individually from pg=7 to pg=69. Read as an obstacle that is annoying. Read as a specification it is a gift: it tells you the permitted depth exactly, which tells you the discovery strategy has to be facet expansion — robots.txt as a specification.
A site with vague rules does not give you more freedom. It gives you less information, and every decision you make becomes one you cannot justify by pointing at the site's own file.
Where that leaves each source
For structured analysis of the Spanish used-car market — depreciation, regional pricing, badge composition, below-market screens — a specialist marketplace with server-rendered typed fields, a published valuation and explicit crawl rules is the cheapest source to work with reliably. That is the case coches.net happens to fit, and it is why it is the one covered here.
General classifieds are the better source for one specific thing: the segment of the market that does not appear on specialist sites at all, which is mostly cheap older cars sold privately. Different question, different tool, and considerably more personal data to be careful with.
Cross-border sources are worth a look if your question is European rather than Spanish, with the caveat that a pan-European site's Spanish inventory is not representative of the Spanish market — it is representative of the Spanish inventory that a pan-European site attracted.
