What to decide before you scrape US car listings

The job is a comparable inventory file. You want active US vehicle listings — auction, classified, dealer — in one schema, on a cadence someone uses before they bid or buy.

Teams often skip the brief and ask for “all the cars.” Then the file mixes trims, drops a source, and a price column nobody can explain. Consultancy writes the output first.

The questions that decide the build

Which channels are in scope this quarter: auctions, classifieds, dealer sites, or a named short list. Which fields must exist on every row (year, make, model, trim, price, mileage, location, listing age). How stale is allowed — a day, a few hours, next morning. What you will do when a page layout changes. Who owns a re-run when a chunk fails.

If you cannot name the grain (one row per listing, per day, per price change), we will not start a pipeline. If a licensed feed already has those fields, we will say so. Scraping is a last resort, not a personality.

When the brief is real, the work looks like the USA car market case study: extract, normalize, refresh, a file a dashboard can load. We will not treat one desk’s source list as a nationwide promise.

This is not the Bayut property coverage job. Buildings and asking rents are a different schema. It is also not the Facebook Marketplace watch — that job is volatile consumer stock and time-to-sold, a different freshness rule.

What this is not

This is not a pricing model and not a buying bot. Those are later, and only if the table is honest. We will not start with “every US site” if you cannot name the ten you already open by hand.

If the constraint is “we need the file,” start from AI consultancy or the scraping service once the columns are written down.

Bring the sheet you already use to compare two cars, even if half the rows are missing. Book a consult or email hello@jamilglobal.com.