Working with job data

What a job data API actually gives you

The pitch is “millions of listings”. The work is in the fields nobody advertises.

Every provider in this market leads with a count: millions of postings, hundreds of thousands of companies, dozens of sources. The count is the least useful thing about the product, because raw postings are cheap and the hard parts are elsewhere.

What you are actually buying

  1. Collection. Someone runs the crawlers, handles the blocks, and keeps up when a source changes its markup. This is continuous work, not a one-off.
  2. Deduplication. One role cross-posted to five boards must become one record. Without this your counts are inflated by a factor you cannot estimate.
  3. Normalisation. Free-text locations into city, region, country. Salary strings into a number, a currency and a period. Titles into something groupable.
  4. Expiry. Knowing when a listing came down is harder than knowing when it went up, and more valuable.
  5. A stable schema. Fields that mean the same thing next quarter.

The fields that separate providers

FieldWhy it is hardWhy it matters
Normalised salaryFree text in dozens of formats and currenciesWithout it you cannot compute anything
Deduplication groupSame role, different wording, different boardsEvery count depends on it
First seen / last seenRequires storing history, not snapshotsThe only way to compute posting age
Posted by agencyRarely stated; must be inferredChanges what the posting means
Normalised locationAmbiguity, remote, multi-site rolesEvery geographic cut depends on it

Questions worth asking before you buy

  • What share of postings have a normalised salary, not just a salary field?
  • How do you decide two postings are the same role, and what is your error rate?
  • Do you keep first-seen timestamps, or only the current snapshot?
  • When a listing disappears from the source, how quickly do I learn?
  • Where does the data come from — the employer's own systems, or another vendor's feed?

That last one matters commercially and legally. A provider re-selling another vendor's feed inherits that vendor's risk. Proxycurl, a well-known provider of adjacent data, shut down in July 2025 following litigation from LinkedIn. Where the data comes from is not a detail.

What "real time" usually means

Rarely what it sounds like. Ask for the distribution, not the claim: what share of postings are discovered within an hour, within a day, within a week. A provider who cannot answer has not measured it.

More

Related reading

Got a correction?

If something here is wrong or out of date, tell us and we will fix the post.