# Amazon Jobs API > Query Amazon's postings through one API with ?source=amazon — plus why amazon.jobs' posted_date parses a day early east of Greenwich. Source: https://hyperjobs.io/ats/amazon Amazon is not a platform. Every other source on this site is software that hosts many employers' boards — one [Greenhouse](/ats/greenhouse) adapter reaches every company that bought Greenhouse. This one reaches Amazon. That's the whole of it. It's here because Amazon is large enough that its jobs matter on their own, and because it runs a bespoke careers site rather than buying an ATS, so no platform adapter will ever find it: there's no `jobs.lever.co/` to detect, no vendor fingerprint to match. If you want Amazon's postings, you write an adapter for Amazon and you own it forever. We wrote it, and it's ours to own: query it with `?source=amazon`. [`/v1/meta`](/docs/api-reference/meta) is the authoritative live vocabulary, with the current count per source. ## How amazon.jobs actually works Of the bespoke giants, this is the friendly one. Unauthenticated JSON, offset pagination, full job objects with descriptions inline: ```bash curl "https://www.amazon.jobs/en/search.json?offset=0&result_limit=100&sort=recent" ``` No token, no handshake, no browser, no N+1. The response carries `hits` for the total and `jobs` for the page, and you walk `offset` until the batch comes up short or the total is exhausted. ### What the adapter reads The description isn't one field. Amazon splits a posting into three, and they have to be reassembled in order: ``` description +

Basic qualifications

basic_qualifications +

Preferred qualifications

preferred_qualifications ``` Read `description` alone and you get the pitch with none of the requirements — which looks like a populated field, so nothing alerts you. It's the same trap [Lever](/ats/lever)'s three-part body sets, in a different arrangement. `job_category` (`"Medical, Health, & Safety"`) and `job_family` (`"Pharmacy"`) are both consistently filled, giving a department and a finer team grouping with nothing to infer. The adapter keeps `job_family` only when it differs from the category, so a row doesn't carry the same string twice. The id field is called `id_icims` — a fingerprint of what's underneath the careers site. It's the stable id and the adapter uses it, but nothing else about this API behaves like a standard iCIMS tenant. Don't read the name as a shortcut. ### The coordinates are hidden inside strings `locations[]` is an array of **JSON-encoded strings**, not objects. Each has to be parsed a second time to reach a `"coordinates": "lat,lng"` field inside it. JSON inside JSON, with unparseable entries skipped rather than thrown. It's worth the trouble: exact coordinates mean no geocoding, so a location is a point rather than a string somebody has to resolve into a guess. ## What bites you Amazon posts bare calendar dates: `May 6, 2026` — often with a double space, and with no timezone anywhere. `Date.parse` reads a non-ISO date-only string as **local midnight**. On a machine in Europe/Rome that makes `May 6, 2026` into `2026-05-05T22:00Z`, and every posting reads as a day earlier than it was posted. This was live in production on this source. It throws nothing, it looks completely normal, and it reproduces only east of Greenwich — so it survives review on any laptop set to UTC or US time. The fix is to build the date from its parts with `Date.UTC` and never `Date.parse` a bare date. ISO date-only strings aren't affected, because the spec pins those to UTC — which is exactly why this hides: the common path is correct. **One adapter, one company, forever.** This is the honest cost of a bespoke source. When Amazon reshapes `search.json`, nothing else in your pipeline notices and nothing else covers the gap. Multiply that by every giant you want and you see the shape of the problem. Same deal as [Apple](/ats/apple), [Uber](/ats/uber), [TikTok](/ats/tiktok), [Google](/ats/google), [Microsoft](/ats/microsoft) and [IBM](/ats/ibm). **There's no salary.** Nothing structured. Whatever comp appears is prose inside the qualifications HTML. **A missing `job_path` means no URL.** Rows without one are dropped rather than guessed at, since the posting URL is built from it. ## Or use ours The fetch is genuinely easy here and this page won't inflate it: pagination, a three-part body reassembled, JSON parsed out of JSON strings for the coordinates, and dates built at UTC instead of parsed. What we add is mostly what surrounds it — Amazon's rows in the same [`Job`](/docs/objects/job) shape as every other [source](/docs/reference/sources) rather than in Amazon's, dedup against the same role on [LinkedIn](/ats/linkedin), expiry by board diffing since the feed declares none, and the adapter's upkeep being ours instead of yours. [Sources](/docs/reference/sources) · [Quickstart](/docs/quickstart)