The public job posting APIs of Greenhouse, Lever, Ashby, Recruitee and Personio are dependable once you know a board exists: on 19 Nov they returned a job list for 19,768 of 19,805 known boards (99.8%), with no API key. The waste is in finding boards: 36.0% of 30,928 board addresses taken from Common Crawl answered "not a board". Workable is the exception: it stopped answering after 3,834 to 3,840 boards in each pass, and the recorded stop was an HTTP 429 asking for a 24-hour wait.
Download the per-ATS numbers: CSV · JSON
How many discovered board addresses are real job boards?
On the five ATSs read on 18 Nov, between 57.1% (Personio) and 75.5% (Ashby) of the addresses were live boards on their first check. Workable, first read on 19 Nov, was at 96.0% of the addresses it answered. Every address came from Common Crawl's public URL index, which keeps links to boards that have since closed or moved. "Not a board" means HTTP 404, 410 or 401, or a response that is not the job API; such an address is not read again.
| ATS | Addresses | Not a board | Share |
|---|---|---|---|
| Greenhouse | 10,154 | 3,654 | 36.0% |
| Lever | 3,855 | 1,560 | 40.5% |
| Ashby | 4,915 | 1,202 | 24.5% |
| Recruitee | 4,807 | 1,619 | 33.7% |
| Personio | 7,197 | 3,088 | 42.9% |
| Workable* | 10,785 | 153 | 4.0% |
*Workable's lane stopped on 18 Nov before reading any board, so its row is from 19 Nov: the share is of the 3,834 addresses answered in the second pass of that day before its rate limit (the first pass answered 3,840).
How reliable are the APIs on a daily re-check?
Very, for boards already known to exist. On 19 Nov the crawl read every board not ruled out the day before; Greenhouse alone returned 191,853 of the 461,985 jobs, a total that includes Workable's 67,009.
| ATS | Job list returned | Jobs per board | No jobs | Seconds per board |
|---|---|---|---|---|
| Greenhouse | 99.9% | 29.5 | 14.6% | 0.08 |
| Lever | 99.7% | 29.7 | 23.0% | 2.15 |
| Ashby | 99.9% | 16.3 | 10.0% | 0.08 |
| Recruitee | 99.5% | 14.8 | 13.6% | 0.25 |
| Personio | 99.9% | 7.8 | 13.6% | 1.03 |
| Workable | 96.0% of 3,834 read, then stopped | 18.2 | 41.7% | 1.01 |
Job list returned: share of known boards whose API answered with a valid job list, empty or not. No jobs: share of boards whose list was empty. Seconds per board: each ATS's pass time divided by its boards, at 4 requests at a time for Greenhouse, Ashby and Recruitee and one a second for the others. That is the pace the crawl allowed itself, not each API's limit. Lever took 2.15 seconds per board under the same one-a-second rule as Personio and Workable; the reports do not show why.
Why requests failed
- Workable stops after a few thousand requests. On 18 Nov it answered HTTP 429 with
Retry-After: 84362(23.4 hours) before any board was read, and the crawler stops a lane told to wait over an hour. Yet on 19 Nov it answered again before that wait was up: 3,840 boards in the first pass and 3,834 in the second, which began 2.4 hours after the first, both at one request a second. The second pass's stop was HTTP 429 withRetry-After: 86400(24.0 hours). Its other 6,951 checks failed, and at least 6,785 of them ended within 0.8 seconds, without a request. Budget for a few thousand Workable requests per run rather than for the stated wait. - Personio blocks bursts. Its first pass, several requests at a time, stopped after 114 seconds, when 20 of the last 50 answers were HTTP 403 or 429; only 515 boards had answered. At one request a second the other 6,682 boards answered with no errors in 2.8 hours. That pass also made 3,232 company-name lookups on the same lane: at least 9,914 requests, or no more than 1.01 seconds per request on average. The 19 Nov pass had no errors either.
- Boards vanish overnight. 6 Greenhouse, 8 Lever, 3 Ashby, 8 Recruitee and 3 Personio boards that were live on 18 Nov answered "not a board" on 19 Nov.
- Some addresses never answer. The same 9 Recruitee addresses failed in every pass on both days without returning 404: timeouts, 5xx or 403 (the reports do not split them).
- A job list is not always enough. For 126 Lever, 21 Ashby and 15 Personio boards the API returned jobs, but the hosted board page answered 404, so job links would not open. The index leaves those boards out.
Endpoints and the request rate that held
- Greenhouse:
https://boards-api.greenhouse.io/v1/boards/{token}/jobs, 4 at a time, no failures. - Lever:
https://api.lever.co/v0/postings/{token}?mode=json, one a second (its robots.txt setsCrawl-delay: 1). - Ashby:
https://api.ashbyhq.com/posting-api/job-board/{token}, 4 at a time, no failures. - Recruitee:
https://{token}.recruitee.com/api/offers/, 4 at a time. - Personio:
https://{token}.jobs.personio.de/xml(some accounts use.com), one a second. - Workable:
https://apply.workable.com/api/v1/widget/accounts/{token}, one a second, still rate-limited.
Field lists are in the vendors' docs: Greenhouse Job Board API, Lever Postings API, Ashby Job Postings API.
Plan a crawl with these rates
The planner applies the rates above to your own number of board addresses, assuming they behave like the ones in this crawl. It reads the same JSON file you can download.
Whole number you plan to read.
Expected at this crawl's rates
Day over day
Totals barely moved. From the 18 Nov index to 19 Nov, live boards changed by -6 at Greenhouse, -4 at Lever, -3 at Ashby, -8 at Recruitee and -3 at Personio, and jobs by -0.05%, -0.53%, -0.01%, +0.09% and +0.03%. Lever lost 8 boards, but its boards without a public page fell from 130 to 126, so its index shrank by 4. For the five ATSs that finished both 19 Nov passes, 2.4 hours apart, job counts differed by 38 or fewer. These are net counts: the reports do not show how many jobs opened and closed in between.
Method and limits
- What ran: the refresh behind ATS Jobs Search, from one machine, with a generic user agent and no login, key, proxy or browser. It retries 429, 5xx and timeouts up to 4 times with backoff, and stops a lane when 20 of its last 50 answers are HTTP 403 or 429 or when a 429 asks to wait over an hour.
- Addresses: board URLs in Common Crawl's index. Shares describe those addresses, not how many companies use each ATS.
- Dates: reports exist for two days, 18 Nov 2026 and 19 Nov 2026 (UTC): a day-over-day change, not a trend.
- Excluded: BambooHR, read on 18 Nov and then dropped from the crawler because its developer terms bar crawling.
- Reproducible: build_data.py rebuilds the CSV and JSON from the refresh's report files, records each file's SHA-256 hash, and fails if a number on this page is not in that data. The report files themselves are not published; the CSV and JSON carry every per-ATS count used here. This page was drafted with AI tools from those files.
Read these feeds yourself
Scaleturn, which publishes this site, also makes the open-source ats-job-feed library, which reads the Greenhouse, Lever, Ashby and Workable public APIs into one schema, and the hosted Career Site Job Feed, which runs it on Apify. These are not affiliate links. Both use the same public endpoints as the refresh; the figures on this page come from the refresh, not from them.
Three more reports count the jobs in the ATS Jobs Search index: what day of the week most jobs are posted, which AI job titles employers post and how old open job postings are.
Sources
ScrapeTools job-index refresh reports for 18 Nov 2026 and 19 Nov 2026, summarized in ats-board-crawl-2026-11.csv and ats-board-crawl-2026-11.json.
Board addresses: Common Crawl URL index.
Endpoint documentation: Greenhouse, Lever, Ashby, checked 19 Nov 2026.