What this measures, and what it does not
Every other figure we publish comes from someone else's research. This one does not. When someone wants to reach the human behind a job posting, they paste that posting's URL into this site — so the domain in that URL records where the job was found, which is a thing job boards know about themselves and nobody publishes across all of them at once.
The population is people motivated enough to look up a hiring manager, not job seekers in general. That is a real narrowing and it runs in a predictable direction: someone chasing a specific person is more likely to have come from a company's own careers page than someone scrolling a feed. Read the shares as a picture of considered applications, not of the whole market.
Where the postings come from
- Job board50%
- Company careers page29%
- Applicant tracking system19%
- Freelance marketplace2%
Roughly half of these postings were not found on a job board at all. Company career pages and applicant tracking systems together account for 50% — people going directly to an employer they already had in mind, or landing on a Greenhouse or Ashby page from a link somewhere. Any advice that treats the job search as a board-shaped activity is describing half of it.
Which platforms, specifically
- LinkedIn41.3%n=862
- Company careers pages29.5%n=615
- Workday5.1%n=106
- Wellfound5.0%n=105
- Greenhouse4.4%n=92
- Ashby3.6%n=76
- Indeed1.8%n=38
- Oracle Recruiting1.4%n=30
- Lever1.3%n=28
The gap between LinkedIn and Indeed is the number that surprised us most. LinkedIn appears 23 times as often as Indeed here, despite Indeed carrying more total job traffic worldwide. The likeliest explanation is not that Indeed is unused but that it is used differently: a LinkedIn posting shows you a company and often a poster, which is what makes someone think there is a person to find.
The long tail is most of the story
Those 2,086 postings came from 553 distinct domains, and 406 of those appeared exactly once. Below the 9 platforms large enough to publish individually sit 26 more, holding 134 lookups between them.
A job search does not happen in four places. It happens in one big place, a handful of medium ones, and several hundred single-employer pages that no aggregator covers.
Every number on this page
| Platform | Type | Lookups | Share |
|---|---|---|---|
| Job board | 862 | 41.3% | |
| Company careers pages | Company careers page | 615 | 29.5% |
| Workday | Applicant tracking system | 106 | 5.1% |
| Wellfound | Job board | 105 | 5.0% |
| Greenhouse | Applicant tracking system | 92 | 4.4% |
| Ashby | Applicant tracking system | 76 | 3.6% |
| Indeed | Job board | 38 | 1.8% |
| Oracle Recruiting | Applicant tracking system | 30 | 1.4% |
| Lever | Applicant tracking system | 28 | 1.3% |
| All other platforms (26) | mixed | 134 | 6.4% |
Why this is measured the way it is
Every choice below narrows what the data can say. They are listed so you can disagree with a specific one rather than with the whole thing.
Why the domain of a pasted URL is a fair proxy
Nobody types a job URL from memory. It arrives by copy and paste from the page the person was reading, which makes the domain a record of where they were rather than a claim about where they usually look. That is a narrow fact, and narrow is the point: surveys asking “where do you look for jobs?” measure what people recall and are willing to say, and recall over-reports the famous names. This measures an action instead of an opinion about an action.
Why the unit is a lookup, not a person or a paste
One row is one job posting someone went to the trouble of researching. Counting people would need identity we deliberately do not hold on the free tier. Counting pastes would let one determined user with forty applications outweigh forty users with one each, and it is the second group whose behaviour generalises. A lookup sits between the two and is the only unit the system actually records.
Why an unrecognised domain counts as a company careers page
The alternative is a bucket called “other”, and “other” at 29% would be the largest and least informative row in the table. We checked the long tail by hand: it is overwhelmingly employer domains — jobs.company.com, careers.company.com,company.com/careers/…. The mapping errs toward that reading, which slightly overstates company pages whenever a small regional board slips through unrecognised. That direction is stated rather than hidden because it is the one error this design can make.
Why 25 is the floor for publishing a row
A share computed over nine observations moves eleven points when one more arrives. Publishing it invites a reader to quote a number that the next refresh contradicts, which costs more credibility than the row was ever worth. Everything below the line is still counted — it appears as a single total so the percentages add up.
Why no employer is named
The aggregate of where people look is a fact about the job market. A list of which employers our users are chasing is a fact about our users, and it is not ours to publish. Every company careers page collapses into one row for that reason, not because the breakdown would be uninteresting.
Why a snapshot instead of a live query
A page built to be cited must not move underneath the citation. Each refresh is a single commit, so the numbers you quote stay quotable, the change between versions is a readable diff, and there is no way for the page to quietly disagree with itself between two readings.
What we removed, and why
An earlier version of this page also reported how often a hiring manager could be identified from each platform. That column is gone. The spread was real but it tracked how public a company's team is, not anything about the platform hosting the posting — early-stage startups are hard to research wherever you find them. A caveat under a chart does not stop the chart being quoted as a ranking of platforms, so the honest fix was to stop publishing it rather than to annotate it.
Method, and three ways this is biased
One row is one job posting looked up between 2026-05-24 and 2026-08-24. The domain is taken from the URL and mapped to a platform; anything unrecognised is counted as a company careers page, which is what it almost always is. Employers are never named: every company careers page is aggregated into one row. No cell below 25 observations is published on its own.
Three things distort this, and it is better that you know all three.
- The sample is self-selected. These are people who decided to find a hiring manager. That is not the average applicant.
- Some inputs are rejected before they are recorded. The tool turns away search-results pages and bare domains, so those never become a row. Over this window that was 166 rejections against 2,086 recorded lookups — 7.4% of all attempts, split 83 list pages and 83 bare domains. 73 of them came from linkedin.com, which means LinkedIn's true share of attempts is slightly higher than the table shows, not lower.
- Repeat lookups of the same posting collapse. When a posting has already been looked up, the second request reuses the first record. A role that many people chase counts once, which dampens whichever platforms concentrate attention on the same postings.
Three months and 2,086 lookups is enough to state the shape and not enough to track a trend. This page is rebuilt from the source table on each refresh, and each refresh is a commit, so the version you cite stays the version you cited.
Using this data
Free to quote and republish with a link. The full table is available as CSV.
DearHiringManager.io, “The Outreach Index”, snapshot of 2026-08-24. 2,086 job postings looked up between 2026-05-24 and 2026-08-24. https://dearhiringmanager.io/outreach-index
Found an error, or want a cut we did not publish? Write to team@dearhiringmanager.io and it gets corrected with the date of the change.
Related: the externally sourced benchmarks on job search statistics 2026, and what to do once you have the posting — finding the hiring manager.