Job boards overlap far less than you think
The same posting rarely turns up on more than one board. The repetition you notice is one board relisting, not boards copying each other.
Jobzyl3 min read
Figures on this page are current as of
Everybody who has searched for a job in the last decade holds the same belief: the boards are all the same. You open six tabs, you see the same postings, you close five of them.
We can measure that, because merging duplicate postings is something this product has to do anyway. The belief turns out to be wrong, and wrong in a way that changes how you should search.
The same posting almost never appears on two boards
Show the numbers
| Category | Value |
|---|---|
| United States | 1.3% |
| Canada | 0.9% |
| India | 0.8% |
| United Kingdom | 0.4% |
| Australia | 0.3% |
| Germany | 0.1% |
Data as of
In the United Kingdom, 0.4% of postings turned up on a second board. In Germany it is 0.1%. The highest number on the chart does not reach two per cent, and we double-checked the query before believing it.
This is measured across a set that puts the big generalists and the specialists together. In the UK slice alone that means Careerjet, Indeed, Reed, Greenhouse, Ashby, Lever, SmartRecruiters and Jooble, all searched for the same thing at the same time. They mostly returned different jobs.
Then why does it feel so repetitive?
Because it is repetitive. Just not in the way people assume.
Show the numbers
| Category | Value |
|---|---|
| Germany | 18.8% |
| India | 14.5% |
| Canada | 13.6% |
| United States | 13.2% |
| United Kingdom | 13.0% |
| Australia | 12.2% |
Data as of
Between an eighth and a fifth of postings exist more than once, depending on the market. Set that against the chart above and the arithmetic is unavoidable: nearly all duplication is one board listing the same job more than once, under a different URL, not two boards carrying it.
That matches how boards behave. A posting gets relisted when it is refreshed, an employer runs the same requisition for three offices, an agency posts the same role under four slightly different titles to catch more searches. Every one of those produces a duplicate within a single board, and none of them produces overlap between boards.
The tabs are not showing you the same jobs. Each tab is showing you its own jobs, several times.
What follows from this
Searching one board means missing most of the market. If overlap were high, picking a favourite board would be a reasonable strategy, because the others would be mostly redundant. At under two per cent, the others are almost entirely additional, and whatever single site you are loyal to is showing you a slice.
Deduplication turns out to be worth more within a board than across boards, which is the opposite of what we expected when we built it. It is why the merge runs on title, company and location rather than on URL. A URL-based merge would catch almost nothing, since the duplicates that actually exist have different URLs by definition.
And breadth beats depth. Reading three pages deep on one board is a worse use of an hour than reading the first page on ten. That is the whole argument for an aggregator, and until we ran this we could not have told you whether it was true.
Frontend developer roles in the United Kingdom and security engineer roles in the United States both run every board at once, and each row names the boards it came from.
Method, and why this number is a floor
Two postings are the same posting when the title, the company and the location match after accents are folded, punctuation is stripped and company legal suffixes are dropped, so "AGS Ltd" and "AGS" are one employer. That is the same identity the product uses to merge rows on screen. Measuring with it means this post describes the overlap we actually remove, not a theoretical one.
It also means every number here is a floor. An employer who rewrites the title between boards, or a board that appends its own suffix, produces two postings this method will never match. Real overlap is higher than the chart says. It is not plausibly ten times higher, but we cannot bound it from here, and anyone quoting these figures should quote them as a minimum.
Rows without a merge key are excluded rather than folded together. Those rows predate the column, and treating each as unique would have deflated the overlap share silently, which is the more flattering mistake and therefore the one to avoid.
Grouped by the location we searched for, which is not the same thing as the country a job is in. Every posting is counted once per group.
The comparison covers the boards we index and name. Rows that reach us through a scraping fan-out rather than a provider integration are excluded, which costs a little overlap, because those rows do carry postings.
Related
- A job advert is not the same on every boardThe same role can be a two-line teaser on one board and a full brief on another. Which board you read it on decides how much of it you get.
- What employers actually name as skillsPython leads, English is second and Excel outranks SQL. What company career pages name when they list a skill, and what that leaves out.
Every figure here comes from the jobs Jobzyl has indexed, and carries the date it was pulled. Subscribe by RSS, or run the search yourself.