In August we timed 11,219 tech roles from the day they appeared to the day they closed. Half were gone in 13 days, and only 3.6% stayed open longer than 90 days.
So here is a number that should not be possible. At the companies we have tracked since before April, a third of the tech roles open right now have been open for more than 90 days.
Both numbers are true. The gap between them is the most useful thing we know about ghost jobs, and it is rarely what ghost job articles lead with (including, if we are honest, the one we wrote in April).
Everything below comes from Remoet's own data on tech roles, measured on September 17, 2026 unless a date says otherwise, and each section says exactly what it counts.
Old jobs pile up
The trick is the difference between jobs that close and jobs that are open.
A role that gets filled in ten days is on the board for ten days. Most people never see it. A role that sits unfilled for five months is on the board for all five of those months, and everyone who searches in that window sees it. Fast jobs leave. Slow jobs stay and accumulate.
So when you look at what is open today, you are looking at a sample that is heavily tilted toward the stragglers. Across 4,161 open roles (a role posted in several places counts once) at 421 companies we have followed since before April:
| Open for more than | Roles | Share |
|---|---|---|
| 30 days | 2,803 | 67% |
| 60 days | 1,921 | 46% |
| 90 days | 1,408 | 34% |
| 150 days | 848 | 20% |
The median open role there has been up for 55 days. The median closed role was open for 13.
That is the core of the ghost job feeling. You are not imagining that the board is full of old postings. A large share of what you can see at any moment is the slow tail, because the fast part has already left.
Old does not mean fake
It would be easy to stop there and call a third of the board ghosts. The data does not support that.
Here are the companies with the most open roles we first saw more than 90 days ago:
| Company | Roles open 90+ days |
|---|---|
| OpenAI | 201 |
| Databricks | 168 |
| Airwallex | 97 |
| Anthropic | 96 |
| Nebius | 67 |
| Datadog | 60 |
(These are counted by exact title per company, on postings first seen since March.) Our data cannot see intent, so we are not going to guess at any one of these. What we can say is that large, fast-growing companies commonly keep a posting open for months as a standing requisition for a kind of engineer they hire over and over (recruiters call it an evergreen requisition). An old posting there is often a pipeline rather than a single seat.
The same age means something different at a company with eight employees and one open role that has not moved since spring. That one is worth a direct question before you spend an evening on the application.
Some jobs come back
The other pattern people describe as ghosting is the job that disappears and then reappears.
We can see that too. Counting open roles by their exact title at each company, 870 of 12,868 are back under a brand new posting after an earlier posting with the same title closed. That is about one open role in fifteen. For 359 of them the gap was a month or more, and 488 had already been posted and closed at least twice before this version.
We deliberately left out another 1,950 roles that came back at the same web address. Those are almost always on our side (a careers page that failed to load for a few days, then came back) rather than a company reposting anything.
A repost is not proof of a ghost either. A company hiring its second "Senior Software Engineer" this quarter looks exactly the same in this data as a company recycling one it never meant to fill. What a repost does tell you is that the first round did not end in a hire, or ended in one and needed another. Either is worth knowing.
Almost nobody labels a talent pool
There is an honest version of the always-open posting: the talent pool, the expression of interest, the "don't see your role, apply anyway". Those are fine. You know what you are getting.
They are also vanishingly rare. Among open tech roles first seen since March, 11 carry a title that says so, at 6 companies. Everywhere else, a standing pipeline and a real vacancy look identical from the outside.
What to actually do with this
Posting age is a real signal, just not a verdict. What we would do:
Look at the age next to the company. An old posting at a company hiring across dozens of roles is usually a standing door. An old posting that is the only open role at a small company deserves a question first.
Check whether the company is hiring at all. If a company's other roles keep turning over while one sits for months, that one is probably a pipeline. If nothing there ever changes, be more careful.
Treat a repost as information, not an insult. Ask what happened with the previous round. The answer tells you a lot.
Spend your effort where the flow is. In our August sample half the tech roles that closed did so within two weeks, so the new ones are where hiring decisions are most likely being made this week.
If you search through an agent, every role on Remoet carries the date we first saw it and the last time we confirmed it was still open, and you can sort by newest. That makes the flow easy to find. It works without an account.
Where this data is soft
"First seen" is when we first saw a role, not when the company posted it. For a company we started following in March, anything already open then gets March as its start date, so the oldest roles are older than we say. And we only mark a role closed after it has been missing from a careers page three checks in a row, so lifespans skew a little long.
The repost count matches on exact titles. It misses a repost that changed its title, and it counts a genuine second hire for the same title as a repost. The two errors pull in opposite directions, and we cannot tell you which is bigger.
The closed-role figures come from our August measurement, and the open-role figures from September 17. The stock and flow were not measured on the same day, although the pattern between them is large enough that we do not think the timing changes it.
And none of this can see intent. We can tell you a role has been up for five months. We cannot tell you whether anyone is reading the applications.
If you have applied to a role that turned out to be one of the stragglers, or you hire and have a better explanation for the long tail than ours, come tell us in Discord. This is the kind of thing we would rather get right than get first.
