Should I Apply to Old Job Postings? We Aged 3,777 AI Jobs

The question shows up in our inbox in one of two shapes. Either “this job was posted three months ago, is it dead?” or the sadder version, “I found it too late.” Both assume the little grey date under the job title means something.

We read those dates for a living. Every morning a script here pulls the public feeds behind 20 employer career pages — OpenAI, Anthropic, Waymo, Figure AI, Scale AI, xAI and the rest — and files what it finds into our jobs tracker and our jobs board. This week we turned the script on the dates themselves. What came back was uncomfortable enough that we changed our own board before publishing this.

The date on the posting is often not the date it was posted

Most of these companies run their hiring on one of two systems, Greenhouse or Ashby, and both publish a job feed that anyone can read. Greenhouse’s own Job Board API documentation lists three date-ish fields on every job: first_published, updated_at, and application_deadline.

The date you see on the careers page is usually the second one. And updated_at moves whenever anyone on the hiring team touches the record — fixes a typo, adds a location, re-syncs the board. Nothing about the job changes. The clock just resets.

Waymo is the cleanest case. On September 6 every single one of its 344 open roles showed a posted date of three days ago or newer. Read first_published on the same 344 roles and the median is 87 days, with 48% over three months old. We can prove the restamping rather than infer it, because we scanned the same boards on August 29 and kept the file: of the 316 Waymo listings that appear in both scans, unchanged and at the same URL, 316 had moved their date forward. All of xAI’s 233 did too. Across all boards, 716 of 3,454 surviving listings (20.7%) got a fresher date in eight days without being new.

The practical damage runs in one direction. On September 6, 356 roles looked a week old or newer and were actually more than 90 days old. Across the whole set, 1,570 postings (41.6%) showed a date fresher than the truth, by a median of 73 days. Not one showed a date older than the truth. If the date is wrong, it is wrong in the direction that makes you feel early.

How old are these jobs really

Two thirds of open AI jobs have been open for over a month. One in three is older than a quarter. And 213 of them — 5.6% — have been collecting applications for more than a year.

The oldest is a Helix AI Engineer, Perception role at Figure AI, first published on November 15, 2022. That is 1,390 days. We opened it on September 6, along with the four next-oldest at OpenAI, Scale AI and Nuro. All five load, and all five still have an apply button.

Do old postings close sooner? We checked

Age is only worth worrying about if it predicts something. So we ran the obvious test on the eleven boards that publish a true posting date: take every role open on August 29, sort by how old it was that day, and see which ones were gone by September 6.

The pattern people expect — old listing, about to vanish — is not there. Postings over six months old closed at 2.5% over the eight days. Postings under a week old closed at 5.9%, more than double. The middle buckets sit between 3.6% and 7.2% with no trend. Across the whole sample, 102 of 1,937 roles closed, roughly 0.7% a day.

That daily rate is the only clock that actually exists here, and it ticks at the same speed for a role posted yesterday and a role posted last winter. Waiting a week costs you about five percent of the doors. Waiting because a posting looks old costs you nothing you were going to get anyway.

There is no closing date, and that is the real answer

Here is the number that reframed this piece for us. Greenhouse gives every employer an application_deadline field. We read all 1,781 postings on the eight Greenhouse boards we track. The number that filled it in: zero. The number carrying an internal requisition ID: 1,780.

Federal jobs are different — a USAJOBS announcement has a real closing date printed on it, which is why the federal process rewards calendar discipline in a way private hiring does not. On a company board there is no bell. A req is open until a hiring manager closes it, and it closes when they fill it, freeze it, or forget it. “Am I too late?” has no answer because there is no schedule to be late against. The only meaningful question is whether the posting is still in the employer’s feed today, and you can check that in ten seconds.

What an old posting usually is

Old does not mean fake, and it does not mean live, either. In our data the year-plus roles cluster into three recognizable shapes.

Shape	Looks like	What it means for you
Pipeline req	“General Application”, broad titles, no team named	Real, permanently open, low odds per application. Cognition’s general application has been up since March 2024.
Always-hiring role	Technician, operator, data center, service	Real and repeatedly filled. Figure AI’s Robot Service Technician has been posted 758 days.
Hard-to-fill senior role	Narrow research or infrastructure title, high band	Genuinely unfilled. Age is a signal of difficulty, which can favor a non-obvious candidate.

Of the 1,399 roles open more than 90 days, 1,124 are in the US, 812 publish a pay range, 333 are remote and 48 are entry-level. The old pile is not junk. It is where a third of the disclosed-pay market lives — and pay disclosure is its own story, which we took apart in why AI salary ranges are so wide.

Four checks that beat looking at the date

1. Open it on the employer’s own domain. If the role is still listed at greenhouse.io, ashbyhq.com or the company’s careers page, it is in the live feed today. That is the only freshness signal that means anything. A copy on an aggregator can outlive the original by weeks.
2. Ignore “posted 2 days ago” on a company board. Now that you know eight of our twenty boards restamp, treat that line as decoration. Our jobs board now shows the first-published date instead, which is why some listings there suddenly look older than they did last week.
3. Read the title for shape, not the date. “General Application” is a pipeline. A named team and a named manager is a live req. A technician role that has been up 400 days is a role they hire for continuously, so today is as good as any day.
4. Apply within the week you find it, then move on. Roughly 0.7% of postings vanish per day. That is the actual cost of delay, and it is the same whether the job went up yesterday or in 2023. Nothing here says a human is reading — only that the door has not been shut.

Ghost jobs, and where this argument stops

There is a fair objection to everything above: a posting can be in the feed and still be going nowhere. That is the “ghost job” complaint, and our data cannot rule it out. We can see the door. We cannot see whether anyone is behind it.

Note what the ghost-job argument actually implies, though. If a listing is a placeholder, it was a placeholder on day one too. Age is not the tell. The tells are the ones in the CBS segment below: no named team, no requisition number, a company that has posted the same role continuously through a hiring freeze, or a “recruiter” who moves the conversation to a messaging app.

If a posting asks you for money at any stage, it is not a stale job. It is a scam, and the age of it is the least of the problem. We keep the four rules in our piece on AI training work.

What to do with this today

• Go back to the three jobs you skipped because they looked old. Check each one on the employer’s domain. If it loads, apply this week.
• Filter our jobs board to entry-level or remote and sort by pay rather than by date — there are 104 entry-level roles open right now, median true age 81 days.
• If you are aiming at one company, start with the two we have taken apart in detail: Anthropic and xAI in Memphis.
• Check the running counts on the tracker before you write a cover letter that guesses at the market.

If you want to go deeper

• “Which of these companies is actually growing?” — who is hiring right now and which one pays the most.
• “Do I have the credentials?” — whether you need a PhD, whether a certificate is worth it, and what a robotics technician job actually requires.
• “What about academia?” — university postings run on real deadlines and a real season: when faculty jobs get posted, the faculty search tracker, and labs recruiting for next fall.
• “Where do I start?” — the application walk-through, the data center route, and what to say about AI on your resume. New here? Start here or browse the resources shelf.

Questions people actually ask about old listings

Should I apply to old job postings?

Yes. On September 6, 2026 we measured 3,777 open AI and robotics roles and found that postings more than six months old closed at 2.5% over the following eight days, while postings under a week old closed at 5.9%. Age did not predict closing. Check that the role is still on the employer’s own site, then apply.

How long do job postings stay open?

Longer than most people assume. In our September 6 count the median open AI job had been posted 57 days, two thirds were older than a month, 37% were older than 90 days and 213 had been open more than a year. The oldest still accepting applications went up in November 2022.

Why does a job say posted 2 days ago when it has been up for months?

Because the date shown is usually the last-updated timestamp, not the original posting date. Both fields sit in the same public feed. On September 6 we found 356 roles that looked a week old or newer but were really more than 90 days old, and 1,570 in total whose shown date was fresher than the truth, by a median of 73 days.

Is a job posting from three months ago already filled?

Sometimes, and there is no way to tell from the outside. What we can say is that the posting being three months old does not make it likelier to close than a fresh one. If it has been filled, the req usually disappears from the employer’s feed — 262 of the 3,716 roles we saw on August 29 were gone eight days later.

Do job postings have a deadline?

Private ones generally do not. Greenhouse gives employers an application-deadline field, and on September 6 not one of the 1,781 postings we read had filled it in. Federal announcements on USAJOBS are the exception: those carry a real closing date and you should treat it literally.

Is it too late to apply if the posting is a year old?

No, but adjust what you expect. Year-old postings tend to be pipeline reqs, continuously hired technician roles, or genuinely hard-to-fill senior positions. All three accept applications. The first gives you the lowest odds per application, the third can favor an unusual candidate.

How can I tell if an old job posting is still real?

Open it on the employer’s own domain rather than an aggregator, and look for a named team and a requisition number. If the page loads on greenhouse.io, ashbyhq.com or the company careers site today, the req is in their live feed today. Aggregator copies routinely outlive the original.

Should I apply to a job posted on the same day it goes up?

Applying early is fine, and it costs nothing, but our numbers give no reason to panic-apply. Fresh postings closed slightly faster than old ones in our eight-day window, and the overall rate was about 0.7% of postings disappearing per day. Waiting a week costs roughly 5% of your options.

Written by Prof. H, who has spent the week arguing with timestamps. The Jobs Desk publishes daily; the underlying counts refresh every morning on the tracker and the board. Next: what the eight-day disappearance rate looks like company by company, and whether the roles that vanish are the ones that pay.