The Job Ads That Outlived the Jobs

01Five roles nobody was hiring for

The agency's careers page listed five open roles, each with an apply link and an email address. None of them was open any more, and the applications kept coming. Today we closed them properly. Before changing anything, I looked at who had been reading the page. The answer explained the applications better than any job board could. In the fourteen days to 19 October, ChatGPT's live agent fetched the careers page 42 times. Only the homepage was fetched more, 54 times. The most read article on the site came third, with 40. Your careers page is a statement of fact about your company, and assistants repeat it long after it stops being true.

The people it sent

ChatGPT marks the links it hands to people, so its visits can be counted at the door. In those fourteen days there were 10 such visits. Two were bots. Of the other 8, six landed on the careers page. People were asking an assistant about working with us, the assistant read the page, saw five open roles, and sent them to apply. Google was doing the same quietly: the page collected 129 impressions in about three months at an average position of 7.7, 60 of them for a search pairing a short form of our name with the words AI job. All these counts are a floor, since four of the fourteen days fall after the site began caching its pages at the edge, which hides some fetches from the server.

02Stale facts travel furthest

No markup required

The page had no job posting markup at all, nothing that told a machine these were job listings. It did not need any. An assistant reading the page in plain language understood five open roles and an address to write to, and repeated that faithfully. Assistants do not need your structured data to repeat your claims; they only need your claims to be readable. That cuts both ways. The same readability that gets a guide quoted gets an out of date notice quoted.

The fact with the longest life

Of everything a company publishes about itself, a careers page is the easiest to forget and among the most likely to be asked about by name. Prices, team size and hiring status are the facts people check before they write to you, which is why they belong in the checks of entity hygiene. And as the reading counts in the page ChatGPT reads most showed, the pages an assistant fetches about you are often not the ones you think of as content. The page you stopped maintaining is often the page the machines still trust.

03How to close a job page properly

What we changed

We kept the page, because it ranks and because it describes how we work. Each role now carries a clear position filled badge with no apply link. A notice above them says hiring is paused, that applications sent meanwhile are not reviewed, and the date it was last updated. Every application email address is gone from the page. The frequently asked question about whether we are hiring now answers no, plainly, and so does the page description. If an assistant reads the page tomorrow, the first sentence it finds should be the true one.

Deleting the page would have been the quickest fix and the worst one. Searches about working with us would land on an error, and an assistant holding an older copy would have nothing newer to replace it with. A page that says no, clearly and with a date, gives every reader, human or machine, a newer version to repeat.

What the documentation says

For pages that do use job posting markup, Google's guidance is specific: when a job expires, set its valid through date in the past, remove the page so it returns a not found status, or remove the markup, and if a job is filled early, remove the posting. That handles Google's job listings. It does nothing for an assistant reading the visible text, so change the words as well as the markup. Then ask the assistants what they now say about your hiring, one of the questions in the answer audit, and check again after their next visit. Removing the apply button is not enough if the sentence above it still says you are hiring.

Tomorrow: back to the record.