Prompts to Analyse Competitor Job Postings

A careers page is a list of gaps. Every role on it is work somebody decided needs doing and has not found a person for yet. The moment a role is filled it disappears, and so does any trace that it was ever there.

That gives the source an odd shape. It is the only competitor record that describes intentions rather than actions, and it is also the only one that deletes its own history. Six months of hiring leaves nothing behind unless you saved it.

The consequence is the one that matters for reading. A company building a new team and a company replacing people who left publish identical pages. Both show five open sales roles. Asked which it is, a model will pick the interesting answer, because growth is the story the words of a job advert are written to tell.

Open two postings from the same company and read them side by side. The opening paragraph about the company is the same. The culture section is the same. The benefits, the equal opportunity statement and the closing pitch are the same.

On a board of a hundred roles, that paragraph appears a hundred times. It was written once, by one person, probably a while ago. Paste twenty postings into a model and ask what recurs, and the recurring thing is the template.

The output that comes back is confident and reads like a finding. Consistent emphasis on moving fast. A settled way of describing themselves. Neither is evidence of anything, because the repetition is a copy-and-paste rather than twenty separate choices. Count roles, and strip the shared text before you count anything at all.

The line that appears once is the one worth having

There is an uncomfortable consequence. If repetition is mostly template, then the informative material is whatever appears rarely, and every frequency-based reading throws exactly that away.

A single enterprise sales posting defining its territory as organisations of five hundred to five thousand employees tells you where a competitor draws the line between mid-market and enterprise. Their website will never put a number on it. That line appeared once, so a theme count discards it and a careful read finds it. Ask for the specifics that appear only once, by name, or you will not get them.

Three passes, and the second one changes what the first is worth. Stripping the template is a mechanical job and it takes one instruction.

The third prompt asks a question the second capture makes possible and nothing else does: which roles closed. A closed posting can mean filled, cancelled or expired, and the board does not say which. The prompt is told to report that rather than choose.

Two inputs, and the second one is the difference between a description and a signal. Nobody keeps an archive of a job board, so if you did not save it, it is gone.

Paste the postings in full, including the parts you would skip as a candidate. Employing entity, security requirements, language fluency and the pay field all sit in the sections nobody reads, and those are the sections written to satisfy a rule rather than to sell the job.

We took what a real careers board shows on its first screen, and then two full postings from it, and asked the obvious question of each.

The pattern was in the sort order

Six roles, all sellers, spread across five countries. Read alone, that is an international expansion. Read against the board, it is the first page of a hundred and thirty-seven openings, and the list begins with sales.

Nothing in the answer is wrong about the six. The error is in what the six are taken to represent, and it is the same error any analysis makes when the denominator stays outside the context window. Say how many roles are open before you paste any of them, and the model can qualify its own answer.

Everything above describes a board as it stood on the afternoon you copied it. A week later three roles have closed and two have appeared, and the closed ones leave no record at all. That makes this the most perishable source in competitive work.

The loss is specific rather than general. What you cannot reconstruct later is how long a role stayed open, and that is the one thing a board tells you about difficulty. A seller posted for one week and a seller posted for five months are very different facts about a territory, and both look identical in a single capture.

No amount of watching answers the question underneath all of this. Who they actually hired, what they were paid and whether the role survived contact with the business are private, and the board goes quiet at exactly the moment the answer would become interesting.

Flares tracks competitor job boards and keeps each role with the date it appeared and the date it went. A run of openings in one segment then arrives as a pattern you can date, rather than as a page you happen to be looking at.