A job posting is written for a candidate, not for a competitor. That is exactly why it is useful. Where an annual report tells a story about the future, a job posting describes what is actually missing from a team this quarter. Is a competitor still looking for a planner who manually drags schedules around every day, or are they looking for a planner who reviews exceptions that a system has already proposed? That difference in task description is a difference in where the work has already shifted from human to machine, with oversight, or still rests entirely with the human.
The reason this works is that job postings are close to day-to-day practice. A marketing department can write in an annual report that it is at the forefront of automation; a job posting for a content employee that still asks for "writing product descriptions from A to Z" contradicts that. The reverse also holds: a vacancy for a "reviewer who evaluates AI-generated drafts for brand fit" is a concrete signal that the first version of that work is no longer done by a human at all.
The three categories that currently run through every field of work also show up in job descriptions. A task that AI can fully take over disappears from the vacancy or is replaced by a task at a different level. A task that is partly taken over with human oversight remains but changes wording: no longer "drafting", but "reviewing", "correcting", "approving". And work that remains human work stays in the text exactly as it always did, with the same emphasis on experience and personal judgment.
This shift explains why the same job title can be filled completely differently at two competitors. A customer service employee at one company resolves tickets independently; at another, they review draft answers that a system has already prepared. Both vacancies are called "customer service employee". Only the task description shows which part of the work has already moved. Exactly where that difference comes from depends on how mature the underlying systems are at that employer and how much oversight that employer currently deems necessary, not on a fixed standard for the sector.
A job posting is written to recruit, not to report. That makes it just as distortable as an annual report, only in a different direction. A company that wants to make a position attractive can describe tasks that are in fact routine and largely automated as "strategic" or "analytical" to make the role sound more interesting. Conversely, a vacancy that uses little AI language may simply have been written by a recruiter who is not aware of what the department itself has already implemented.
A second pitfall is timing. A job posting is a snapshot of a single vacancy at a single moment; it says nothing about how many positions of that type a company is opening or closing in total. One vacancy for an "AI reviewer" does not prove a company-wide shift. Only a pattern across multiple vacancies, multiple quarters and multiple departments starts to become a reliable signal.
A third pitfall: the number of FTEs a competitor hires or does not hire says nothing in itself about why. What an employer does with its workforce falls under its own legal requirements and its own decision-making; a job posting provides no substantiation for that, nor is it meant to.
For a competitor that changes slowly, going through the vacancies once a year is often enough to keep the picture current. For a competitor in a market where the dimension that decides deals shifts extremely fast, such as delivery time or response time, looking more often is more worthwhile, because a job posting often changes sooner than an annual report or a price list. Job postings are also not a replacement for other sources, but a supplement: what annual reports report about strategy, what price lists show about margins and speed, and what tender documents reveal about delivered capacity only together with job postings give a complete picture of where a competitor stands.
A single signal from a job posting only becomes valuable once it is placed alongside the other signals and held up against the dimensions that actually decide deals in your market. How you do that without assumptions is described on the page about scoring without assumptions, and how you place that next to your own position is described on the page about benchmarking against competitors. The underlying question of which work in your own company can truly be taken over by AI is answered by FTE TO AI's work scan per task.
You can start by placing the job postings of your two or three most important competitors from the past year side by side and noting which task descriptions have disappeared, which have changed from "performing" to "reviewing", and which have remained unchanged. If you want to know whether what you infer from that also holds up, you can use the free dimension check: you name what you believe you are winning on, and you see which of those claims can be defended with evidence. The full benchmark, with the evidence matrix and the peer group built in, is under construction.