A government job posting on our board runs 2,637 words. A software posting runs 906. You would assume the government posting tells you almost three times as much about the job.
It doesn’t. Strip the legal and procedural boilerplate out of both and the government posting has 1,014 words of actual job description against software’s 832. Nearly three times the reading, 22% more information.
Across every industry we measured, total posting length varies more than fourfold. The job description inside varies less than twofold. Almost all the difference in how long a posting looks is disclaimers.
Where the words go
Government leads by a mile: 58.3% of a public-sector posting comes after the boilerplate line. Higher education is second at 40.6%. At the other end, aerospace and telecommunications keep it near 16%.
The market-wide picture: the median posting on our board is about 900 words, and the longest 5% run past 2,300. That’s a four-minute read for the typical posting and ten minutes for the long ones, most of the extra spent on text no employer wrote for you.
Length doesn’t tell you whether it’s worth reading
The obvious next question is whether the long ones at least answer more of your questions. Partly, and unhelpfully.
Postings under 500 words state the pay only 25.6% of the time. Postings over 2,000 words state it 90% of the time. So brevity is genuinely worse for you.
But the reverse doesn’t hold either, because the same length means completely different things in different industries. Consumer goods postings run 754 words and state the pay 84.4% of the time. Professional services postings run 780 words, essentially identical, and state the pay 9.5% of the time.
Turn that into the number that matters, which is how much reading you do before one posting answers the first question you have:
| Industry | Median words | States the pay | Words read per posting that tells you |
|---|---|---|---|
| Consumer goods | 754 | 84.4% | ~890 |
| Telecommunications | 1,021 | 89.9% | ~1,140 |
| Retail and e-commerce | 776 | 51.7% | ~1,500 |
| Software and technology | 906 | 38.7% | ~2,340 |
| Government | 2,637 | 95.3% | ~2,770 |
| Construction and real estate | 1,058 | 29.5% | ~3,590 |
| Professional services | 780 | 9.5% | ~8,210 |
A ninefold spread between the best and worst industries, and the worst is not the wordy one. Consulting writes short postings that tell you nothing.
What this says about who wrote them
Put the two findings together and a picture emerges that most job seekers already suspect.
The job description itself, the part a hiring manager actually thinks about, is roughly 700 to 1,000 words almost everywhere. That’s the real content ceiling across wildly different industries, company sizes, and role types. What varies is how much unrelated text gets stapled to it, and whether anyone bothered to include the pay.
The pattern behind every silence we've measured
That consistency isn’t a sign of care. It’s a sign that most postings are assembled rather than written: a requisition template, a paste of last year’s version for the same title, and a compliance block appended by a system nobody reviews. Half the market can’t be bothered to state compensation. Nearly half names no specific benefit. Half won’t say where you’ll work. A posting produced that way is not a considered description of a job. It’s a form.
We found the same machinery from the other direction when we measured benefits language: health, retirement, and paid time off appear at nearly identical rates because they come from the same HR template, and whether a posting tells you anything at all is often decided by one template decision at one large employer.
Which leads to the practical conclusion. If the posting was written carelessly, you cannot treat it as a reliable description of the work, the team, or the company. The gaps aren’t hiding good news. Doing your own research isn’t optional diligence for the paranoid, it’s compensating for the fact that the document you were given was never designed to inform you.
Do the research the posting didn’t
If the posting is a form, the burden shifts to you: assume the gaps hide nothing good and check the company yourself before spending an evening on an application. That is exactly the work worth handing to an AI, and there are two ways to do it without retyping anything. We wrote a full method in Stop applying blind: use AI to vet companies first, and every job on our board now has a one-click version of it, which we walk through in Ask AI: the button you’ve been missing.
Deciding which postings deserve that attention in the first place is the job upstream, and that is what Fitcheck does: it checks every new posting against the criteria you set and returns Apply, Review, or Skip, with the reasons, so the 900 words you do read are attached to a job worth reading about.
Crossover with the rest of the series
This is the mechanism behind the silences we’ve been measuring all week: postings that won’t name pay, won’t state work mode, and won’t name a benefit aren’t withholding strategically so much as being generated without much thought. The whole series lives at the job posting statistics hub.
How we counted
A snapshot of live postings on the JobWeb Board taken July 29, 2026, covering 726,937 postings across 3,356 employers, restricted for this analysis to postings whose text runs between 500 and 60,000 characters. Word counts are derived from character length at six characters per word including spacing, so treat them as close approximations rather than exact counts.
Boilerplate is measured mechanically as the text following the earliest of several standard markers: equal-opportunity and equal-employment statements, reasonable-accommodation notices, E-Verify and selective-service language, and application-procedure sections such as “how to apply” and “required documents.” This is a floor rather than a ceiling, since boilerplate appearing before those markers, such as company blurbs and benefits blocks, is counted as job description. Marker coverage varies by industry, from 97.4% of government postings to 48.9% of software ones, and the per-industry content figures above are restricted to industries where markers were found in a majority of postings. Industries with sparse coverage, including staffing and supply chain, are excluded rather than reported on thin subsets.
“States the pay” counts any parseable compensation figure of any interval. Everything above is a structural measurement: how long a posting is, how much of it is standard legal text, and whether specific facts appear. Judging whether the remaining words are any good is a different exercise, one that requires actually reading the posting, which is precisely the work this post argues you should hand to an AI rather than do by hand.