Only 5.2% say remote, hybrid or onsite; of those, 60% are remote. Agency ads, visa sponsorship, seniority and AI skills, measured on the whole corpus.
A single job ad is a document an employer wrote to get a specific person to click. Six and a half million of them are something else: a record of what employers are willing to put in writing. Some of it is hiring. Some of it is silence.
Here is what the silences look like, measured on every crawled record — the September 2026 edition, full pass, September 15, 2026.
5.2% of postings say anything about where the work happens. Remote, hybrid or onsite, stated plainly — that is the entire group. Of the ads that do state it, 60.2% are remote, 31.9% hybrid and 7.8% onsite.
The 94.8% that say nothing are not hiding a policy. Most are warehouse, nursing and driving ads where the location is the building you show up at, so nobody writes it down. That is the trap in reading remote percentages: the base is not the job market, it is the small share of ads that bothered.
The pay side is where it gets more useful. Ads that state remote also state a number: mean advertised salary 96,158 against 87,399 for the rest, a gap of about 10%. It rests on 15,402 remote ads with a figure, against 641,993 without. Remote-and-published-pay is a small, self-selected set, and the honest reading is that remote jobs that advertise a number advertise a higher one.
6.2% of postings come from staffing agencies rather than the employer. They are the ones who put a figure in the ad: 19.4% of agency postings state pay, well above the 12.4% overall. The same ads also fan out under several employers at once, which is why 9.3% of our records are duplicates of a posting we already hold under another source. When a job seeker says the same role is everywhere, this is the mechanism.
Visa sponsorship appears in 0.7% of postings, and an explicit work-authorisation requirement in 0.4%. Both are rare enough that a job seeker filtering on them is filtering on almost nothing.
Those three cover almost everything, which says more about how ads are labelled than about the roles. Explicit "entry level" appears in 2.8% of postings, and 18.3% ask for a credential of some kind. 76.5% are written in English.
AI gets its own small pile: 2.1% of postings mention it at all, mostly as "machine learning", "AI-powered" or "data science". Two percent is a lot of postings and a small share of the market.
None of this measures whether the job is good, whether the company pays what it says, or whether it will still exist next quarter. It measures the paper. The gap between paper and reality is where most job-search frustration lives, and a crawl can describe the gap without closing it.
What it can do is tell you which signals are worth reading. Location language and stated pay are worth the thirty seconds. Visa sponsorship, at 0.7%, is not a filter worth setting. And the 94.8% who write nothing at all are the most honest group in the dataset: they know the ad is not going to tell you.
→ Key figures · Across borders
Want a cut we have not published? A region, an industry, a role family, a time series, one country against another — we can pull it from the same corpus these posts are measured on.
Tell us what you are looking at: [email protected]. We answer either way.
Source & licence. All figures here come from our own crawl of public job postings, measured against the corpus published in the monthly report and its JSON (CC BY 4.0).
Cite as: DeUnemployment, “What a job ad tells you when you count six and a half million of them”, https://deunemployment.com/blog/what-six-million-job-ads-actually-say (data as of 2026-09-20).