One in twenty HR job ads requires AI skill. Fewer than one in five mention it.

Source. LinkedIn job listings pulled on 2 and 3 August 2026 in Central European Summer Time, covering ads LinkedIn reported as posted in the preceding seven days, so the window runs from 27 July to 3 August 2026. See the note on the two sources below.

Role families. HR leadership covers CHRO, Chief People Officer, VP HR, VP People, Head of HR, Head of People and HR Director. Business partners cover HR Business Partner, HRBP, People Partner and People Business Partner. Generalists cover HR Generalist, HR Specialist, People Operations, HR Advisor and HR Officer. Recruiters cover Recruiter, Talent Acquisition in all its variants and Sourcer. Learning and development covers Learning & Development, Instructional Designer, Training Manager, Training Specialist and Talent Development. Where a title names two families, for instance Talent Acquisition & HR Business Partner, the family whose words come first in the title takes the ad, and the ad is counted once.

Two sources. The three HR families were pulled from LinkedIn's job index through the Apify actor fantastic-jobs/advanced-linkedin-job-search-api, which offers a server-side full-text filter and a built-in staffing-agency exclusion. Recruiters and learning and development were pulled later, through Monid's LinkedIn job search, because the first source had hit its monthly quota. That second source has no agency filter, so agencies were removed here by employer name and industry, which is a blunter instrument. It also has no server-side text filter, so the full ad body was pulled and matched locally against the AI term list before classification. Treat comparisons between the two groups as directional rather than exact, particularly for recruiters, where residual agency listings are the likeliest source of error.

Geography. The United States plus the United Kingdom, Germany, France, the Netherlands, Sweden, Denmark, Norway, Finland, Spain, Italy, Ireland, Belgium, Poland, Switzerland, Austria, Portugal, Czechia, Romania, Hungary and Greece.

Duplicates are gone. Employers routinely publish the same opening across many locations, and LinkedIn sometimes returns one posting under two different titles. Rows are collapsed first on the LinkedIn posting id and then on employer, title and country, which leaves 3,350 distinct ads. Where one posting had been coded twice, the stronger claim on the candidate is the one kept. Every number on this page counts an opening once, no matter how many times LinkedIn carried it or under how many titles.

Why seven days. Seven days answers the question of what is live right now, in one clean snapshot. A thirty-day pull would cost roughly four times as much, mix expired ads with live ones, and change the shares very little.

Detection runs in sixteen languages. Every one of the 3,350 ad bodies was matched locally against an AI vocabulary covering English, German, French, Spanish, Portuguese, Italian, Dutch, Swedish, Danish, Norwegian, Finnish, Polish, Czech, Romanian, Hungarian and Greek, including the acronyms KI, IA and AI, and the named assistants. Acronyms are matched case-sensitively and only in the language regions where they mean artificial intelligence, which keeps the Italian preposition "ai" and the French "j'ai" out. That pass recovered 43 ads the original English-only search had missed, concentrated in Germany, France and Spain.

Classification. Every match was read in its entirety and classified against a fixed category set, then quality-checked in an independent pass that verified each quote appears verbatim in the original and that the category holds. A further 73 ads were set aside as application-process boilerplate, and thirty-five matches were discarded as false positives, among them the Italian preposition "ai", the French "j'ai", company codes containing IA, scraped page furniture, and the string "anthropic" hiding inside "philanthropic".

Where the numbers still strain. The language gap is closed but two limits remain. Countries below roughly forty ads, Ireland, Austria, Romania, Greece and Sweden, carry wide error bars and should be read as directional rather than precise. And an ad that gestures at AI without naming it in any of the sixteen languages, for instance by describing automation or intelligent tooling in the abstract, is still invisible to this method. The United States, the United Kingdom and Germany are the three steadiest series.