An AI Line in a Job Ad Is Not an AI Job

A bakery manager vacancy can mention artificial intelligence without asking the next manager to train a model. The employer may want someone who can question a forecast, try an inventory assistant, or notice when a suggestion should be ignored. That is a different job from a machine-learning post that names the same word. A new count of October advertisements is useful only if the denominator stays attached.

17.7 percent of a file, not of the economy

Qarera, in an update published with October 2026 data, says AI was named in 17.7 percent of 664,812 job postings first seen between 1 and 31 October 2026. On that page, AI means any mention: AI, generative AI, large language models, machine learning or agentic AI. The set covers postings with 13 countries broken out. About 62 percent came from employer career sites, and the rest mostly from LinkedIn and Indeed. Cross-posted copies of the same job were counted once. The page says the October edition covers far more non-technology jobs than the first edition and uses a newer skill extractor, so the percentages should not be compared directly with that earlier file.¹

The first edition, 360,336 postings from 27 December 2025 to 16 June 2026, found AI in 19.8 percent of postings. That file was mostly technology roles, which is why the page says a higher Python or AI share there is not a trend line. A pattern the page calls robust in both editions is that senior postings name AI more often. In October, 43.2 percent of Principal-level postings named it, against 14.3 percent at entry level. That is still a share of advertisements, not a hiring census.¹

Online advertisements are not all jobs. Some roles are filled without a public ad. One employer can list a skill it never tests. An extractor can still catch a mission statement. Qarera says each figure is how often a skill is requested across postings, not the number of open roles or hires. 17.7 percent is not a national hiring rate.

The role split is the part worth keeping. In the October file, AI was named in 94.7 percent of data scientist and machine-learning postings (5,539 jobs), 52.7 percent of product-manager postings (4,991) and 50.0 percent of software-engineer postings (14,163). It was named in 5.5 percent of nurse postings (6,878), 5.3 percent of teacher postings (3,052), 9.8 percent of customer-service postings (8,306) and 14.1 percent of accountant and finance postings (8,116). Communication was still the most common skill in the whole file, at 53.1 percent.¹

A small employer that copies a technology job title because "half of ads say AI" is reading the wrong row.

Three advertisements, three different skills

Consider three hypothetical ads. A developer is asked to integrate a model and watch failure rates. A recruitment coordinator is asked to use a writing assistant, check for bias and protect applicant data. A shop supervisor is asked to read an assisted sales forecast before ordering food that spoils. All three can contain the word AI. The first needs engineering and testing. The second needs process knowledge, privacy awareness and human review. The third needs local judgment. Counting each as one AI-related advertisement can be fair for a keyword indicator. Treating them as equal demand for specialist talent is not.

"Uses AI" is not a test. "Checks an AI-drafted customer reply against policy before it is sent" can be tested. So can "explains when a forecast should be overridden and writes down why." Those lines also give an applicant without a formal AI credential a way to show related experience. The flood of applications that mention a tool without showing the work is a separate problem, covered in why the application pile hides real skills. Judgment that a coding test does not capture is covered in why AI hiring needs judgment, not only code.

A different dataset is not a second reading of 17.7 percent

PwC's 2026 Global AI Jobs Barometer, in a press release of 15 June 2026, says the study analysed more than one billion job advertisements in 27 countries and territories. The release says jobs requiring specific AI skills, such as prompt engineering or machine learning, are growing almost eight times (69 percent) faster than the total jobs market (9 percent). A later sentence says they are growing roughly eight times (69 percent) as fast as the overall jobs market, at 9 percent. Those sentences do not print the words "since 2019". A supplied draft added that date. This article does not. Dividing 69 by 9 is about 7.7, which is how "almost eight times" is usually built, but the page's own wording is the claim, not a rewritten "69 percent growth since 2019".²

The same release says the average wage premium for workers with AI skills hit 62 percent, up from 57 percent in the previous barometer. The report PDF describes the premium as the difference between salaries offered for roles that require AI skills and salaries offered for comparable roles that do not, averaged across 16 sectors. It says the analysis is not a guarantee attached to one person who adds a keyword to a CV. PwC's "specific AI skills" and Qarera's "any mention" are different definitions. They should not be added, and 62 percent should not be read as the value of writing "AI" into a bakery ad.²³

Randstad's release of 29 October 2026, "Tech talent in transition", combines a survey with a separate posting analysis. The survey was 1,100 interviews, 100 in each of 11 markets, in the first half of August 2026, among senior business and technology decision-makers responsible for workforce planning and digital talent strategy. A bullet says 61 percent of employers report that finding talent with communication and stakeholder-management skills is harder than sourcing technical skills alone. A later section says 61 percent report that finding talent with human capabilities such as critical thinking and communication is harder than sourcing technical skills alone. Those are two sentences on one page. This article does not merge them into a single skill list. The 61 percent is a perception in that selected group. It is not a count of vacancies in the whole economy. The secondary file, compiled in August 2026, covers more than 20 million job postings and about 100 million candidate profiles. The 61 percent is not a result from that file.⁴

Taken together, the sources say AI language is spreading, and that some employers say human capabilities are hard to hire. They do not say AI caused every change in the ads, or that an advertisement becomes a hire.

What a small organization can test

A ten-person nonprofit should start from the work, not from a technology title. Which activities take staff time? Which decisions need professional judgment? Where would a wrong or biased output harm a client? Which data must not go into an unapproved service? Only then does it make sense to recruit, retrain, buy a service, or leave a task alone.

A hypothetical community group might find that staff spend hours on grant drafts. Teaching two people to structure sources, use an approved assistant, check numbers and get a final sign-off can be enough. Hiring a dedicated specialist may be out of proportion. An organization that uses a system to decide who is eligible for a benefit is a different case, because the harm from an error is larger. The scenes are illustrations, not Qarera rows.

A short hiring exercise can stay close to the role. Give a fictional work sample, allow one permitted tool, and ask for the result plus a record of the checks. Look at the decision, the handling of uncertainty, the treatment of sensitive information and the explanation of corrections. Keep tool access equal, and offer an alternative where accessibility requires it. The same steps can show a training gap for current staff. A progression from "recognizes a bad use" to "does a bounded task with review" to "can supervise the workflow" is an editorial sketch, not a validated certificate.

For a European employer, a writing assistant in a job ad is not itself a high-risk system. Annex III, point 4(a), of the EU AI Act lists systems intended to be used for recruitment or selection, including placing targeted job advertisements, filtering applications and evaluating candidates. Point 4(b) covers systems intended to monitor and evaluate performance and behaviour. The Commission service desk says Annex III rules apply from 2 December 2027. A mention of AI in a vacancy is not, by that sentence alone, such a system. Classification depends on the intended purpose. This paragraph is not legal advice.⁵⁶

Employers should also look at what happens after the hire or the training. Error rates, rework, time spent checking, complaints and confidence belong next to any time saved. A faster first draft that doubles the verification can erase the gain. That reinvestment problem is set out in who gets the hours a faster task appears to release.

How hilighter can help

hilighter does not score job advertisements, certify an AI skill, or tell an employer that 17.7 percent of roles must be re-titled. It helps a team write down the work and the evidence a person can show.

hilighter helps teams map skills by role, compare people with those requirements in Skills & Talent Insights, record proof of capability in Skills Evidence & Recognition and set personalized growth paths. Its AI Job Impact view is indicative. It is not a reading of Qarera, PwC or Randstad. A free plan covers up to three job roles, with unlimited employees.

Does 17.7 percent mean 17.7 percent of jobs are AI jobs?

No. It is the share of 664,812 postings first seen in October 2026 in Qarera's file that named AI in any of the forms the page lists. Cross-posts were counted once. The page says the figure is not the number of open roles or hires. A nurse posting and a machine-learning posting are both inside the 17.7 percent when they mention the word, and the role table shows they do not mention it equally often.¹

Can the 17.7 percent be compared with the earlier 19.8 percent?

Not as a trend. Qarera says the October file includes many more non-technology jobs and uses a newer extractor, so direct percentage comparisons with the first edition are inappropriate. The first edition was mostly technology postings.¹

Does the 62 percent wage premium apply to anyone who writes AI on a CV?

No. PwC's release reports an average premium of 62 percent for workers with AI skills, up from 57 percent in the previous barometer. The report describes it as a difference in offered salaries between roles that require AI skills and comparable roles that do not, across the sectors in the study. It is not a promise that a keyword raises one person's pay, and it is not Qarera's "any mention" definition.²³

Sources

1. Qarera, The Most In-Demand Skills of 2026, October 2026 update. 664,812 postings first seen 1 to 31 October 2026; AI named in 17.7 percent, defined as any mention of AI, generative AI, large language models, machine learning or agentic AI; 13 countries broken out; about 62 percent from employer career sites; cross-posts counted once; newer extractor, so percentages are not a trend against the first edition. Principal 43.2 percent, entry level 14.3 percent. Role shares and posting counts as in the table above. Communication 53.1 percent. First edition: 360,336 postings, 27 December 2025 to 16 June 2026, AI in 19.8 percent, mostly technology roles. The page says each figure is a share of postings, not open roles or hires.
2. PwC, 2026 Global AI Jobs Barometer press release, 15 June 2026. More than one billion job advertisements in 27 countries and territories. Jobs requiring specific AI skills growing almost eight times (69 percent) faster than the total jobs market (9 percent); later, roughly eight times (69 percent) as fast, at 9 percent. The words "since 2019" are not on those sentences. Average wage premium 62 percent, up from 57 percent. Not the same definition as Qarera's any-mention count.
3. PwC, 2026 Global AI Jobs Barometer (PDF), "Workers with AI skills command a 62% wage premium". Premium described as the difference in offered salaries for roles that require AI skills versus comparable roles that do not; average across 16 sectors; not an analysis of one person's keyword. The Belgium-hosted PDF was the copy read on 8 November 2026.
4. Randstad, 86% of employers are growing tech teams as AI reshapes demand, 29 October 2026. Survey: 1,100 interviews, 11 markets, 100 per market, first half of August 2026, senior decision-makers on workforce planning and digital talent. Bullet: 61 percent on communication and stakeholder management. Later section: 61 percent on critical thinking and communication. Not merged here. Secondary research: more than 20 million job postings and about 100 million candidate profiles, August 2026. The 61 percent is the survey, not that file. The chief executive's remarks that use a contraction are not quoted.
5. European Commission, AI Act Service Desk, Annex III, point 4(a) and 4(b). Read on 8 November 2026. Not legal advice. A vacancy that mentions AI is not, by that text alone, a listed system.
6. European Commission, AI Act Service Desk, When does enforcement start?. Annex III rules apply from 2 December 2027. Prohibited practices, transparency requirements for certain systems, and general-purpose AI model rules are enforceable from 2 August 2026.

Sources last reviewed: 8 November 2026. Qarera's 17.7 percent is a share of its October file under an any-mention definition. PwC's 69 percent and 9 percent lines are a different definition, and "since 2019" was not on the sentences read. The two Randstad 61 percent sentences name different skills. The wage-premium method is from the report PDF, not a promise to one candidate. EU dates follow the Commission service desk. This article is not legal advice. Neither an AI-detector score nor the absence of Google penalties can be guaranteed.