> ## Content Index > Fetch the complete content index at: https://www.notatechguy.com/llms.txt > Use this file to discover other available public pages before exploring further. # Job postings mention AI 66% more, but the work barely changed - URL: https://www.notatechguy.com/job-postings-mention-ai-66-more-but-the-work-barely-changed/ - Published: 2026-11-09T05:56:01.000Z - Updated: 2026-11-09T05:56:01.000Z - Description: A study of 50,000 job ads from 33 firms finds AI mentions up 66%, but most growth sits outside actual work activity descriptions. - Author: Marcello Babbili - Tags: Technology & AI, OpenAI, Anthropic Employers across 33 firms added 'AI' to job ads 66% more often between 2023 and 2025, but most of that growth sat outside the work descriptions [S¹](https://arxiv.org/abs/2611.09870v1?ref=notatechguy.com). The finding, from a preprint posted on arXiv on 6 November 2026, cuts against the popular reading that AI is rapidly rewriting job descriptions from the inside out [S¹](https://arxiv.org/abs/2611.09870v1?ref=notatechguy.com). If the work itself were changing fast, the listed responsibilities would shift. They barely did. **My read:** The 66% figure is the kind of number that gets quoted in a keynote and loses all nuance. What the paper actually shows is more interesting: employers are name-checking AI in job ads without rewriting the underlying work. The signal I would watch is whether the mild decline in AI-exposed responsibilities accelerates in 2026 data, because a gentle dip over two years could be the start of a slow substitution or could be noise. The dataset is narrow, 33 firms is not the labour market, and the exposure measure only captures Copilot and Claude, not the full range of tools on the market. ### What the researchers actually did The team built a pipeline that pulls responsibility statements out of job advertisements and maps each one onto O\*NET's Intermediate Work Activities taxonomy, a US government standard that classifies the component tasks of every occupation [S¹](https://arxiv.org/abs/2611.09870v1?ref=notatechguy.com). They applied it to 50,000 job advertisements from 33 firms spanning 2023 to 2025 [S¹](https://arxiv.org/abs/2611.09870v1?ref=notatechguy.com). To measure AI exposure at the activity level, they drew on usage data from Copilot and Claude, two of the most widely deployed workplace AI assistants [S¹](https://arxiv.org/abs/2611.09870v1?ref=notatechguy.com). This matters because most prior research on AI and labour has worked at the occupation level, asking whether "software developer" or "paralegal" is exposed. Activity-level measurement gets closer to the real question: which specific tasks within a job are changing, and which are not. OpenAI released its own contribution to this line of work on 26 October 2026\. It published "Signals" data covering 165 work activities across nearly 100 countries, drawn from how consumers use ChatGPT for work [P³](https://openai.com/index/chatgpt-expanding-world-of-work/?ref=notatechguy.com). Microsoft separately maintains a public repository, "working-with-ai," with 45 stars on GitHub, accompanying a paper on measuring generative AI applicability to occupations [P⁴](https://github.com/microsoft/working-with-ai?ref=notatechguy.com). The arXiv preprint sits in a fast-growing body of research trying to connect granular AI usage data to what employers actually ask for. ### The number and what it sits against The 66% increase in explicit AI mentions is the headline figure, but the paper's more telling finding is where those mentions appear [S¹](https://arxiv.org/abs/2611.09870v1?ref=notatechguy.com). Much of the growth occurred outside the descriptions of actual work activities, meaning employers are adding "AI" to boilerplate, benefits sections, or company-overview language rather than rewriting the day-to-day responsibilities [S¹](https://arxiv.org/abs/2611.09870v1?ref=notatechguy.com). When the researchers looked specifically at advertised responsibilities, activities with greater AI exposure tended to decline in prevalence, but only mildly as a share of the total number of responsibilities listed [S¹](https://arxiv.org/abs/2611.09870v1?ref=notatechguy.com). At the same time, the AI-exposed activities that remained were increasingly paired with explicitly AI-related responsibilities [S¹](https://arxiv.org/abs/2611.09870v1?ref=notatechguy.com). The authors read this as evidence that employers want to hire people who can perform AI-exposed tasks using AI, rather than dropping those tasks entirely [S¹](https://arxiv.org/abs/2611.09870v1?ref=notatechguy.com). ### What would have to hold The preprint has not been peer-reviewed [S¹](https://arxiv.org/abs/2611.09870v1?ref=notatechguy.com). The dataset covers 33 firms and 50,000 advertisements, a substantial sample but not a nationally representative portrait of the labour market. The AI exposure measures derive from Copilot and Claude usage specifically, so they may not generalise to other tools or platforms [S¹](https://arxiv.org/abs/2611.09870v1?ref=notatechguy.com). The study identifies associations between AI exposure and changing responsibilities, not causal relationships [S¹](https://arxiv.org/abs/2611.09870v1?ref=notatechguy.com). For a hiring manager at a mid-size professional services firm, the practical check is simple: pull your last 20 job postings and count how many times "AI" appears in the responsibilities section versus the company blurb. If the ratio mirrors what this study found, most of your AI mentions are decoration. The question worth asking before the next round of hires is whether the responsibilities you list reflect how the work is actually done now, or whether they are copied from a template written before Copilot existed. The next checkpoint for this research is peer review and, if the authors release their pipeline code, independent replication on a broader set of firms. OpenAI's Signals data, released 26 October 2026, gives other researchers a new source of activity-level AI usage to cross-check against [P³](https://openai.com/index/chatgpt-expanding-world-of-work/?ref=notatechguy.com). --- *Sources: [S1 — AI Exposure and the Changing Responsibilities in Job Postings](https://arxiv.org/abs/2611.09870v1?ref=notatechguy.com) · [P2 — Responsible AI in Open Ecosystems: Reconciling Innovation with Risk As](https://arxiv.org/html/2409.19104?ref=notatechguy.com) · [P3 — ChatGPT’s expanding world of work | OpenAI](https://openai.com/index/chatgpt-expanding-world-of-work/?ref=notatechguy.com) · [P4 — microsoft/working-with-ai](https://github.com/microsoft/working-with-ai?ref=notatechguy.com)* --- *Written from 4 sourced items, 3 of them primary.* ![AI mentions in job ads, 2023 to 2025](https://storage.ghost.io/c/6e/89/6e896869-22ef-4281-a213-b4c462c17cff/content/images/2026/11/chart_8a84292ed2b0aa1fd924.png) ## More from Not A Tech Guy - [Rust compiler repo trends on GitHub at 121,536 stars](https://www.notatechguy.com/rust-compiler-repo-trends-on-github-at-121-536-stars/) - [MonkeyCode AI office workbench trends on GitHub with 5,093 stars](https://www.notatechguy.com/monkeycode-ai-office-workbench-trends-on-github-with-5-093-stars/) - [RBA's Sarah Hunter asks what cheaper AI means for the economy](https://www.notatechguy.com/rba-s-sarah-hunter-asks-what-cheaper-ai-means-for-the-economy/)