Job titles are supposed to describe what a person actually does at work. New data from OpenAI suggests that assumption is breaking down faster in human resources than in almost any other function, and the tool doing the breaking is the one HR teams already have open in a browser tab.
The finding: job titles are getting less predictive of job tasks
OpenAI analyzed more than 800,000 work related ChatGPT messages from U.S. business users, classifying each request against the U.S. Department of Labor’s O*NET occupational database, which maps specific tasks to specific job profiles. Common activities like writing, summarizing, and scheduling were excluded so the analysis could isolate specialized, occupation specific work.
The result: 43.5 percent of occupation specific requests involved tasks that O*NET assigns to a different job than the one the requester actually holds. Among HR professionals, that crossover rate was 69 percent, the third highest of any occupational group OpenAI studied, trailing only customer experience workers (77 percent) and designers (75 percent). When HR professionals used ChatGPT for work outside their own function, the largest shares went to marketing tasks (23 percent), engineering tasks (18 percent), and finance tasks (16 percent).
OpenAI framed the shift plainly: “AI can change the division of work within an organization by allowing users to perform work using the tool themselves rather than handing it off to others.” A separate summary of the report noted that OpenAI treats the usage data as an early signal that job profiles are shifting even before job titles or descriptions catch up.
Why HR specifically
HR is a plausible place for this to show up first. HR teams, especially at small and mid-size companies without dedicated specialist support, have long absorbed adjacent work by necessity: drafting internal comms that read like marketing copy, building headcount models that lean on basic financial modeling, evaluating HR technology vendors in ways that overlap with procurement and engineering due diligence. The difference now is that AI lets an HR generalist execute that adjacent work directly rather than routing it to a specialist and waiting. Reporting on the same OpenAI data found the crossover effect was stronger at smaller companies that lack dedicated specialist teams for functions like design or engineering, which tracks with what HR departments have dealt with for years.
What it means for the HR leader
This is not a story about ChatGPT usage statistics. It is a story about whether the scaffolding HR built around fixed job scope, job architecture, leveling frameworks, comp bands, still describes the work being done underneath it.
Three practical exposure points follow directly from the data:
Job architecture and leveling. If an HR generalist is regularly doing finance modeling and a designer is regularly doing engineering troubleshooting, job descriptions and comp bands built around narrow, fixed task lists are measuring something that no longer matches reality. HRTech Edition has covered how leadership readiness, not tooling, is the real bottleneck to AI adoption, and this data adds a second layer: even employees who are technically capable are already operating outside their formal scope, whether HR has measured it or not.
Governance and tool access. Security and data governance policies for generative AI tools are still frequently scoped by department, on the assumption that marketing uses the marketing tools and engineering uses the engineering tools. If 44 percent of occupation specific AI use already crosses those lines, governance built around departmental silos is auditing the wrong boundary.
Performance management. Review cycles and manager check ins are built around the job description on file. When a chunk of an employee’s actual output falls outside that description, and OpenAI’s data says that is now common rather than rare, performance conversations risk missing real contributions or, worse, penalizing time spent on unsanctioned but genuinely useful work.
The measurement gap
Most HR technology stacks currently answer “who uses AI” at the level of tool seats and department assignment. OpenAI’s task level data argues that is the wrong unit of measurement. The more useful question is what work is actually being done with the tool, regardless of who is nominally assigned to do it. That is a harder question to answer, but it is also where HRTech Edition has previously reported HR’s own confidence in AI governance is under strain: Culture Amp’s research, covered here, found HR buy in on AI is slipping even as adoption grows, with governance gaps identified as a bigger obstacle than budget or tooling. Task crossover is exactly the kind of blind spot that kind of governance gap produces.
What to do next
HR and people analytics teams do not need to wait for a formal org redesign to respond to this data. Four steps are available now:
First, audit AI usage logs rather than relying on seat counts or department rosters. Seat assignment shows who was given access; usage logs show what work is actually happening and where it crosses functional lines.
Second, revisit job architecture and leveling with actual task data, not the job description on file, as the input. Where crossover is concentrated, that is a signal for either formal cross training investment or a redesign of how the role is scoped and compensated.
Third, extend AI governance and access policies past the functions traditionally seen as heavy AI users. The data shows HR itself, along with customer experience and design, absorbs more outside work through AI than most departments, which means HR cannot assume its own AI use is low risk simply because it is not engineering or finance.
Fourth, loop learning and development into this data rather than leaving it as an IT or security exercise. Cross functional AI literacy, not department siloed training, is the direct response to a workforce that is already operating across job boundaries with or without formal sign off.
Source: OpenAI
