OpenAI's Own Data Shows HR Teams Use ChatGPT for Almost Everything Except HR

HR professionals are using ChatGPT to do work that sits outside their own job description more often than almost any other profession, according to an OpenAI analysis reported by HR Dive. The company analysed more than 800,000 messages submitted by US-based users and found that about 44% of all occupation-specific messages involved tasks not included in the user's occupation. Among HR professionals that share reached 69%. At Metaintro we read findings like this as a signal about job design rather than about software adoption, because a number that high usually means the role has quietly expanded beyond what it says on the contract.

What did OpenAI actually find?

The headline comparison is the useful part. Across all occupations studied, roughly 44% of occupation-specific messages concerned tasks outside the user's own field. For HR that figure was 69%, which placed the profession third overall, behind only design and customer experience roles. That is not a marginal difference. It means that for every ten work-related questions an HR practitioner brought to the tool, close to seven were about something other than HR.

The breakdown of what that outside work consisted of is equally telling. Marketing accounted for 23% of occupational messages submitted by HR practitioners. Engineering followed at 18% and finance at 16%. Those are not adjacent HR disciplines. They are three separate functions, each with its own team somewhere in most organisations, and HR people were doing pieces of all of them.

OpenAI framed the finding as evidence that AI can change the division of work inside an organisation, by allowing people to perform work themselves rather than handing it to someone else. The company also noted that task crossover appeared more frequently among users at small businesses, which likely reflects a tendency in smaller organisations to solve a problem directly rather than delegate it. That caveat matters, because it separates two very different situations that the headline number blends together.

Why is HR doing so much work that is not HR?

Because the function has been absorbing scope for years, and the pattern predates the technology entirely. HR teams have long been asked to perform work that falls outside HR, and the trend accelerated after the pandemic, when organisations frequently called on HR to stand up remote work policies, wellbeing programmes and a range of other initiatives that had no obvious owner elsewhere. Once a function acquires a reputation for picking up unowned problems, it tends to keep acquiring them.

The strain that produces is measurable. In data published in March 2025, Perceptyx found that more than 40% of surveyed HR leaders said their jobs had become more difficult during the preceding year, and 30% had considered leaving the profession altogether. A SHRM report released around the same time found that 57% of HR professionals believed their departments were understaffed. Those two findings together describe a function that is stretched rather than merely busy.

Read against that backdrop, the 69% looks less like enthusiastic experimentation and more like coping. An understaffed team handed marketing, engineering and finance tasks has two options. It can decline the work, which is politically difficult and often career-limiting, or it can find a way to produce something adequate quickly. A language model makes the second option viable for the first time. That is a genuine productivity gain, and it is also how a job description quietly stops describing the job.

Is this good news or bad news for HR careers?

Both, and which one it becomes depends almost entirely on whether the expanded scope is acknowledged. The optimistic reading is straightforward. An HR professional who can draft a campaign, interrogate a budget and hold a credible conversation with an engineering team is more valuable than one who cannot, and that breadth is exactly the profile that moves into senior operating roles. There is a real argument that the function could grow into something closer to a chief work officer, responsible for deciding which tasks are best performed by automated systems and which should stay with people. That is a considerably larger job than administering policy.

The pessimistic reading is equally plausible. Scope that arrives without headcount, budget or a title change is not a promotion. It is unpaid expansion, and it is the mechanism by which roles become impossible. If 57% of the profession already believes it is understaffed, absorbing three more functions through a chatbot does not fix the staffing problem. It conceals it, and a concealed problem does not get resourced. Our reporting on why workers push back on AI, and why the reason is rarely the technology itself covers the resentment that builds when tools are used to absorb overload rather than reduce it.

The difference between those outcomes is visibility. If nobody knows the marketing work is being done by HR, then HR gets neither the credit nor the resource, and the only person who knows the true scope of the role is the person exhausted by it. This is the single most important practical point in the whole study.

Does this mean AI is replacing HR work?

Not on this evidence, and the distinction is worth holding onto because the headline invites the opposite conclusion. Nothing in the data shows a model performing HR tasks in place of an HR professional. It shows HR professionals performing other people's tasks with a model's help. Those are close to opposite phenomena. One is substitution. The other is expansion, with the human firmly in the loop and taking on more, not less.

That distinction changes what you should worry about. The risk in this pattern is not redundancy, it is dilution. A specialist who spends most of their week producing adequate output in three other disciplines is not building depth in any of them, including their own. Over a few years that can erode the thing that made the person valuable in the first place, and it does so invisibly, because the work is getting done and nobody is complaining.

It also raises a quality question that organisations rarely ask out loud. Work produced quickly by a capable person outside their specialism is usually good enough to pass and rarely as good as what the specialist team would have produced. That trade is sometimes correct, particularly under deadline. It becomes a problem when it turns into the default operating model without anyone deciding that it should. The honest version of this conversation is about where good enough is genuinely fine and where it is not, and that is a leadership decision rather than a tooling one.

How should you use this in your own career?

Start by documenting the crossover rather than absorbing it silently. If a meaningful share of your week goes to work outside your function, write down what it is, roughly how long it takes and what it produced. Not as a complaint, but as a record. That record is what turns an invisible expansion into a negotiable fact, and it is the difference between a scope conversation and a vague sense of being overworked. Our guide to comparing total compensation rather than just salary is a useful companion here, because expanded scope is a compensation question even when it never appears as one.

Second, be deliberate about which crossover you take on. Doing finance work makes you more valuable if it builds toward a role you actually want, and simply makes you busier if it does not. The three areas showing up in the data, marketing, engineering and finance, have very different career implications for an HR professional. Commercial and financial literacy travels well into senior people roles and general management. Ad hoc marketing production usually does not. Choosing on that basis is one of the few genuinely high-leverage decisions available in a stretched job.

A fourth move is to check whether the crossover is actually solving anything. If you are producing marketing output because the marketing team is under-resourced, the organisation has a capacity problem that your effort is currently hiding. Naming that plainly, once, in a conversation with your own manager is usually more valuable than another quarter of quietly absorbing it, and it is a materially different conversation from asking for help.

Third, invest in direction rather than delegation. A study by researchers at the University of Texas at Austin and KPMG found that simply prompting an AI system is distinct from the ability to direct one to perform complex tasks, and that workers who excel at collaborating with the technology, providing direction, oversight and judgement rather than solely handing tasks over, tend to outperform their peers. That is a specific, learnable skill and it is the one that holds its value. Our reporting on why mentoring matters more in the AI era makes a related point, which is that judgement is developed through people rather than tools.

What does this mean for how organisations design jobs?

It means job descriptions are now a lagging indicator, and leaders should treat them that way. If the average worker is spending a substantial share of their AI-assisted effort outside their nominal function, and HR specifically is spending 69%, then the formal structure of the organisation no longer describes how work actually flows. That gap has consequences for hiring, because you end up recruiting against a description that does not match the job, and for retention, because people leave roles that quietly became something else.

There is also a workforce planning consequence that most organisations have not caught up with. If HR is producing marketing output, the marketing team's true workload is not what its own numbers say. Capacity is being borrowed across functional lines without appearing in any plan. Our analysis of the skills mismatch that defines the AI workforce debate argues that the mismatch is rarely about raw capability, and this data supports that. The problem here is not that HR lacks skills. It is that the organisation has no visibility into where the work is going.

One further consequence is worth naming, because it lands on the people least able to push back. Junior staff in stretched functions are the most likely to be handed the crossover work, and the least able to decline it. A senior practitioner asked to produce a marketing campaign can negotiate. Someone two years into their career usually cannot, and will simply do it. Over time that produces a cohort with broad shallow exposure and thin professional foundations, which is precisely the profile that struggles when the market tightens and employers start hiring for demonstrable depth.

The fair caveat is the small business one. In a company of thirty people, someone in HR doing marketing is normal, sensible and not evidence of dysfunction. OpenAI specifically noted higher crossover among small business users. The finding should worry large organisations with dedicated functions far more than it should worry small ones, because in a large company the same behaviour means the handoffs between teams have stopped working. Our reporting on how employers are chasing increasingly specific skills in hiring shows the tension clearly, with organisations recruiting for narrow specialisms while their existing staff quietly go broad.

People Also Asked

Q: How much of HR's ChatGPT use is for non-HR work?

A: About 69% of occupation-specific messages submitted by HR professionals involved tasks outside their own occupation, according to OpenAI's analysis of more than 800,000 messages from US-based users. The average across all occupations was about 44%.

Q: What non-HR tasks are HR professionals doing?

A: Marketing accounted for 23% of occupational messages from HR practitioners, followed by engineering at 18% and finance at 16%.

Q: Is HR understaffed?

A: A SHRM report found that 57% of HR professionals believed their departments were understaffed. Separately, Perceptyx data published in March 2025 found more than 40% of HR leaders said their jobs had become more difficult over the preceding year, and 30% had considered leaving the profession.

Related Articles

• The AI Workforce Skills Mismatch Is Not a Job Apocalypse
• The Real Reason Workers Push Back on AI
• Why Mentoring Matters More in the AI Era
• Older Workers and the AI Upskilling Edge
• Leadership Upskilling Beats Culture Change
• The AI Skills Worker Shortage
• The AI Skills Salary Premium
• Could AI Raise Your Salary
• Employers Chase Increasingly Specific Skills
• Compare Total Compensation, Not Just Salary
• How AI Is Reshaping HR Decisions on Pay

Future-proof your career. The most valuable skill in this data is not prompting, it is knowing which expanded scope is worth taking on and which is quietly costing you, and that judgement is easier to make when you can see what the wider market values. Metaintro matches your profile against roles worldwide so you can benchmark where your real skill set sits, and you can create your free account here to start.