The HR Assistant job is 11 per cent reachable by AI, and that is the problem

Nineteen tasks make up the Human Resources Assistant role in the O*NET taxonomy. Two of them can be done with a chat window today.

That is 11 per cent, and it is the second lowest of the nine People occupations in this report. If the question is whether ChatGPT is coming for this job, the published answer is that it mostly has not, and mostly cannot.

Which makes almost everything currently being sold to these people beside the point.

Where the exposure actually sits

The three bands for Human Resources Assistants, Except Payroll and Timekeeping:

• 11 per cent reachable today, which is two tasks
• 68 per cent needs software building first, which is thirteen tasks
• 21 per cent cannot be sped up, which is four

The two reachable tasks are explaining policy and answering questions about benefits and eligibility. Real work, and the smallest part of the job.

The thirteen in the middle band are the job. Processing and verifying personnel documentation. Recording employee data. Gathering records from other departments. Examining files to answer queries. Compiling reports. Processing applications and screening them. Informing candidates. Arranging job postings.

Every one of those is something a purpose-built tool could take on. None of them is something a chat window does, because they are not writing problems. They are move-this-record-from-here-to-there problems, and the tool has to be connected to the places the records live.

The number that makes this unusual

O*NET sorts occupations into Job Zones by how much preparation they need. Zone 3 is the band of jobs that need some training but not a degree. Zone 4 is the graduate band.

Across all 923 occupations in this report, Zone 3 work has a middle band of 30 per cent. That is the average share of work needing purpose-built software for jobs at that level of preparation.

Human Resources Assistants is Zone 3, and its middle band is 68 per cent. Two and a third times the typical figure for its own band, and the fourteenth highest of the 212 Zone 3 occupations in the taxonomy.

This is a job with a Zone 3 entry requirement and a Zone 4 exposure profile.

That combination is rare and it is the whole point of this piece, because of what else is true about the role. In the O*NET incumbent survey, 35 per cent of people doing this job hold a degree, from a sample of 37. Every other People occupation in this report except Payroll and Timekeeping Clerks sits in Zone 4, with degree shares from 67 to 100 per cent.

So this is one of two doors into the People profession that does not require a degree, and it happens to be one of the most software-exposed jobs at its level of preparation in the entire economy.

What a second, independent model says

Digital Planet at Tufts published its own occupational risk index in March 2026, built by combining three separate academic exposure measures and layering on observed use of Claude and Copilot. Across 757 occupations it agrees with this report at r = 0.91, which is worth knowing before quoting either.

In their median scenario, Human Resources Assistants carries the highest projected job loss of the nine People occupations, at 17.2 per cent, or 17,430 of roughly 101,000 US jobs.

Now the part their headline does not carry. Their three scenarios for the same nine occupations give 42,332 jobs, 197,255 jobs and 521,915 jobs. That is a spread of more than twelve times, and their data booklet does not define anywhere in text what separates the three runs. Anyone quoting one of those numbers without the other two is quoting a preference.

Read it as direction rather than forecast. The direction is consistent, and it points at this role.

The risk is not the one being sold

Put those together and the danger is not a redundancy announcement.

Eleven per cent of this job is reachable with tools that exist. The other 68 per cent needs somebody to build something, and mostly nobody has. Building it is not hard any more, which is precisely why it will happen. When it does, the work does not vanish from under the people doing it. It stops being a reason to hire the next one.

Nobody is made redundant. There is no consultation, no announcement, no news story. A role quietly stops appearing in the budget, and a profession that could be entered without a degree can be entered only by graduates. It takes about five years and nobody notices, because at no point does anything happen.

The first thing AI takes out of HR is not a person. It is the way in.

That is a problem for the person in the job. It is a bigger problem for whoever runs the function, because the entry-level role is where the profession makes people, and the degree share tells you exactly whose route it is.

What actually follows from this

The standard advice, which is to learn prompting and put AI on your profile, addresses two tasks out of nineteen. It is worth an afternoon. It is not a plan, and anyone selling it as one has not looked at where the exposure sits.

The 68 per cent is waiting on a specification, not on a model. Somebody has to write down what the tool needs to do: which records, from which systems, in what order, with what exceptions, and what happens when the reference does not come back. That document is the scarce thing, and building the software from it is now the cheap part.

The person best placed to write it is the person doing the job. They are the only one who knows that the fifth line of the process is where it always breaks.

That is not a comforting answer, because it asks for something harder than a course. It is the accurate one. The choice this data actually presents is between specifying the tool and being specified by it.

What this piece cannot tell you

The ratings were made in 2023 against the model capabilities of that year, so 11 per cent is a floor and not a current reading. Nineteen tasks is a small denominator, and one task moving band is five points. The Tufts projections are a model, not a measurement, and their own range spans more than an order of magnitude. Nothing here is a prediction about any individual job, and no figure in it should be quoted as one.