Why the judgment part stays with people
Will AI replace historians? Look at what the job actually involves and the answer gets clearer. Language models read fast and draft well, so the reading and summarizing side of historical work is exposed. The part that decides whether a claim holds up is different. Determining the authenticity and significance of a document, and weighing it against the rest of the record, is the core skill the job is paid for.
Much of the raw material is also still physical. Court records, parish registers, diaries, letters and local news files sit in boxes, in handwriting, in languages and scripts that change by decade. Collecting historical data from those sources takes travel, permission and patience. Oral history pushes it further: interviewing a person about what happened, and reading what they avoid saying, is not a text task.
The pressure shows up earlier in a career than at the top of it. Literature reviews, first-pass transcription and background summaries were how new researchers learned the craft and earned hours. Those are exactly the tasks a model can take a run at, which is why fewer junior briefs get assigned. The entry-level hiring tracker follows that pattern across occupations, and it matters here because the field is small: about 3,450 historians were employed in the US, with median pay of $76,750 (BLS, 2025).
What AI does, what it assists, what it leaves alone
Some tasks already sit on the machine side. Pulling together records from digitized archives and producing a clean first draft of a report or publication summary are the clearest examples. That slice is 0% of task time on our last run, and the share of all task time AI can handle today is 39 out of 100 on the Can AI do it? measure.
A bigger group is assisted work. Organizing and indexing archival materials moves faster with machine transcription and search. Tracing how a topic developed over time is still a person’s argument, but the tools surface candidates and spot patterns a reader would miss. Assisted tasks come to 78% of task time, which is the honest middle of this job.
Then there is the work that holds. Judging whether a source is genuine and what it is worth, and advising institutions or communities on historic preservation, both turn on accountability: someone has to stand behind the call. That group is 22% of task time.
What the evidence actually supports
There is no direct, measured test of an AI system against a qualified historian doing the full job. That is why the evidence grade here is D and why we publish no parity number for this occupation. Grade D means not measured, not proven equal and not proven worse.
What would settle it is specific: a blind comparison on archival research with unseen or partly handwritten material, scored by subject specialists on citation accuracy, provenance checks and whether the argument is new rather than restated. Until something like that exists, claims that a model can do most of the job rest on demos, not measurement. Our quality parity method explains how a result like that would be graded, and the wider scoring method covers how the three questions fit together.
When the picture could shift
Most likely between 2037 and 2049 (8 in 10 of our scenarios). The replacement-year method sets out exactly what that window is and is not.
Two things could pull it earlier. Reliable machine reading of handwritten and multilingual manuscripts at scale would unlock collections that are now effectively closed to search. Wider digitization with open licensing would do the same from the other direction. Cost is not the brake: the job is almost entirely desk and archive work, so no robot is needed, and running a model is cheap next to a salary.
Two things hold it back. The paper problem is stubborn, and access to restricted or private collections depends on relationships and permissions. The second is the standard of proof. History is judged by whether a citation checks out and whether a peer reviewer agrees, and a confident invented footnote fails both. Demand is steady rather than growing, with employment projected to rise 3% between 2025 and 2035 (BLS, 2025), so most change will arrive as task erosion inside existing roles.
What to do: keep a record of the sources you located and verified yourself, because that is the part of the work a reader cannot get from a model.
How to stay needed
Lean into the tasks that sit on the human side of this job. First, authentication and significance: be the person who checks provenance and explains why a document matters. Second, undigitized and restricted collections, where getting in is half the skill. Third, oral history and public-facing advice, including preservation consulting for museums, courts, agencies and local groups.
Two skills carry the most weight. One is source criticism applied to machine output, so you can catch a fabricated citation or a flattened chronology in seconds. The other is writing for decision-makers, where a short, sourced brief beats a long summary.
If you are weighing options, close relatives of this work include History Teachers, Postsecondary, Anthropologists and Archeologists and Social Science Research Assistants. The social scientists job family shows how the group compares, and the education sector page covers where many of these roles sit. You can put two of them side by side on our job comparison tool, or see where research work lands on the list of jobs that most need a person.