Why paper records keep a person in the room
File clerk work splits into two halves. One half is information handling: reading an incoming document, deciding where it belongs, adding it to a record, and logging what was filed or taken out. Software has been chipping away at that half for years. The other half is physical: placing materials in cabinets, boxes and bins, pulling a folder when a colleague asks for it, and boxing up outdated material for destruction. That work moves at the speed of hands, shelves and hallways.
Asking whether AI will replace file clerks is really asking how fast those two halves separate. Capture software can scan a stack, read it, tag it and drop it into the right folder in a document system. It cannot find the one misfiled deed in a row of banker’s boxes, or notice that a contract was returned without its signature page. When an office is half paper and half digital, someone has to hold the two systems together.
The job is also shrinking for ordinary business reasons, not just AI. The US employs about 73,440 file clerks, with median pay of $43,600 a year, and the Bureau of Labor Statistics projects employment falling 15.8% between 2025 and 2035 (BLS, 2025). Most of that comes from offices going paperless and from fewer entry-level filing roles being backfilled when someone leaves. You can see the same pattern across clerical work in our list of jobs expected to shrink.
What software handles, what it assists, and what stays with people
Start with the automatable end. Of the task time AI touches in this job, 12% is work it can run with light supervision: classifying incoming materials by content, and creating or updating the index entry when a new record is opened. Optical character recognition plus a trained classifier does both faster than a person, once documents are already in digital form.
The assisted share is 22%. That is retrieval and housekeeping. A search tool can surface likely matches for a records request, and a retention rule can flag material that has passed its keep-by date — but a person confirms the match, checks the file is complete, and signs off before anything is purged. On our coverage scale, which asks only can AI do it, this job sits at 30, where higher means more task time AI can handle today.
That leaves 66% of total task time with people. It is the hands-on and judgment work: shelving and boxing physical materials, tracking which files have been removed and by whom, chasing down records that were never returned, and dealing with the request that does not match any index term. Our headline figure, the Still needs a human score, comes out at 71 out of 100 (higher is safer); how that score is built explains the weighting.
What the evidence does and does not show
There is no direct head-to-head test of AI against working file clerks. Our evidence grade for quality parity is D, which means the question is not measured, so we publish no parity number for this job. Claiming software beats a clerk here would be a guess dressed up as data.
What exists is adjacent. Document capture accuracy is a well-studied engineering problem, and office digitization has been tracked for decades. Neither tells you how often a records request gets answered correctly in a mixed paper-and-digital archive, which is the measure that matters. A clean test would score people and software on the same batch of real requests: correct file found, complete, logged, and returned on time. Until something like that is published, treat confident predictions about this job with care. Our scoring method shows exactly where the gaps sit.
When the picture could change
Most likely between 2036 and 2059 (8 in 10 of our scenarios). The replacement-year method explains how that window is built and why it is a range rather than a date.
Two things could pull it earlier. First, cheap document capture: running software on a page costs far less than paying for the same hour of human handling, so once an archive is scanned, the ongoing filing work largely goes away. Second, retirement and attrition — when a clerk leaves a small records team, the tasks are often split between remaining staff and a document system rather than refilled.
Two things hold it back. The physical side is stubborn: most of this job’s task time involves moving and storing real objects, and the automation that handles that is fixed equipment built for one layout, not a general-purpose robot you can drop into a basement file room. And legal retention rules in law firms, insurance and government mean original paper has to be kept, located and produced on demand, with someone accountable for the chain of custody.
What to do: if your office is mid-digitization, ask to own the scanning and indexing project rather than only feeding it.
How to stay needed in records work
Lean into the tasks that stay human. Take charge of tracking materials removed from files, so the audit trail is yours. Handle the awkward retrieval requests nobody can phrase properly, because that skill is judgment, not search. And own the purge: deciding what can go, under which retention rule, with a record of the decision.
Two skills raise your floor. Learn the document management system you already use well enough to build taxonomies, retention rules and permissions, not just file into it. Then add records compliance — retention schedules, privacy rules, and how to respond to a subpoena or public-records request. That combination is the usual path from clerk to records coordinator.
Nearby jobs worth a look, and how they score, are Correspondence Clerks, Library Assistants, Clerical and Office Clerks, General. The wider information and record clerks family shows where the steadier roles sit, and the administrative support sector page puts this job next to its neighbors. To weigh one move against another, put two of them side by side in our job comparison tool.