Why typing stopped being the whole job
The question of will AI replace word processors starts in the wrong place, because the keystrokes left first. Entering text from drafts, dictation, and recordings is the part machines handle best. Speech-to-text and page scanning took that work years before chat assistants arrived. What kept the occupation going was everything around the typing.
That remainder is real work. Formatting a long document to a house or court style. Checking a finished draft against the source it came from. Handling confidential material under a named person’s responsibility. Going back to the author when a sentence makes no sense. Templates break. Tables pick up stray styles. A filing has a deadline, and a clerk who will bounce it over margins or a missing signature block.
The market picture is blunt. BLS counts about 35,010 of these jobs in the US, with median pay of $49,280, and projects a 34.4% decline between 2025 and 2035 (BLS, 2025). That is fewer seats, not an empty room. Shrinking occupations usually stop hiring juniors before they let go of the person who knows how the document set works.
What software drafts, what it assists with, and what stays with a person
Software now carries the typing core: transcription from audio, first-pass entry from scanned or handwritten sources, and routine spelling and grammar checks. The share of task time AI can handle on its own is 22% of the job. That figure is one half of the coverage score we publish, which measures task time rather than headcount.
A second slice is shared work, where a tool drafts and a person decides: building documents from templates, cleaning up layout and numbering, and converting files between formats without wrecking the styles. Assisted time comes to 42% of the job. Speed goes up here, but someone still owns the output.
The rest sits with people: verifying a document against its source, deciding what can be released and to whom, coordinating with the author on unclear wording, and triaging which rush job goes first. That block is 36% of task time. Small, specific, and hard to hand over.
What the evidence actually tests
Evidence grade for this job: D. That is our lowest grade, and it means something plain. No published study has put AI head to head with a typical qualified word processor on the full job, so we give no quality figure against a person here. How the parity grade works explains why an ungraded result is left blank rather than estimated.
Transcription benchmarks exist, but they test clean audio and short passages. They do not test a 90-page document with tracked changes, numbered exhibits, a style guide, and a filing deadline. A test that would settle it is straightforward to describe: real document packages, timed, with error rates, rework, and rejected filings counted. Until that exists, treat confident claims about this role, from any source, as estimates.
What would move the timing, and what holds it back
Most likely between 2035 and 2049 (8 in 10 of our scenarios). The replacement-year method sets out what that window does and does not mean.
Two things could pull it earlier. Cost is one: operator figures put AI tooling for this work at roughly $90 to $8,900 a year against $15,750 to $27,880 for the human side of the same tasks. Hardware is the other, or rather the lack of it. About 36% of the work has a physical element, and that element sits in the fixed-automation tier, meaning scanners, printers, and mail equipment that offices already own. No new robot has to be invented.
Two things hold it back. Accountability comes first: a person signs off on confidential and legally filed documents, and software does not absorb liability when a version goes out wrong. Source quality is second. Poor audio, handwriting, legacy file formats, and inconsistent internal templates still produce output that a person has to repair line by line.
What to do: keep a written record of documents you caught, corrected, or rescued before filing, because that work is the part a job description never lists.
How to stay needed in document work
Lean into the three tasks that stayed with people. Final verification against the source, so you are the last check rather than the first typist. Deadline and compliance formatting, where a rejected filing costs real money. Confidential handling, where who sees what matters as much as what the file says.
Two skills raise the floor. The first is document engineering: styles, templates, cross-references, redlining, and the macros that stop a 200-page set falling apart. The second is reviewing machine output quickly and knowing where it fails, which is a proofreading habit applied to a draft nobody wrote by hand. AI skills employers ask for covers that second one across office roles.
If you are weighing a move, the nearest work is usually proofreaders and copy markers, desktop publishers, and data entry keyers. Put any two side by side on the job comparison tool before you commit to retraining. The wider office support job family and the law firms sector page show where document work is still concentrated, and the jobs most exposed list puts this role in context with its neighbors. Our full scoring approach is set out in the methodology.