Why the record still runs through a person
Court reporters and simultaneous captioners do two jobs at once. They capture spoken words as they happen, and then they stand behind the result. In a deposition or a hearing, the reporter keeps the record usable: stopping counsel when two people talk over each other, asking a witness to repeat a name, reading testimony back when the judge asks for it. Software does not get to interrupt a courtroom.
The second half is certification. A transcript becomes an official record because a qualified person attests that it is true and complete. In many depositions, that same person swears in the witness and marks the exhibits. Accountability here is a legal arrangement, not a technical one, and it does not transfer to a model.
Live captioning has its own version of the problem. Captioners handle accents, crosstalk, proper nouns and specialist vocabulary in real time, with no second pass and no chance to clean the file later. Ask whether AI will replace court reporters in that setting and the honest story is task erosion: drafting gets cheaper and faster, while the parts that carry legal weight stay with people.
What the software drafts, what it assists, and what stays human
Machine transcription is strongest at first-pass drafting. It turns recorded audio into searchable text, labels speakers, and indexes a file so a paralegal can find a passage in seconds. In our task split, the share of task time sitting in the group AI can handle on its own is 22%. Coverage, our measure of how much of the work AI can touch today, reads 38 out of 100; how coverage is measured sets out what that counts.
Assisted work is the bigger day-to-day change. Speech tools feed a rough real-time draft that the reporter corrects, dictionaries expand legal terms and party names on the fly, and automated passes flag likely misheard words before a transcript goes out. The share of task time in that assisted group is 45%.
Then there is the work that needs the person in the room: certifying the transcript, controlling the record when speech overlaps or a witness trails off, and reading testimony back on request. That group comes to 33%.
What has actually been tested
Not much, directly. Our evidence grade for this job is D, which means there is no published head-to-head test of AI against certified reporters under real courtroom conditions in our evidence set. So we publish no parity number for this occupation. Word-error-rate scores on clean, single-speaker audio do not settle it, because the hard minutes are the messy ones.
What would settle it is a blind comparison on genuine multi-speaker proceedings: objections, accented testimony, technical exhibits, and names nobody spells out. The output would need scoring against the accuracy standard courts require for a certified transcript, with disputed passages reviewed by qualified reporters. Until a test like that exists, the parity column stays open. Our scoring method explains how the grades are assigned and why a missing grade never becomes a number.
When this could shift
Most likely between 2036 and 2049 (8 in 10 of our scenarios). The replacement-year method explains what that window measures and how wide it is meant to be.
Two things could pull the change earlier. Deployment needs no hardware: the work is audio in, text out, so there is nothing to build beyond software and better microphones. And where a court simply cannot staff a reporter, the fallback is already recording plus later transcription, which normalizes the machine draft as the starting point.
Two things hold it back. Rules of court and state statutes often specify a certified reporter for an official record, and those rules change slowly, jurisdiction by jurisdiction. Liability is the other brake, because a disputed line in a transcript can move a case and someone has to answer for it. Federal projections also point to a steady, small occupation rather than a shrinking one: BLS 2025-35 projections put the ten-year employment change for this job at about -0.1%. If you want context on jobs moving the other way, see our list of jobs most at risk.
How to stay needed in this work
Lean into the tasks that only work with a qualified person attached. Certifying the record is the first: the signature, the oath, the chain of responsibility. Second, controlling and reading back the live record, which means interrupting, clarifying and fixing a misheard name before it hardens into the transcript. Third, real-time captioning of unpredictable speech for deaf and hard-of-hearing audiences, where there is no second pass and no editing window.
Two skills carry the most weight. One is realtime output with a well-tuned personal dictionary, so your draft is usable the moment the hearing ends. The other is fast, accurate correction of machine drafts, which is a different skill from typing: it is hearing the gap between what the file says and what was said, and knowing which errors matter legally.
What to do: keep your certification current and get quick at editing machine output, because that is the part of the workflow firms and courts are paying for.
Nearby work worth a look: Interpreters and Translators, Proofreaders and Copy Markers, and Court, Municipal, and License Clerks. You can put any two of them side by side on our job comparison tool, see the wider media and communication workers family, or read how exposure plays out across law firms.