Not fully, and the reason is legal rather than technical. Speech recognition is now accurate enough to produce a usable rough transcript of most courtroom audio. What it cannot do is certify that transcript as the official record, swear an oath, or take responsibility when a word that changes the meaning of testimony is wrong. The Bureau of Labor Statistics projects employment of court reporters to show little or no change from 2025 to 2035, holding at about 19,900 positions with median pay of $72,420.
Flat is an unusual projection in a field this exposed to automation, and understanding why it is flat rather than falling tells you most of what matters about this career.
Key points
- The official record needs a certified human. Courts require a sworn reporter to attest that the transcript is accurate and complete, and no jurisdiction has transferred that to software.
- Employment is projected flat, at roughly 19,900 jobs and 0 percent change to 2035, with median pay of $72,420.
- The profession is small and ageing, so openings come from retirements rather than growth, and several jurisdictions report shortages rather than surpluses.
- AI has already changed the workflow, with recognition producing a first pass that the reporter corrects and certifies.
- The work is splitting in two. Certified courtroom and deposition reporting holds up; uncertified general transcription is being competed away.
What court reporters do and where AI already sits
Describing the job as typing what people say understates it in ways that matter for the automation question.
A reporter produces a verbatim record of proceedings in real time, at conversational speed, with multiple speakers who interrupt each other, use specialist vocabulary, mumble, and speak in accents the system has never encountered. They mark exhibits, note who is speaking, capture non-verbal events that affect the record, and interrupt the proceeding when something genuinely cannot be heard. That last power is important and often overlooked: a reporter can stop a court to ask a witness to repeat themselves, which is the mechanism that keeps the record complete.
Beyond the live capture, they certify the transcript. A certified transcript is a legal instrument. It supports appeals, it is relied on by judges who were not present, and an error in it can change an outcome. The certification carries the reporter’s professional liability.
Three kinds of work with different exposure
Real-time courtroom reporting. A judge or counsel reads the text as it appears, and deaf or hard-of-hearing participants rely on it. This demands both speed and immediate accuracy, and it is the least exposed part of the profession.
Deposition and hearing reporting. Often the largest share of private practice work. Recording is more controlled, and this is where digital reporting models have made the most progress, usually with a human still present to monitor and later certify.
Post-event transcription. Turning existing audio into text. This is the most exposed category, because a recognition system produces a draft in minutes and the human role shrinks to correction.
The technology already in the workflow
The technology is not arriving; it arrived. Most reporters now work alongside it in some form.
Automatic speech recognition generates a rough transcript from courtroom or deposition audio, often within minutes of the session ending. Speaker diarisation attempts to label who said what, which is genuinely useful and genuinely unreliable when people talk over each other. Legal vocabulary models handle case citations, Latin terms and drug names far better than general recognition. Search and indexing across transcripts lets counsel find a passage across thousands of pages in seconds, which is a capability the profession simply did not have before.
The honest assessment is that recognition handles clean single-speaker audio very well and degrades sharply in exactly the conditions courtrooms produce: crosstalk, poor microphone placement, strong accents, emotional testimony, and specialist terminology used without explanation. The reporter’s value has shifted from producing every word to catching the words the system got wrong, which is a different skill and in some ways a harder one.
There is also a category of error that matters more here than in most transcription work. A recognition system that mishears a word produces something plausible rather than something obviously broken. “Cannot” and “can” differ by three letters and reverse the meaning of testimony. A human reading for sense catches that; a system optimising for acoustic likelihood may not.
This is why raw accuracy percentages mislead in legal settings. A transcript that is 98 percent accurate sounds excellent until you consider that a long deposition runs to tens of thousands of words, which puts hundreds of errors in the document, distributed unpredictably. In most transcription work those errors are cosmetic. In a legal record, a handful of them sit on words that decide something, and there is no way to know in advance which handful.
Why the record still requires a person
Four things keep certification with a human, and none of them is close to changing.
The oath and the liability. A reporter swears that the record is true and complete. That statement has legal weight, and it needs a person who can be examined about it. Software cannot take an oath, and no vendor has offered to accept the liability that would come with certifying court records.
The right to challenge the record. Parties can dispute a transcript, and resolving that dispute involves questioning the person who produced it about what they heard and why they transcribed it that way. Removing the human removes the mechanism.
Accessibility obligations. Real-time text for deaf and hard-of-hearing participants is a legal requirement in many proceedings, and courts have generally required the accuracy standard that certified reporting provides rather than the standard automatic captioning delivers.
Courts move slowly and cautiously. Rules of procedure change through committees and pilot programmes over years. Even where digital reporting has been approved, it has typically been introduced with a human monitor rather than as an unattended system.
Several jurisdictions have expanded digital reporting, usually because they could not recruit enough stenographers rather than because they preferred the technology. That distinction matters. The shortage is driving the change, not the capability.
It also sets a ceiling on how far the change goes. A court that adopts digital reporting because no stenographer was available has solved a staffing problem, not replaced a function. The monitor still attends, the certification still happens, and the cost saving is smaller than the technology alone would suggest. Jurisdictions that expected to remove the human entirely have generally found that the review time needed to make a machine draft citable eats most of what they hoped to save.
What to know before deciding
| Measure | Court reporters, 2025 |
|---|---|
| Median annual pay | $72,420 |
| Number of jobs | 19,900 |
| Projected change, 2025 to 2035 | 0 percent (Little or no change) |
| Projected employment change | 0 |
| Typical entry-level education | Postsecondary nondegree award |
Two things about that table are worth drawing out.
A flat projection in a highly exposed field is a stronger signal than it looks. Comparable clerical and transcription occupations are projected to decline: medical transcriptionists by 4 percent and editors by 1 percent. Court reporting holding steady while adjacent transcription work falls is the certification requirement showing up in the numbers.
The occupation is very small. Twenty thousand positions nationally means limited geographic flexibility and a job market where reputation and local relationships matter more than in larger fields. It also means the training pipeline is thin, which is why shortages persist.
For context on how exposure is being measured, the Bureau publishes AI exposure categories for 831 occupations and states that exposure “does not imply job loss, productivity gains, automation probability, or wage effects.” High exposure means a model can do overlapping tasks. It says nothing about whether the law permits it to.
What is actually changing for reporters
- Output shifts from production to verification. Less time capturing every word, more time correcting a draft and certifying it.
- Speed requirements stay, accuracy requirements rise. The errors that survive a recognition pass are the subtle ones, which are harder to catch than obvious gaps.
- Real-time work becomes the premium segment. It is the hardest to automate and commands the best rates.
- Uncertified transcription work disappears. If the job does not require certification, price competition with software wins.
- Technical fluency becomes part of the craft. Knowing where a recognition system fails predictably is now a professional skill.
Decision framework
Five questions before entering or leaving this field.
- Are you willing to certify? The whole defensibility of this career rests on holding a certification the law recognises. Uncertified transcription is not a viable long-term position.
- Can you work at real-time speed? The premium segment requires it, and it is the part of the profession with the clearest future.
- Is there demand in your jurisdiction? This is a local market. Some jurisdictions have shortages and good rates; others have moved substantially to digital reporting with monitors.
- Do you want employment or self-employment? A large share of the work is freelance deposition reporting, which changes the risk profile considerably.
- Will you learn the systems you work with? The reporters doing best are the ones who understand where recognition fails and can correct efficiently rather than retyping from scratch.
That last point generalises well past this profession. Knowing how these systems reason, and where their confident output is wrong, is a transferable skill. If you want a structured route in, explore Coursiv AI lessons and check current plan details on the official site.
Your next step
If you are considering this field, check your own jurisdiction before anything else. Court reporting is regulated locally, certification requirements differ, and the balance between stenographic and digital reporting varies enormously between states and countries. National projections tell you the profession is stable; only local research tells you whether it is stable where you live.
If you are already working, the highest-return move is getting fluent at editing recognition output rather than resisting it. The rate you can charge depends on turnaround, and reporters who correct a draft efficiently finish faster than those producing every word from scratch.