The question keeps coming up. At conferences, in Facebook groups, on Reddit threads with hundreds of comments. Every time a tech company demos a new speech-to-text model, someone posts it with a caption like “RIP stenographers.”
So let’s talk about it honestly.
What AI Can Do Now
Modern speech-to-text has improved dramatically. Models like Whisper, Deepgram, and the transcription engines built into Zoom and Teams can handle clear, single-speaker audio with remarkable accuracy. Podcasts, lectures, meetings with decent microphones — AI transcribes these well enough to be genuinely useful.
And it’s getting better. Real-time captioning has gone from laughable to functional. Speaker diarisation (figuring out who said what) has improved. Background noise handling is better. Latency is lower.
If you only read the press releases, it sounds like the problem is solved.
What AI Can’t Do Yet
Courtrooms are not podcasts.
A courtroom has multiple speakers, some of whom are deliberately quiet. Witnesses mumble. Lawyers talk over each other. Judges speak softly from the bench. The acoustics are designed for authority, not clarity. The microphone setup is rarely optimised for transcription.
Then there’s the content itself.
A court reporter doesn’t just transcribe words. They produce a verbatim record — a legal document that may be used in appeals, cited in judgments, and scrutinised word-for-word years later. The standard isn’t “close enough.” It’s exact.
That means:
- Every false start matters. “I — no, I didn’t say that” is not the same as “I didn’t say that.”
- Homophones are lethal. “Eluded” vs. “alluded.” “Council” vs. “counsel.” An AI that picks the wrong one hasn’t made a typo — it’s changed the meaning of a legal record.
- Proper nouns are everywhere. Case names, statutes, place names, technical terms, names of parties who may share common surnames. An AI model doesn’t know that “Nguyen” in this case is the defendant and “Wynn” is the expert witness.
- Speaker identification must be exact. Attributing a statement to the wrong person in a legal transcript isn’t an error. It’s a potential miscarriage of justice.
AI doesn’t understand the stakes. It optimises for the most probable word. Court reporting requires the actual word.
The Shortage Nobody Talks About
Here’s the part that gets lost in the “AI will replace everyone” narrative: there aren’t enough court reporters to replace.
The profession has been facing a shortage for years. Training takes two to four years. Pass rates for certification exams are low. The average age of working court reporters is climbing. Courts in some jurisdictions are already struggling to staff proceedings.
AI transcription isn’t arriving into a profession with a surplus of workers competing for jobs. It’s arriving into a profession that can’t fill the seats it has.
Some courts have already moved to digital recording as a stopgap — not because it’s better, but because there’s no reporter available. In those courts, the question isn’t “will AI replace the reporter?” It’s “will AI make the recording actually usable?”
Where AI Actually Helps
The most interesting development isn’t AI replacing court reporters. It’s AI working alongside them.
Real-time rough drafts. Some reporters use AI-assisted tools to generate a first-pass transcript that they then correct and certify. The AI handles the bulk transcription; the reporter applies the judgment, verifies the record, and takes responsibility for accuracy.
Proofreading and review. After producing a transcript, AI can help catch the kinds of errors that slip past fatigued eyes — inconsistent spellings of a name, homophones used incorrectly, formatting anomalies. Not as a replacement for the reporter’s review, but as an additional check.
Scoping. Scopists (the people who edit and format rough transcripts from reporters) are already using AI tools to speed up their workflow. Punctuation suggestions, paragraph breaks, formatting consistency — these are tasks where AI saves time without undermining accuracy.
Training. New reporters can use AI-generated feedback to improve their speed and accuracy during training, comparing their output against AI transcriptions to identify weak spots.
The pattern is the same one playing out across professional fields: AI is most useful as a tool for skilled people, not as a replacement for them.
The Real Threat Isn’t AI
If court reporting faces an existential risk, it’s not from technology. It’s from economics.
Courts operate on tight budgets. Administrators face pressure to cut costs. If a jurisdiction can replace a salaried reporter with a recording system and a contract transcription service, some will — not because the quality is equivalent, but because the budget demands it.
That’s a political and institutional decision, not a technological one. And it’s already happening in places, independent of AI capability.
The reporters who thrive will be the ones who can articulate why their work produces a better record — and demonstrate it. The ones who can show that their transcripts are more accurate, delivered faster, and stand up to legal scrutiny in ways that automated alternatives don’t.
So Is It Cooked?
No. But it is changing.
The court reporter of 2030 will probably use AI tools routinely — for first-pass transcription, for proofreading, for formatting, for catching errors in review. The skill set will shift. Pure stenographic speed will matter less than judgment, accuracy under pressure, and the ability to produce a certified record that holds up.
The profession isn’t cooked. But the job description is evolving. The reporters who treat AI as a threat to resist will struggle. The ones who treat it as a tool to master will be the ones still working.
That’s not a comfortable answer. But it’s the honest one.
Stampede is a PDF annotation app for iPad used by court reporters and transcript proofreaders. On-device AI catches homophones, spelling errors, and inconsistencies — without your documents leaving your device. See how it works.