■ CRITICAL RISK ■ Legal
Yes, and it largely already has. Automatic speech recognition now does in real time what shorthand was invented to approximate, and the remaining stenography jobs survive on certification requirements, not on any task a machine can't do.
“Speech-to-text is faster than your shorthand and doesn't need a translation.”
Our AI replacement risk score — how we score jobs
A stenographer's day is transcription under pressure: capturing dictation, meetings, depositions, or proceedings verbatim, usually on a steno machine whose chorded keystrokes get translated into English by dictionary software. The skill takes years to build — court-reporting programs famously wash out most students before they hit the 225-words-per-minute certification speed — and the work product is a clean, certified transcript with speakers correctly attributed and jargon spelled right.
Modern speech-to-text eats the core of that job. ASR systems transcribe faster than any human can type, handle multiple speakers with diarization, time-stamp everything automatically, and cost pennies per audio hour instead of dollars per page. Digital reporting — an audio recording plus a transcriber cleaning up the machine's output later — has already displaced live stenographers in many deposition rooms and corporate settings, driven as much by a shortage of trained stenographers as by the technology itself. The dictation side of the job, taking a letter from an executive, essentially no longer exists.
What resists is narrow but real. Courtrooms are conservative: many jurisdictions still require a certified human reporter as the official record-keeper, and a stenographer can interrupt to say 'the witness is nodding' or stop two lawyers talking over each other — an audio file just captures the mush. Heavy accents, technical terminology, and bad acoustics still trip ASR in ways that matter when a transcript decides a lawsuit. But those are shrinking edge cases plus regulatory inertia, and our risk score of 90 reflects that inertia is not a business model.
Automatability: our editorial assessment of current and near-term AI capability
This one isn't a forecast, it's a status report. Office dictation work evaporated with speech-recognition software years ago, and digital reporting is taking deposition work now. The remaining stenographer roles are concentrated in courtrooms where certification rules hold the line, and those rules are being rewritten jurisdiction by jurisdiction this decade — often because there aren't enough stenographers left to hire anyway.
As a general career, no — speech recognition has absorbed most transcription work. The exception is certified court reporting and realtime captioning, where legal requirements and accessibility standards still demand humans, and a shortage of qualified reporters keeps pay high. If you pursue it, aim squarely at those niches, because freelance transcription pricing has already collapsed against software.
Partly law, partly function. Many jurisdictions legally require a certified reporter to produce the official record. Functionally, a human can stop the proceeding when testimony is inaudible, note non-verbal responses, and swear in witnesses. An audio recording that turns out garbled can't be fixed after the fact, and appeals have been complicated by exactly that failure.
Faster in raw throughput — software transcribes audio quicker than real time and never fatigues. Top stenographers still edge it on accuracy in messy conditions: overlapping speakers, heavy accents, specialized jargon. But the gap narrows every year, and for most commercial transcription the machine's output plus a light human edit is cheaper and good enough.
Move up the value chain. Realtime captioning, CART accessibility work, and certified courtroom reporting pay better and resist automation longer than general transcription. Learn the ASR tools rather than ignoring them — agencies increasingly want people who can produce a certified transcript from hybrid workflows. And treat any purely audio-typing income stream as temporary.