Court reporter: has AI replaced it?

AI tools doing this job

Skribe.aiRecords depositions over Zoom and generates transcripts using automated speech recognition, with a human reviewing the result. A law firm article on JD Supra (June 9, 2026) describes its use in a Texas deposition; the trial court struck that transcript in April 2025 because it was not prepared by a certified shorthand reporter, and the Texas Supreme Court heard the appeal on October 6, 2026 without a ruling found as of November 13, 2026. No source gives usage figures.source ↗

Task breakdown

Task	Status	Tool	Source
Producing the official transcript of court proceedings	Not yet	Automatic speech recognition (ASR)	GovTech
Transcribing depositions	Not yet	Skribe.ai	ABA Journal
Live captioning of television programs	Augmented	Automatic speech recognition captioning	Linguistica Antverpiensia

How we rate →

Evidence

This page covers court reporters and simultaneous captioners (BLS code 27-3092), who capture spoken dialogue word for word in courts and depositions, and who caption live television and meetings for people who are deaf or hard of hearing. Before the 2018 code change, BLS's OEWS survey listed them as 23-2091 Court Reporters; its 2016 to 2018 pages already define the group to include stenocaptioners. The site's medical transcriptionist page covers speech recognition in medicine separately. In BLS's OEWS national estimates, employment was 17,130 in May 2005, 18,430 in May 2010, 17,700 in May 2016, 15,220 in May 2017, 14,490 in May 2018, 14,530 in May 2019, 12,300 in May 2021, 12,390 in May 2023 and 12,870 in May 2025. That is about 30% fewer jobs in 2025 than in 2010. Almost all of the fall came between 2016 and 2021, before the AI court transcription efforts named on this page (2025 onward), and the count has been roughly flat since. Automatic TV captions were being tested from 2018, but no source ties the fall to them. The 2018 to 2019 step crosses the code change and is flat. OEWS is a survey estimate that excludes self-employed workers, and BLS says it does not encourage the use of OEWS data for time-series analysis. OEWS also switched to a new model-based estimation method called MB3 with the May 2021 estimates, which affects comparisons across that year.

The BLS Occupational Outlook Handbook (last modified August 27, 2026) gives a larger count, 19,900 jobs in 2025, so I use it only for the projection and do not compare it with OEWS. OEWS leaves out the self-employed, and an NCRA-sponsored 2013 study put only about 9,000 of roughly 32,000 stenographers, roughly 28 percent, in official court jobs, with the rest mainly freelance. The Handbook projects 0% change (little or no change) from 2025 to 2035, with about 1,800 openings a year. Its duties list includes capturing speech with stenography machines and digital recording devices, and it says digital recording devices create an audio or video file rather than a written transcript. On court reporters, it says technology may make transcription work more efficient; however, court reporters will continue to be needed to review and edit digitally produced documentation, and that judicial budgets may affect employment. Its only AI sentence sits in the paragraph on simultaneous captioners, after the expected growth in CART demand as the population ages: transcription generated by artificial intelligence (AI) tools is expected to dampen employment demand. That is a forecast, not a measured loss. The page does not mention speech recognition.

The main non-AI force is a shortage of stenographers, which means headcount fell while demand for them did not. A 2013 Ducker Worldwide study sponsored by the National Court Reporters Association (NCRA), the stenographers' group, counted about 32,000 stenographic court reporters, found supply and demand balanced in 2013, and projected a shortfall of nearly 5,500 positions by 2018 as retirements outpaced new entrants; it expected about 70 percent of reporters to retire within 20 years. A 2025 study commissioned by the American Association of Electronic Reporters and Transcribers (AAERT), which represents digital reporters, says certified stenographers fell 21 percent over the last decade to about 23,000, that student enrollments at stenography schools dropped 74 percent, and that about 42 percent of programs or schools closed. Both sponsors have a stake in the answer, and the two counts use different methods. California shows the strain: CalMatters reported on August 11, 2026 that more than 3 million proceedings since 2023 had no written transcription, according to the Judicial Branch of California, that Los Angeles Superior Court offers a $50,000 signing bonus and had 92 vacant court reporter positions (down from 106 the summer before), and that the median pay with benefits for a California court reporter is more than $200,000. The reporters' lawyer said court reporter schools are full and have waiting lists.

Digital audio recording is the other long-running substitute, and it is plain recording, not AI. The 2013 NCRA-sponsored study said more than 45 states use some form of digital recording, at least for routine matters, and that Kentucky and Utah had fully implemented it. In August 2026 the California Supreme Court ruled that courts must allow electronic recordings of hearings when a certified professional reporter is not available (CalMatters, August 11, 2026). The recordings let parties later have a hearing transcribed by a certified professional, so this ruling is about recording, not AI.

Automatic speech recognition (ASR) transcripts are AI under this site's rules, and I found no court that accepts them as the official record in everyday practice. In North Dakota, where digital recording systems have been used for over 20 years, the Supreme Court's rulemaking committee voted unanimously in January 2025 to adopt changes to transcript preparation; the proposal behind it would mandate audio recordings starting January 1, 2026, after which transcripts could be generated with ASR (Bismarck Tribune via GovTech, January 31, 2025). The state court administrator said ASR isn't yet ready to replace official court records, that the transcript would remain the official record in most cases, and that no current staff would lose jobs. The state reporters' association president said a state-sponsored pilot had aimed for 93% accuracy. I could not confirm whether the 2026 recording mandate took effect. In Texas, a lawyer recorded a deposition with Skribe.ai, which generates transcripts using automated speech recognition, with human review (Hicks Johnson on JD Supra, June 9, 2026). The trial court struck the transcript in April 2025 because it was not prepared by a certified shorthand reporter, and the Ninth District court of appeals affirmed (ABA Journal, October 7, 2026). The Texas Supreme Court heard argument in In re Patrick Hughey on October 6, 2026, and the justices seemed inclined to allow AI-generated transcripts (ABA Journal; Bloomberg Law, October 6, 2026), but I found no ruling as of November 13, 2026. In Indiana, a judge wrote in a July 23, 2026 decision that a court transcript contained errors that looked a lot like generative AI (404 Media).

In live captioning, AI is on the air but has not been shown to replace captioners. The FCC set four caption quality standards (accuracy, synchronicity, program completeness and placement) in an order adopted February 20, 2014, and said it was not persuaded that quality standards will impede new methods such as speech recognition software. The same order let stations outside the largest markets keep using the Electronic Newsroom Technique, which turns teleprompter scripts into captions; that is plain software, not AI. In 2019, deaf and hard of hearing groups petitioned the FCC for rules on live caption quality and for a ruling on the use of automatic speech recognition for live TV (FCC public notice, August 14, 2019); I found no FCC decision on that petition. The largest study I found, by Romero-Fresco and Fresno (Linguistica Antverpiensia, December 2023), measured about 17,000 live captions from 2018 to 2022 in the UK, U.S. and Canada. It says human and automatic captions co-exist now on television. In its three North American case studies, human captions averaged between 98.9% and 99.4% accuracy, and unaided automatic captions averaged 95.7% to 96.3% in 2018 to 2021 tests, below the 98% threshold of the NER model the study uses. Some Canadian broadcasters reached the threshold consistently by the end of 2021, but the authors say automatic captions currently need help from a human operator or captioner. One U.S. case study was commissioned by the captioning company Vitac, and Enco's developer supplied the automatic captions for another. The study does not measure jobs.

Accuracy on legal speech and on different speakers is the main open question. A 2023 workshop paper tested generic ASR on two UK Supreme Court hearings: Amazon Transcribe and OpenAI Whisper averaged word error rates of 12.3 and 12.4, and they correctly transcribed court forms of address at ratios of only 0.66 and 0.69. A peer-reviewed PNAS study (Koenecke et al., 2020) of five commercial ASR systems, using interviews rather than legal speech, found an average word error rate of 0.35 for Black speakers and 0.19 for white speakers. Human reporters are not error-free either: a 2019 study in Language tested 27 Philadelphia court reporters, who are required to be certified at 95% accuracy, and found they transcribed 82.9% of words in African American English sentences accurately.

Limits: OEWS excludes the self-employed, who appear to be a large share of this occupation, and BLS advises against using OEWS as a time series. I found no study that separates the effects of the stenographer shortage, digital recording, court budgets and AI on headcount, and the 2016 to 2021 fall predates the AI tools named here. The shortage figures come from groups with a stake (NCRA and AAERT), and the California figures come from one state. I found no dated, independent evidence on AI replacing CART providers, or on how many TV stations caption live programs with ASR alone. The Bloomberg Law article was paywalled, so I read only its opening. Several state court pages, the Texas docket and the California opinion blocked my access, so the Texas and California rulings are reported from news coverage. A Texas Supreme Court ruling allowing AI deposition transcripts could change this page.

The status is not yet. In court work, which is most of this occupation, AI transcripts are being proposed, piloted and litigated but are not accepted as the official record in any court I found, and the headcount fall happened mostly by 2021 alongside a stenographer shortage and plain digital recording. Automatic captions are on air, but the best study finds they still need human help, captioners were fewer than 1,000 of about 32,000 reporters in 2013, and BLS's AI sentence is a forecast. When evidence is mixed the weaker status wins. Confidence is low: the evidence is thin, and BLS's forecast, wider ASR captioning or a Texas ruling could soon support augmented or partial.

References

Suggest a correction on GitHub →