The United States had 41,550 medical transcriptionists left in May 2025, down 28.2% from 57,830 a decade earlier, and the Bureau of Labor Statistics projects another 5% decline through 2034 (BLS, Occupational Employment and Wage Statistics, May 2025). The work itself did not disappear; it changed owners. Physicians still spend nearly two hours on electronic records and desk work for every hour of direct patient care (Annals of Internal Medicine), the global medical transcription market still moves roughly 80 billion dollars a year (Mordor Intelligence, 2025), and ambient AI scribes documented more than 3 million patient conversations in a single month across 600 healthcare organizations (Microsoft, March 2025). This analysis consolidates data from the Bureau of Labor Statistics, NEJM AI, the Peterson Health Technology Institute, KLAS Research, and 15 other primary sources to map where medical dictation and transcription stand in 2026.
TL;DR
- 41,550 medical transcriptionists were employed in the U.S. in May 2025, down from 57,830 in May 2015 (BLS OES).
- BLS projects employment to fall 5% from 2024 to 2034, from 43,900 to about 41,800 jobs (BLS Occupational Outlook Handbook).
- All of the roughly 7,400 projected annual openings come from replacement needs, none from growth (BLS OOH).
- Median pay was 40,410 dollars per year, or 19.43 dollars per hour, in May 2025 (BLS OES).
- Physicians spend 49.2% of their office day on EHR and desk work versus 27% on direct clinical face time (Annals of Internal Medicine).
- Primary care physicians log 5.9 hours of an 11.4-hour workday in the EHR, including 86 minutes of after-hours “pajama time” (Annals of Family Medicine).
- Raw speech recognition drafts contain 7.4 errors per 100 words; after human review, final signed notes contain 0.3% (JAMA Network Open).
- Physician burnout fell to 45.2% in 2023 from a record 62.8% in 2021 (Mayo Clinic Proceedings).
- Kaiser Permanente’s ambient AI rollout covered 7,260 physicians and 2.5 million encounters, saving the equivalent of 1,794 working days in one year (The Permanente Medical Group, NEJM Catalyst).
- Abridge raised 300 million dollars at a 5.3 billion dollar valuation in June 2025, four months after a 250 million dollar round (Fierce Healthcare).
- The global medical transcription market is worth roughly 80 to 84 billion dollars, with outsourcing at 58.95% of procurement (Mordor Intelligence; IMARC).
1. The Shrinking Medical Transcriptionist Workforce
The occupation that once defined medical documentation is contracting on every measure. The BLS Occupational Employment and Wage Statistics program counted 41,550 medical transcriptionists in May 2025, a loss of 16,280 jobs, or 28.2%, since May 2015. The BLS Occupational Outlook Handbook projects a further 5% decline from 2024 to 2034, and every one of the roughly 7,400 projected annual openings is expected to come from replacing workers who leave, not from growth. The handbook is explicit about why: a core duty of the modern transcriptionist is to “review and edit drafts prepared by speech recognition software”, meaning the technology that displaced the keystroke work now defines what remains of it.
| Metric | Value | Source |
|---|---|---|
| U.S. medical transcriptionists employed (May 2025) | 41,550 | BLS OES, May 2025 |
| Employed a decade earlier (May 2015) | 57,830 | BLS OES, May 2015 |
| Ten-year employment change | -16,280 (-28.2%) | BLS OES |
| Employment in projection base year (2024) | 43,900 | BLS OOH |
| Projected employment (2034) | ~41,800 (-5%) | BLS OOH, 2024-34 |
| Projected numeric change, 2024-34 | -2,200 jobs | BLS OOH |
| Projected annual openings (all replacement) | ~7,400 | BLS OOH |
| Official outlook category | ”Decline” | O*NET Online, 31-9094.00 |
Context: court stenographers face a similar squeeze from speech technology, quantified in our court reporting statistics.
2. Wages: What Human Transcription Pays in 2026
Pay tells the same story from a different angle. The May 2025 median of 40,410 dollars sits just 15.8% above the May 2015 median of 34,890 dollars in nominal terms, a decade of wage growth that trails cumulative inflation over the same period. The pay band is unusually flat: the bottom 10% of medical transcriptionists earned less than 28,250 dollars while the top 10% earned more than 55,950 dollars, a ceiling only about 1.4 times the median. Entry still requires only a postsecondary nondegree award, per the BLS Occupational Outlook Handbook, which keeps the field accessible but also keeps wages compressed as volume shifts to software.
| Metric | Value | Source |
|---|---|---|
| Median annual wage (May 2025) | 40,410 dollars | BLS OES, May 2025 |
| Mean annual wage (May 2025) | 41,710 dollars | BLS OES, May 2025 |
| Median hourly wage (May 2025) | 19.43 dollars | BLS OES, May 2025 |
| 10th percentile annual wage | 28,250 dollars | BLS OES, May 2025 |
| 90th percentile annual wage | 55,950 dollars | BLS OES, May 2025 |
| Median annual wage (May 2015) | 34,890 dollars | BLS OES, May 2015 |
| Nominal median wage growth, 2015-2025 | +15.8% | BLS OES |
| Typical entry-level education | Postsecondary nondegree award | BLS OOH |
Note: the OOH’s separately published 2024 median (37,550 dollars, or 18.05 dollars per hour) reflects the May 2024 survey; the May 2025 figures above are the most recent available.
3. The Documentation Burden Behind the Demand
Transcription demand has always been a symptom of a larger problem: clinical medicine generates more paperwork than clinicians can absorb. The landmark Annals of Internal Medicine time-motion study that followed 57 physicians for 430 hours found they spent 49.2% of the office day on EHR and desk work and only 27% on direct clinical face time. For every hour with patients, physicians spent nearly two additional hours on records and administrative work, and diary data showed another 1 to 2 hours of EHR tasks each night. Event-log research puts hard numbers on the same pattern: 142 family physicians spent 5.9 hours of an 11.4-hour workday in the EHR (Annals of Family Medicine), and a study of roughly 100 million outpatient encounters measured 16 minutes and 14 seconds of EHR time per visit (Annals of Internal Medicine, 2020).
| Metric | Value | Source |
|---|---|---|
| Office day spent on EHR and desk work | 49.2% | Annals of Internal Medicine, 2016 |
| Office day spent on direct clinical face time | 27.0% | Annals of Internal Medicine, 2016 |
| Extra EHR/desk hours per hour of patient time | ~2 hours | Annals of Internal Medicine, 2016 |
| Primary care workday spent in the EHR | 5.9 of 11.4 hours (355 min) | Annals of Family Medicine, 2017 |
| After-hours “pajama time” in the EHR per day | 86 minutes | Annals of Family Medicine, 2017 |
| Clerical share of total EHR time | 44.2% (157 min) | Annals of Family Medicine, 2017 |
| EHR time per outpatient encounter | 16 min 14 s | Annals of Internal Medicine, 2020 |
| EHR activity split per encounter | 33% chart review, 24% documentation, 17% ordering | Annals of Internal Medicine, 2020 |
Context: virtual visits generate the same notes as in-person ones, so this burden follows care wherever it moves; see our telehealth statistics.
4. Burnout and the Cost of Clerical Work
Documentation load is not just an efficiency problem; it tracks directly with clinician well-being. The triennial Mayo Clinic Proceedings survey of 7,643 participants found 45.2% of U.S. physicians reported at least one symptom of burnout in 2023, an improvement from the pandemic record of 62.8% in 2021 but still elevated against the general working population. Even inside the exam room, physicians spent 37% of their time on the EHR while the patient was present, leaving 52.9% for direct interaction (Annals of Internal Medicine). The system-level price is enormous: the Peterson Health Technology Institute’s April 2026 administrative AI report attributes 350 billion dollars of annual U.S. healthcare spending to administrative waste, 266 billion dollars of it to administrative complexity alone.
| Metric | Value | Source |
|---|---|---|
| Physicians with at least one burnout symptom (2023) | 45.2% | Mayo Clinic Proceedings, 2025 |
| Burnout at the 2021 peak | 62.8% | Mayo Clinic Proceedings, 2025 |
| Burnout in 2011, for comparison | 45.5% | Mayo Clinic Proceedings, 2025 |
| Satisfied with work-life integration, 2023 vs 2021 | 42.2% vs 30.3% | Mayo Clinic Proceedings, 2025 |
| Exam-room time on EHR with patient present | 37.0% | Annals of Internal Medicine, 2016 |
| After-hours work reported per night | 1-2 hours | Annals of Internal Medicine, 2016 |
| Annual U.S. healthcare administrative waste | 350 billion dollars | PHTI, 2026 |
| Waste attributed to administrative complexity | 266 billion dollars | PHTI, 2026 |
Note: burnout has many drivers, but documentation is the one health systems can most directly engineer away, which is exactly the pitch behind the ambient AI wave in section 6.
5. Speech Recognition Accuracy in Clinical Documents
The classic case for keeping humans in the loop is an error-rate cascade documented in JAMA Network Open. Researchers analyzed 217 clinical notes, 83 office notes, 75 discharge summaries, and 59 operative notes, dictated by 144 physicians using speech recognition software. Raw speech recognition drafts contained 7.4 errors per 100 words; professional transcriptionist review cut the rate to 0.4%, and final physician-signed notes contained 0.3%. Errors were not all cosmetic: 5.7% of the raw draft errors were clinically significant. That 25-fold reduction between machine draft and signed note is the entire remaining business case for human medical transcription, and it explains why the job survives as an editing role rather than a typing one.
| Metric | Value | Source |
|---|---|---|
| Error rate in raw speech recognition drafts | 7.4 per 100 words | JAMA Network Open, 2018 |
| Error rate after transcriptionist review | 0.4% | JAMA Network Open, 2018 |
| Error rate in final signed notes | 0.3% | JAMA Network Open, 2018 |
| Raw draft errors that were clinically significant | 5.7% | JAMA Network Open, 2018 |
| Sample analyzed | 217 notes from 144 physicians | JAMA Network Open, 2018 |
Note: newer ambient AI systems are evaluated differently; in a 2025 NEJM AI trial, ambient-generated documentation scored 3.97 to 4.99 out of 5 across quality domains on the PDSQI-9 instrument, with clinician sign-off still required on every note.
6. Ambient AI Clinical Documentation Goes Mainstream
The fastest-moving story in medical documentation is ambient AI: software that listens to the visit and drafts the note without dictation. The largest disclosed deployment is at Kaiser Permanente’s Northern California medical group, where an analysis published via NEJM Catalyst covered 7,260 physicians and more than 2.5 million patient encounters. The tool saved those physicians the equivalent of 1,794 working days in one year, 84% reported positive effects on patient interactions, and 82% reported improved work satisfaction. Microsoft’s Dragon Copilot announcement reported DAX Copilot documenting over 3 million ambient conversations in a single month across 600 organizations, while randomized evidence arrived in December 2025: an NEJM AI trial of 238 physicians found a 9.5% reduction in time-in-note for one scribe and improved burnout scores for both tested products, and a second pragmatic NEJM AI trial measured a 0.36 hour per day drop in note time and a significant reduction in work exhaustion. The Peterson Health Technology Institute, surveying leading systems including Mass General Brigham, Providence, and Intermountain, called ambient scribe uptake “on track to be one of the fastest in recent healthcare history”, and KLAS Research published its first vendor-performance comparison of the category in February 2025.
| Metric | Value | Source |
|---|---|---|
| Kaiser Permanente physicians using ambient AI | 7,260 | The Permanente Medical Group / NEJM Catalyst |
| Encounters documented in the evaluation | 2.5+ million | The Permanente Medical Group / NEJM Catalyst |
| Physician time saved in one year | 1,794 working days | The Permanente Medical Group / NEJM Catalyst |
| Patients noticing less physician screen time | 47% | The Permanente Medical Group / NEJM Catalyst |
| DAX Copilot ambient conversations in one month | 3+ million across 600 orgs | Microsoft, Mar 2025 |
| Time saved per encounter (DAX Copilot) | 5 minutes | Microsoft, Mar 2025 |
| Clinicians reporting reduced burnout (DAX survey, n=879) | 70% | Microsoft, Mar 2025 |
| Patients reporting a better overall experience | 93% | Microsoft, Mar 2025 |
| Notes generated via ambient AI in pragmatic RCT | 38% (27,092 of 71,487) | NEJM AI, 2025 |
Outlier: adoption is concentrated among heavy users; at Kaiser Permanente, the top third of users accounted for 89% of all ambient AI activations, with no correlation between physician age and adoption. For the broader clinical AI picture, see our AI in healthcare statistics.
7. Market Size, Outsourcing, and the Money Flowing In
The transcription services market remains large even as the occupation shrinks, because volume moved to outsourcers and software rather than vanishing. Mordor Intelligence values the global medical transcription services market at 80.41 billion dollars in 2025, and IMARC lands nearby at 83.5 billion dollars; both project about 5.1% annual growth. Outsourcing accounts for 58.95% of the market by procurement mode, and North America holds 41% to 46% of global revenue depending on the firm. Investor money, meanwhile, is chasing the AI layer: Abridge raised a 300 million dollar Series E at a 5.3 billion dollar valuation in June 2025, only four months after a 250 million dollar Series D at 2.75 billion dollars, bringing its total raised to roughly 800 million dollars. The company reports deployments in more than 150 health systems across 55 specialties and 28 languages, on pace to support over 50 million medical conversations in 2025, and its 2026 announcements added WVU Medicine’s rural network and evidence partnerships with NEJM and JAMA.
| Metric | Value | Source |
|---|---|---|
| Global medical transcription market (2025) | 80.41 billion dollars | Mordor Intelligence |
| Same market, second estimate (2025) | 83.5 billion dollars | IMARC |
| Forecast values | 108.5 billion dollars by 2031; 133.2 billion dollars by 2034 | Mordor Intelligence; IMARC |
| Projected CAGR | 5.12% | Mordor Intelligence; IMARC |
| Outsourcing share of procurement (2025) | 58.95% | Mordor Intelligence |
| North America share of global market | 41.05% to 45.8% | Mordor Intelligence; IMARC |
| Abridge total raised / valuation (mid-2025) | ~800 million dollars / 5.3 billion dollars | Fierce Healthcare, Jun 2025 |
| Abridge health system deployments (mid-2025) | 150+, up 50% in four months | Fierce Healthcare, Jun 2025 |
Context: the same ambient-capture pattern is spreading beyond medicine into everyday work conversations, tracked in our AI meeting assistant statistics.
Summary: Medical Transcription by the Numbers
| Metric | Value | Source |
|---|---|---|
| U.S. medical transcriptionists (May 2025) | 41,550 | BLS OES, May 2025 |
| Ten-year workforce change (2015-2025) | -28.2% | BLS OES |
| Projected employment change, 2024-34 | -5% (-2,200 jobs) | BLS OOH |
| Projected annual openings (all replacement) | ~7,400 | BLS OOH |
| Median annual wage (May 2025) | 40,410 dollars | BLS OES, May 2025 |
| 90th percentile annual wage | 55,950 dollars | BLS OES, May 2025 |
| Office day on EHR and desk work | 49.2% | Annals of Internal Medicine, 2016 |
| Primary care workday in the EHR | 5.9 of 11.4 hours | Annals of Family Medicine, 2017 |
| After-hours EHR time per day | 86 minutes | Annals of Family Medicine, 2017 |
| EHR time per outpatient encounter | 16 min 14 s | Annals of Internal Medicine, 2020 |
| Physician burnout, 2023 vs 2021 | 45.2% vs 62.8% | Mayo Clinic Proceedings, 2025 |
| U.S. healthcare administrative waste per year | 350 billion dollars | PHTI, 2026 |
| Raw speech recognition error rate | 7.4 per 100 words | JAMA Network Open, 2018 |
| Final signed-note error rate after review | 0.3% | JAMA Network Open, 2018 |
| Kaiser physicians using ambient AI / encounters | 7,260 / 2.5+ million | The Permanente Medical Group / NEJM Catalyst |
| DAX Copilot monthly ambient conversations | 3+ million across 600 orgs | Microsoft, Mar 2025 |
| Clinicians reporting reduced burnout with DAX | 70% | Microsoft, Mar 2025 |
| Global medical transcription market (2025) | 80.41-83.5 billion dollars | Mordor Intelligence; IMARC |
| Outsourcing share of market | 58.95% | Mordor Intelligence |
| Abridge valuation (Jun 2025) | 5.3 billion dollars | Fierce Healthcare |
Methodology and Sources
Data was gathered by aggregating figures directly from primary government datasets, peer-reviewed studies, vendor disclosures, and named market research releases published mostly in 2024-2026, cross-referencing overlapping metrics across two or more sources and flagging older baseline studies (2016-2020) that remain the most recent rigorous measurements of documentation time.
- BLS, Occupational Employment and Wage Statistics, Medical Transcriptionists (31-9094), May 2025 (retrieved via the BLS public API, Jul 2026)
- BLS, Occupational Outlook Handbook, Medical Transcriptionists (2024-34 projections)
- BLS, Occupational Employment Statistics, May 2015 (historical employment and wages, archived edition)
- O*NET Online, Medical Transcriptionists 31-9094.00
- Sinsky et al., Allocation of Physician Time in Ambulatory Practice, Annals of Internal Medicine (2016)
- Arndt et al., Tethered to the EHR, Annals of Family Medicine (2017)
- Overhage and McCallie, Physician Time Spent Using the EHR During Outpatient Encounters, Annals of Internal Medicine (2020)
- Zhou et al., Analysis of Errors in Dictated Clinical Documents, JAMA Network Open (2018)
- Shanafelt et al., Changes in Burnout 2011-2023, Mayo Clinic Proceedings (2025)
- Ambient AI Scribes in Clinical Practice: A Randomized Trial, NEJM AI (Dec 2025)
- A Pragmatic Randomized Controlled Trial of Ambient AI, NEJM AI (Dec 2025)
- The Permanente Medical Group, NEJM Catalyst ambient AI analysis summary
- Peterson Health Technology Institute, Adoption of AI in Healthcare Delivery Systems (Mar 2025)
- Peterson Health Technology Institute, Administrative AI: Current Use and Potential Impact (Apr 2026)
- KLAS Research, Ambient Speech 2025 (Feb 2025)
- Microsoft, Dragon Copilot announcement with DAX Copilot usage data (Mar 3, 2025)
- Fierce Healthcare, Abridge 250M dollar Series D (Feb 2025)
- Fierce Healthcare, Abridge 300M dollar Series E (Jun 2025)
- Mordor Intelligence, Medical Transcription Services Market (2026-31)
- IMARC Group, Medical Transcription Market (2026-34)
- Abridge, press releases (2025-2026)
Data watch: BLS publishes new OES employment and wage data each spring and refreshes its 10-year employment projections annually, so a 2025-35 round is expected in late 2026; the Mayo Clinic Proceedings burnout survey runs on a roughly triennial cycle, with the next wave covering 2026; PHTI and KLAS both update their ambient documentation research on rolling cycles; and market research firms revise transcription market sizing annually.
Last updated: July 20, 2026. We review and update this page quarterly as new data is published.