Is transcription still in demand now that AI exists?

Demand for typing-only medical transcription is projected to shrink, and some of the remaining work is reviewing and correcting a machine draft against the recording. The clearest official figures are for medical transcription, where the US Bureau of Labor Statistics (BLS) projects employment to fall 4 percent ¹ from 2025 to 2035. That projection is for the occupation as a whole, so it does not guarantee steady work for any one person.

For the pay side of the same question, see our guide to transcription jobs and the question on how much transcriptionists make.

Medical transcription is projected to shrink, not grow

The BLS counted about 42,000 medical transcriptionist jobs in 2025 ². It projects employment to fall 4 percent from 2025 to 2035. The openings that do come up are about 6,200 a year on average ³, and the BLS expects all of them to come from replacing workers who transfer to other occupations or leave the labor force ⁴.

Read that carefully. A field that is only replacing its workers is not adding seats. A new entrant competes for openings that appear when someone retires or changes jobs. The projection can change with the pace of AI adoption, and it is a forecast rather than a count of jobs lost.

The work is moving from typing to editing machine drafts

The BLS lists reviewing and editing drafts prepared by speech recognition software among the usual duties of medical transcriptionists ⁵. That changes the skill. You need to judge whether a machine draft matches the recording, not only to type quickly.

Drafts still need checking. Automatic transcripts make errors that a reviewer has to catch, and our question on how accurate AI transcription is covers the measured error rates. One limit is worth knowing. In a 2024 study of Whisper, about 1 percent of transcriptions contained entire phrases that were not in the audio ⁶. That study used one dataset and an older model version, so treat it as a warning rather than a rate you should expect today. A person who listens to the recording is the safeguard.

Review changes the result a great deal. In a clinical study, editing a speech recognition draft cut its error rate from 7.4 to 0.3 errors per 100 words, about 25 times fewer ⁷. That is the case for keeping a person in the loop.

Court and legal work faces the same pressure

The BLS expects transcription generated by AI tools to dampen employment demand for court reporters and simultaneous captioners ⁸. The BLS presents that as an expectation, not a measured effect. Legal transcription has its own demand picture, covered in our question on whether legal transcription is in demand.

Is it still a good side hustle?

Demand is only half the answer. A side hustle works when the pay covers your real time. General platforms pay per audio minute, and a beginner's hourly rate can be well below the posted figure. Before you commit, measure your own rate on ten files, as explained in the pay question. If the number is low, a side hustle in editing or a specialist field may suit you better than general typing.

What to do if you want to work in transcription now

1. Learn to edit against audio. Practice checking a machine draft line by line against the recording, because the BLS lists that task among the usual duties ⁵.
2. Pick a field with a clear credential path. The question on how to become a medical transcriptionist and the question on how to become a legal transcriptionist set out the routes we could verify.
3. Check the employer's terms before you test. Avoid any offer that asks you to pay to start. Our guide to transcription jobs explains the checks.

For the broader question of what AI does to the work, see the transcription overview. A transcriptionist still does the listening, judgment and checking that a machine draft cannot replace on its own, and that is where the remaining demand sits.