Transcription work for AI: languages, accuracy standards and pay

Transcription for AI is language work more than typing. The audio is in your language, the standard is set by project guidelines, and the pay moves with the language. The figures below come from the open postings on micro1 as they stood on 13 October 2026.

Pay moves with the language more than with the task

All 14 open transcription roles are on micro1, they pay by the hour, and none is for English alone. Across them the advertised bands span $8–36 an hour.

The postings share one template, so nothing in the tasks explains the spread. The low bands sit on South Asian languages and the high ones on East Asian and European languages. The postings do not say why.

The seats are modest. Korean and Japanese state the most, 15 each. Every other posting that states a number states fewer.

What you produce

Of the 13 postings that carry a full description, the scope covers five things:

• A transcript of the audio. The Mandarin posting asks for verbatim text. Others ask for accuracy to meaning, tone and context. Dialects or regional speech come up in 7 postings.
• Tags on top. Metadata is asked for in 5 postings, and speaker identification or speaker changes come up in 7. One mentions timestamps.
• Flags where the audio fails. Some ask you to mark unclear or ambiguous passages and suggest an interpretation.
• Edits to someone else’s draft. The Bengali and Japanese postings say you will correct machine-generated transcripts, not only write your own.
• Feedback on the guidelines. Several ask you to report problems with the transcription process.

Error-free work is asked for in 5 postings. None states an accuracy figure or an error rate. No posting contains a percentage at all. The postings point to project guidelines for the standard, and do not print them.

Two Korean roles with the same title

micro1 has two postings called Korean Language Expert. One is transcription work and states $20–36 an hour. The other states $65–98 an hour.

The higher one asks for translation and localisation between Korean and English, linguistic annotation and help writing language guidelines. The transcription one asks for transcripts, metadata and proofreading. The bilingual speakers guide covers the higher kind. Read the scope before you compare titles.

English is part of some of these jobs

The Mandarin, Dutch, Korean and Swiss German postings include English audio as well, and the Japanese posting expects occasional English-language content.

What the postings ask you to bring

• Native or near-native fluency. Every full description uses the word "native". No posting names a test or a certificate.
• Transcription experience. It is preferred, not required. No AI experience is needed, say 13 of the 13 full descriptions.
• Tools. The postings ask for familiarity with transcription or annotation tools. None names a product.
• Discretion. Handling audio securely and keeping it confidential is asked in 4 postings, including Norwegian, German and Swiss German.

The postings do not say where the audio comes from, whether the speakers agreed to it, or whether it contains private conversations. The Dutch posting describes professionally recorded samples. For the rest you cannot tell. Ask before you start, and stop if the material feels like someone’s private call.

Country and screening

None of the postings names a country. This site lists 13 of the 14 as open to a reader in the US. None is limited to the US. The guide to working from outside the US explains how to read a country line.

micro1 screens with an AI interviewer, covered in the Zara interview guide. The postings do not say whether it runs in your language or in English.

Where to start

Korean Language Expert and Japanese Language Expert combine the highest ceiling with the most stated openings. The postings otherwise differ only in band and seats, so choose the language you can transcribe at native level. micro1’s certification and skills guide explains how it rates skills after you apply.

What the postings do not say

• Hours a week, or how long a project runs.
• Whether you are paid for hours worked or for audio completed.
• How much audio there is, how long each clip runs, and which tool you use.
• The accuracy standard, and what happens when you miss it.
• Whether the speakers consented to the recording being used.

This site is not affiliated with any platform named here, and passing a screening is not a promise of work.