JustTranscribe

Report · September 2026

What Latin America Transcribes

Six weeks of real JustTranscribe usage, in numbers: 1,909 recordings and videos turned into text by people in Argentina, Peru, Colombia, Brazil, Mexico and Chile. What they upload, how long it runs, in which language and at what hour.

UGBy Unai GoikoetxeaData from 4 August to 15 September 2026 · published 15 September 2026

1,909
transcripts
35,416
minutes of audio (590 hours)
60
countries
42
languages detected

More than half are files, not links

57% of transcripts start from a file someone had on their phone or computer. YouTube links are 36% of transcripts but half of all minutes: they are long lectures and talks. TikTok and Instagram, for all the noise, are 4%.

% of transcripts% of minutes
  • Uploaded file57 % · 1,08846.3 %
  • YouTube link36.4 % · 69446.4 %
  • Google Drive2.7 % · 526.3 %
  • TikTok2.7 % · 520.9 %
  • Instagram1.1 % · 210.1 %

A third of uploads are WhatsApp voice notes

Of every ten uploaded files, three are WhatsApp voice notes exactly as they leave the phone (.opus), three are video files (MP4/MOV) and the rest are audio recordings: memos, interviews, classes recorded on a phone. Nobody converts formats before uploading; they upload what they have.

  • WhatsApp voice note (.opus)30.2 % · 329
  • Video (MP4, MOV)28.7 % · 312
  • Phone recording (.m4a)14.2 % · 154
  • MP324.9 % · 271
  • WAV2 % · 22

Half run under 8 minutes; the longest 22% take two thirds of the time

The typical recording is 8.4 minutes. But files over 30 minutes, one in five, hold 66% of all minutes transcribed: university lectures, meetings and hearings. 137 recordings ran past an hour; the longest, 148 minutes.

% of transcripts% of minutes
  • Under 2 min21.9 % · 4191.2 %
  • 2 to 10 min32.2 % · 6158.7 %
  • 10 to 30 min24.1 % · 46123.8 %
  • 30 to 60 min14.5 % · 27732.3 %
  • 60 to 90 min4.1 % · 7916.2 %
  • Over 90 min3 % · 5817.9 %

Median length by source: Google Drive 39.5 min · YouTube 13.7 · uploaded file 4.7 · TikTok 1.6 · Instagram 1.0.

Spanish in two of every three; 42 languages in total

Language is detected automatically, never chosen by the user. Two in three recordings are Spanish, one in seven English, one in thirteen Portuguese. Then Hebrew, Arabic, Hindi, Russian, Chinese, Japanese and Catalan: 42 distinct languages in six weeks.

  • Spanish67.1 % · 1,280
  • English14.6 % · 278
  • Portuguese7.5 % · 144
  • Hebrew1.6 % · 31
  • Arabic1.4 % · 27
  • Hindi0.7 % · 14
  • Russian0.6 % · 12
  • Chinese0.6 % · 12
  • Japanese0.5 % · 9
  • Catalan0.3 % · 6

The model labelled 33 recordings as Galician; on review most were Argentine or Chilean Spanish with a strong accent, a known error we fixed on 12 September. That is why Galician is not in the list.

Argentina first, and Lima is the city that transcribes most

A third of transcripts start in Argentina; Peru and Colombia follow. Brazil arrived in September with the Portuguese version and is already fourth. By city, Lima leads, then Bogotá, Buenos Aires, Córdoba and Arequipa.

  • Argentina33.2 % · 820
  • Peru14.6 % · 361
  • Colombia11.3 % · 280
  • Brazil8.1 % · 199
  • Mexico5.6 % · 139
  • Spain5.3 % · 132
  • United States3.3 % · 82
  • Chile2.1 % · 53

Cities with the most transcripts started

  1. Lima · 205
  2. Bogotá · 135
  3. Buenos Aires · 62
  4. Córdoba · 46
  5. Arequipa · 40

Device

  • Desktop · 72.3 %
  • Phone · 26.7 %
  • Tablet · 1 %

Afternoons, Mondays and Sundays

The peak hour is 17:00 to 18:00 in Bogotá and Lima (19:00 to 20:00 in Buenos Aires), with a second peak mid-morning. Almost nobody transcribes between 3 and 5 in the morning. Mondays and Sundays are the busiest days: what was recorded during the week gets turned into text before the next one.

00h03h06h09h12h15h18h21h
Transcripts started by hour of day (Bogotá/Lima time, UTC−5)
SunMonTueWedThuFriSat
Completed transcripts by weekday

Word wins: half of all exports

More than half of accounts came back to transcribe at least once more and 8% made five or more transcripts; the most active account is at 46. When people export, it is Word half the time, PDF one in four, TXT one in six; SRT subtitles are 5%. Speaker labels are used on 1.6% of transcripts, almost always interviews and meetings.

Export formats

  • Word (DOCX)50.4 % · 143
  • PDF22.9 % · 65
  • Text (TXT)17.3 % · 49
  • Subtitles (SRT)5.3 % · 15
  • CSV2.1 % · 6
  • Markdown1.8 % · 5

What the file names say

We do not read content for this report; we do count which words appear in file names. WhatsApp appears in 374 names, interview in 43, class in 27 and meeting in 16. There are also church sermons and podcasts, few but steady.

  • WhatsApp374 % · 374
  • Interview43 % · 43
  • Class / lecture27 % · 27
  • Meeting16 % · 16
  • Church / sermon5 % · 5
  • Podcast3 % · 3

How it was made

  • Source: the JustTranscribe database, completed transcripts between 4 August and 15 September 2026 (1,909), team accounts excluded. Country, city and device come from PostHog (2,471 transcripts started, IP geolocation, no cookies).
  • Aggregates only. No title, text or personal data was read or is published; city counts are shown only from 40 upwards.
  • Language is detected by the transcription model from the first seconds of audio; length is that of the original file or video.
  • Six weeks of beta with search ads in Argentina, Peru, Colombia, Mexico, Chile and Brazil: the country mix also reflects where ads ran.
  • Updated monthly with the same method; changes are listed below.

Cite or reuse

Figures and charts may be quoted freely with a link to this page. Suggested citation: JustTranscribe, "What Latin America Transcribes", September 2026.

Download the tables (CSV)

Changes

  • 15 September 2026 — first edition: 1,909 transcripts, 4 August to 15 September.

Questions about the data: bramontiventures@gmail.com