What 580 Transcription Jobs Reveal About Audio Use
BrassTranscripts looked at 580 completed transcription jobs from its own production database and published exactly what it found: how long files actually run, how many speakers show up, which formats people upload, and what they paid for it. Every number comes from a 180-day window of real completed jobs, not from a survey or an industry estimate.
Worth saying upfront: BrassTranscripts is built by Copper Sun Content and Creative, the same company that builds Brass-SEO. This post curates BrassTranscripts' own research. Brass-SEO did not run this analysis and has not independently verified the underlying data beyond reading the published methodology.
Quick Navigation
- Where This Data Comes From
- The Median Job Isn't What Marketing Decks Assume
- Duration: Files Cluster at Both Ends
- Speaker Count Predicts the Recording Type
- Five Buyer Patterns in the Data
- File Formats: What Device Was in Someone's Hand
- What People Actually Paid
- What This Data Doesn't Tell You
- Why a Business Owner Might Care
- Frequently Asked Questions
Where This Data Comes From
BrassTranscripts pulled its numbers from a 180-day window of its own completed, paid jobs ending May 2026 — 580 jobs, 252 hours of audio processed. The company excluded deleted, silent, and failed jobs before analyzing anything.
This is the kind of number a survey can't produce. Nobody was asked what they thought they uploaded or how long they guessed their file ran. BrassTranscripts pulled it from what customers actually paid to have transcribed. Speaker counts come from automated AI diarization rather than manual review, and BrassTranscripts is upfront that this is BrassTranscripts' own production data, not an independent study. Read the full methodology and every underlying number in BrassTranscripts' original post, What Real Transcription Buyers Look Like in 2026.
To be clear about where the line sits: BrassTranscripts collected these numbers and published the findings. Brass-SEO is republishing and citing that work because it's a genuinely sourced data point, not because Brass-SEO ran any part of the analysis. Every figure in the sections below traces back to that one BrassTranscripts post — we haven't added, rounded, or extrapolated beyond what BrassTranscripts stated.
The Median Job Isn't What Marketing Decks Assume
The median BrassTranscripts job runs 5.3 minutes with 2 speakers, uploaded as an M4A file. That's a phone call or a short voice conversation, not the hour-long board meeting that transcription tool marketing tends to picture.
The average tells a different, more misleading story. BrassTranscripts' mean file length comes out to 29 minutes, pulled upward by the 35 files that ran longer than 2 hours. Report the mean alone and you'd conclude the typical buyer is transcribing long recordings. Report the median and the picture flips: most jobs are short.
Duration: Files Cluster at Both Ends
Nearly half of BrassTranscripts' jobs run under 5 minutes, and roughly a third run between 30 minutes and 2 hours. Very little falls in between.
| Duration | Jobs | Share |
|---|---|---|
| Under 5 min | 287 | 49% |
| 5–15 min | 53 | 9% |
| 15–30 min | 43 | 7% |
| 30–60 min | 94 | 16% |
| 1–2 hours | 68 | 12% |
| Over 2 hours | 35 | 6% |
Short files dominate. A voice memo, a quick call, a two-minute note — that's the single biggest bucket by far. The next-biggest cluster is the 30-minutes-to-2-hours range: real meetings and interviews, not quick notes.
Notice the gap in the middle. The 15-to-30-minute range sits at just 7%, one of the two smallest buckets in the table alongside files over 2 hours (6%). Recordings tend to be either quick and disposable or long and deliberate, with comparatively little in between. That's useful if you're trying to guess where your own recordings fall: a 20-minute call is closer to an outlier length than a typical one.
Speaker Count Predicts the Recording Type
Speaker count sorts BrassTranscripts' jobs into recognizable recording types more cleanly than duration alone does. Solo recordings run short. Group recordings run long, and the more speakers, the longer the average file.
| Speakers | Jobs | Avg. Duration |
|---|---|---|
| 1 | 179 | 5.6 min |
| 2 | 118 | 35.3 min |
| 3–4 | 83 | 45.8 min |
| 5–8 | 61 | 71.1 min |
| 9+ | 37 | 96.6 min |
One speaker means a voice note or a solo recording. Two speakers means an interview or a one-on-one call. Once a recording has three or more voices, average length climbs steadily — 45.8 minutes for a small group, 96.6 minutes once a job hits 9 or more speakers. BrassTranscripts calls speaker identification "the single most valuable AI transcription feature" for professional buyers, and this table is why: speaker count is doing most of the work of describing what kind of recording someone made.
Five Buyer Patterns in the Data
BrassTranscripts grouped its jobs into five recognizable buyer types, built from the duration and speaker-count data above. These aren't labels customers chose at checkout — BrassTranscripts inferred them from the patterns in the data itself. Treat them as a description of what showed up, not a menu of official customer segments.
Voice-Note Buyer
Solo speaker, averaging 5.6 minutes, roughly 179 jobs — about 31% of the total. Founders, sales reps, and researchers turning a quick voice note into text, usually wanting fast turnaround and a plain TXT file to feed into another AI tool. Priced at the $2.50 tier.
Interview Buyer
Two speakers, averaging 35.3 minutes, roughly 118 jobs — about 20% of the total. Journalists, podcast hosts, researchers, and recruiters who need reliable speaker labels and a JSON output for pulling timestamped clips. Priced at $6.00.
Meeting Buyer
Three to eight speakers, averaging 45.8 to 71.1 minutes, 144 jobs combined — about 25% of the total. Standups, board meetings, client calls. Needs speaker attribution and SRT output for captioned clips. Priced at $6.00.
Long-Form Content Buyer
One to two speakers, 60 to 120 minutes, roughly 60 jobs. Podcast producers and lecturers repurposing a long recording into show notes or video captions. Priced at $6.00.
Institutional Buyer
Five or more speakers (BrassTranscripts describes this group as 5–15+), averaging 96.6 minutes, 37 jobs — about 6% of the total. Conferences, depositions, municipal meetings. Needs maximum speaker accuracy and JSON output for further processing. Uses BrassTranscripts' bulk pricing, $1.25 to $6.00 per file depending on batch size.
File Formats: What Device Was in Someone's Hand
Across 624 files in BrassTranscripts' dataset, M4A leads by a clear margin, followed by MP3 and WAV.
| Format | Files | Share |
|---|---|---|
| M4A | 221 | 35% |
| MP3 | 175 | 28% |
| WAV | 104 | 17% |
| MP4 | 70 | 11% |
| OGG | 48 | 8% |
M4A's lead traces back to iPhone Voice Memos, QuickTime, and AirPods becoming standard recording tools for professionals who never open a dedicated audio app. MP4 at 11% suggests most buyers extract audio from video or record audio-only, rather than upload raw video files. OGG's 8% share points to Telegram voice messages and Linux-based recording workflows, both of which use the format by default.
What People Actually Paid
Most jobs landed in BrassTranscripts' cheapest tier. Only a small share needed bulk pricing.
| Pricing Tier | Jobs | Share |
|---|---|---|
| $2.50 (under 15 min) | 340 | 59% |
| $6.00 (15 min–2 hrs) | 205 | 35% |
| Bulk, $1.25–$6.00 (2+ hrs) | 35 | 6% |
BrassTranscripts translates this into rough monthly spend by use case: about $75 a month for someone transcribing a daily voice note, $24 to $30 a month for a weekly interview or meeting, and around $200 a month for a bulk research project. Fifty-nine percent of jobs cost $2.50 or less. That's not the pricing profile of a tool built mainly for enterprise meeting transcripts.
Worth noting for scale: that $75-a-month estimate for a daily-voice-note habit sits in the same neighborhood as a single monthly SaaS subscription, and Brass-SEO's own subscription runs $25 a month, for comparison. Neither product is priced against the other — they solve different problems — but it's a useful anchor if you're trying to picture what BrassTranscripts' pay-per-file model costs against a flat monthly fee.
What This Data Doesn't Tell You
BrassTranscripts is direct about the limits of what a production database can show. The 580 jobs aren't necessarily 580 different customers: as BrassTranscripts puts it, "the database doesn't link single-file uploads to a stable user identity unless the customer signed up for an account." The five buyer patterns above are BrassTranscripts' own inferred labels, not something customers selected at checkout. And there's no geographic data attached to any of it.
BrassTranscripts also names where AI transcription stops being the right tool at all. By its own estimate, automated transcription with speaker identification fits about 95% of business and content-creation use cases — and fails outright for certified court reporter transcripts, real-time live captioning, and adversarial audio where 100% verbatim accuracy is a legal requirement. That's BrassTranscripts drawing its own boundary, not a claim Brass-SEO is making on its behalf.
One more limit worth flagging: this is one company's platform over one 180-day window ending May 2026. It says something real about who uploaded audio to BrassTranscripts specifically during that stretch, and nothing definitive about transcription buyers as a whole, or about what a different transcription tool's customers look like. Read the numbers as a snapshot of BrassTranscripts' own customer behavior, not as an industry-wide benchmark.
Why a Business Owner Might Care
If you record client calls, podcast episodes, or webinars, this data says something concrete about what happens to that audio next: most of it never gets transcribed, and most of what does get transcribed is short. A five-minute voice note is easier to turn into a blog paragraph or a set of show notes than an hour of unedited meeting audio, and BrassTranscripts' own numbers suggest that's the file most people are already sitting on.
That connects to a point we've made about SEO tools generally: most of the value in a workflow comes from a small, specific piece of it, not from every feature a platform offers. If you're already using Brass-SEO's dashboard to find pages and topics that need content, a short recording you already have — a client call, a quick explainer, a voice memo answering a question a customer asked — is raw material for that next page, once it's text.
Look at the Voice-Note Buyer pattern again: solo speaker, 5.6 minutes average, roughly 31% of BrassTranscripts' jobs. That's not a special recording setup. That's someone talking into their phone for less time than it takes to read this paragraph twice. If a customer question keeps coming up in sales calls, or an idea for a blog post is sitting in a voice memo you haven't opened, the gap between "audio that exists" and "text you can publish or feed into an SEO content plan" is smaller than most people assume.
Frequently Asked Questions
Is this Brass-SEO's own research?
No. This is BrassTranscripts' data, pulled from its own production database of completed transcription jobs. Brass-SEO and BrassTranscripts are both built by Copper Sun Content and Creative, but Brass-SEO did not conduct or independently verify this research — this post curates and cites what BrassTranscripts published.
How many jobs did BrassTranscripts actually analyze?
580 completed, paid transcription jobs, totaling 252 hours of audio, pulled from a 180-day window ending May 2026. BrassTranscripts excluded deleted, silent, and failed jobs before running the analysis.
What's the median transcription job length?
5.3 minutes, with 2 speakers, uploaded as an M4A file, according to BrassTranscripts' data. The mean is higher, at 29 minutes, because a small number of files over 2 hours pull the average up — median is the more representative number for what a typical job looks like.
Was this a survey of transcription buyers?
No. BrassTranscripts describes it as production data: "no surveys, no estimates, no extrapolation." Every number comes from jobs customers actually paid to have transcribed, not from asking people what they thought they'd upload.
Does AI transcription work for every kind of audio?
BrassTranscripts says its own tool fits about 95% of business and content-creation use cases, based on this data, but not certified court reporter transcripts, real-time live captioning, or adversarial audio that requires guaranteed 100% verbatim accuracy. That's a limit BrassTranscripts states about its own product, not a general claim about AI transcription tools.