Who actually makes money with AI transcription and dictation apps

Speech-to-text looks like an easy business: open models are free and a working app takes a weekend. The revenue data says otherwise. A handful of apps launched in 2023 earn serious money, while most newcomers with verified numbers make a few hundred dollars a month.

The money in AI transcription and dictation sits with a few apps that launched in 2023, soon after OpenAI released the open Whisper speech model: MacWhisper, Superwhisper, AudioPen and BoltAI. Newer apps launched in 2025–2026 mostly earn hundreds of dollars a month, not thousands. Prices are falling, and the winners now make money through one-off licences for local apps, subscriptions for cloud apps, and team seats.

Below is what is known about each group, how they charge, and why the price floor keeps dropping. All figures were collected on 16 November 2026, with the source next to each one.

The 2023 leaders: what they earn and how they charge

MacWhisper is a Mac app by one developer from the Netherlands. It transcribes audio and video files on the computer itself, without sending them to a server. The Pro version is a €64 one-off licence sold through Gumroad, and a separate subscription version sits in the App Store (Voibe review). A Gumroad sales tracker estimates its lifetime revenue at about $2.2 million (profitable.app). Treat that as an estimate: the tracker counts free downloads too, so the real number is unknown (Steal What Works). The developer gives 30% off to students, non-profits and journalists, which is a good hint about who pays €64 for a transcriber (Voibe, pricing).

Superwhisper does dictation with local or cloud models. It costs $8.49 a month, $84.99 a year or $249.99 for life (TechCrunch, 2 May 2026). Secondary sources say it reached "seven-figure" annual revenue without investors and now has about six full-time staff (aiwiki). There is no primary confirmation.

AudioPen turns a rambling voice note into clean text. Its first version was built in about 12 hours (Rosetta, Starter Story recap). Today it costs $99 a year or $159 for two years (TechCrunch). The founder reports $15–20K a month from more than 5,000 paying users (Indie Hackers). Self-reported.

BoltAI is not a transcriber but a close neighbour: a Mac client for local models and the user's own API keys. It reached $30K a month in March 2025 on perpetual licences, with margins near 90% because it does not resell AI credits, according to the founder (Indie Hackers). The first push came from an X demo video retweeted by Pieter Levels; Setapp came later (aidevcases).

AppWhat it doesPriceRevenueHow reliable
MacWhisperLocal file transcription, Mac€64 one-off; App Store subscription~$2.2M lifetimeThird-party estimate
SuperwhisperDictation, local + cloud$8.49/mo, $84.99/yr, $249.99 lifetime"Seven figures" a yearSecondary sources
AudioPenVoice note → clean text$99/yr, $159 for 2 years$15–20K/moFounder's words
BoltAIMac client for local modelsPerpetual licence, ~$79$30K/mo (Mar 2025)Founder's words

All four share one thing: they arrived early, when Whisper was new and there were few alternatives.

The 2025–2026 newcomers: hundreds of dollars

The picture changes when you look at apps whose revenue is checked against their payment provider. TrustMRR, a startup directory that connects directly to Stripe and similar systems, shows these numbers as of 16 November 2026:

  • Speechyou, meeting recording and transcription: $1,286 in the last 30 days from 84 subscriptions (TrustMRR).
  • Fast-Transcriber: $560 in 30 days (TrustMRR).
  • VoxBee, offline speech-to-text for macOS and Linux: $54 in 30 days (TrustMRR).
  • Yapper, Windows dictation for a €20 one-off licence: $22 in total, all time (TrustMRR).

One more data point outside TrustMRR: the cloud dictation app Voicy started charging in February 2025 and reached about $1,600 a month by August, according to its founder (Voibe).

This is not unusual for small software in general. Across all TrustMRR projects, 67.4% show less than $1,000 a month (TrustMRR stats). But in dictation the gap between the 2023 cohort and everyone after is especially sharp.

Why the price of dictation keeps falling

Three forces push from different sides.

Free open-source code. Handy is a free, open dictation app for Mac, Windows and Linux (TechCrunch). VoiceInk is open source under GPL, sells lifetime licences for $25–49, and its code has already been forked by others (ecosyste.ms). When the core can be copied in an afternoon, nobody can charge much for the core.

Lifetime licences at a coffee price. Dictato sells offline dictation for €9.99 for life, with two years of updates, using open models and Apple's on-device speech engine (TechCrunch). TechCrunch counts "dozens" of dictation apps on the market.

Investor money at the top. Wispr Flow raised $30M and then $25M during 2025 (thebridge.jp). It offers a free desktop tier of 2,000 words a week and paid plans from $15 a month (TechCrunch). Latka estimates its revenue at about $10M a year (Latka). A funded company can afford a generous free tier; a solo developer cannot.

So plain dictation is squeezed from below by free code and €10 licences, and from above by a well-funded leader. The space in between is thin.

One-off licence, subscription or team seats?

The payment model follows the cost of serving one more user.

One-off licence when processing is local. If transcription runs on the user's computer or on the user's own API key, an extra customer costs the developer almost nothing. That is why MacWhisper (€64), BoltAI (about $79) and VoiceInk ($25–49) can sell once and still keep healthy margins (TechCrunch, zekaiwork).

Subscription when processing is in the cloud. Every minute of audio costs money on a server, so cloud apps charge $8–15 a month. AudioPen even moved from a $19 lifetime deal in 2023 to an annual plan (Indie Hackers).

Team seats for scale. The clearest example is next door to transcription. TypingMind, a chat interface that started with personal lifetime licences at $39–99, got more than half of its $130–160K monthly revenue from team plans by late 2025, according to a secondary source (nomadpreneur). In meeting notes, Meetily, an offline open-source recorder, charges $10 per user per month on annual billing (Meetily), and Granola starts at $18 a month (Granola).

Bundles like Setapp. Setapp is a subscription that gives Mac users a catalogue of apps. It pays developers 70% of each subscriber's fee, split by how much each app was used, plus 20% more if the developer brought that subscriber in. For single-app sales the developer keeps 85% of a subscription and 75% of a one-off purchase (Setapp, membership revenue, Setapp, marketplace). No public Setapp revenue figure exists for any single transcription app.

The larger small players now mix models: lifetime for individuals, a subscription in the App Store, seats for teams. Superwhisper sells a monthly plan, a yearly plan and a lifetime licence side by side.

Does language quality still matter?

For English and the largest languages, open models are already good. In one test Whisper large-v3 made errors in 5.2% of English words and only 2.7% of Spanish words (inferenceapis). The share of wrong words, called word error rate (WER), is the standard quality measure.

Smaller languages are a different story. The same test lists 27 of 83 languages as not usable without heavy correction. For Serbian, Croatian and Bosnian, a benchmark posted in a GitHub pull request shows Whisper large-v3-turbo at 13–22% WER, while a cloud service scored 2.8% on Serbian (LokalBot, PR #197). For these languages the gap is not "no service exists" but "no good quality offline".

That matters for the business question. For English dictation the model is a commodity. For some smaller languages, careful tuning still makes a visible difference.

What this means if you are choosing a tool

For a user, the falling price is good news. If you just need a recording turned into text once, a browser tool is enough; for example, online audio-to-text handles a file without installing anything. If recordings must not leave your computer, the market is full of offline apps: MacWhisper on Mac, open-source options like Handy and Meetily, or our own offline transcription program Konspekt.

For a builder, the lesson from this data is less comfortable. The winners of 2023 had timing and early press: Om Malik and heise wrote about MacWhisper (om.co, heise), and BoltAI got a retweet from a large account. A new general-purpose Whisper wrapper starts with neither. The products that spread on their own are easy to copy, and the ones that are hard to copy, such as a niche language or a specific profession, have to be sold by hand.

How reliable are these numbers

Only the TrustMRR figures are checked against a payment provider. AudioPen, BoltAI and Voicy numbers come from their founders. MacWhisper's figure is a third-party estimate that includes free downloads, Superwhisper's comes from secondary sources, and TypingMind's team-plan share is also secondary. Wispr Flow's revenue is a Latka estimate. The language benchmarks come from a blog test and a GitHub pull request, not peer-reviewed studies.

And there is survivorship bias: we hear about apps that made money. Failed transcription apps rarely publish anything, so the real odds for a new entrant are probably worse than this list suggests.

What next

More guides

All blog posts