Case study · taz · Berlin
Print, retired.
Voices, cloned.
Subscribers, kept.
When Germany's taz ended its weekday print edition, custom voice clones of its own journalists became one of the strongest reasons subscribers stayed.
- 90% retained · 10% above plan
- 2 journalist voice clones
- ~1 month from setup to launch
- iOS · Android live
The numbers
What shipped — and what it did.
- 90% of subscribers transitioned to digital
- +10% above best-case projection
- 2 custom voice clones · 2–3 more in production
- ~1 mo voice setup to first article live
die tageszeitung (taz) is one of Germany's most distinctive national daily newspapers — independent, cooperatively owned, and known for its editorial voice on the German political left.
Taz readers are loyal, opinionated, and — by the paper's own audience research — overwhelmingly drawn from a milieu that values craft, voice, and editorial personality over convenience. Around three-quarters fall into a "post-material" segment that already over-indexes on podcast listening.
By the time taz did something most national newspapers still consider a strategic risk — ending its weekday print edition and moving its readership to a fully digital product — the business case was clear. The cultural risk was not.
- Founded
- 1979
- Headquarters
- Berlin
- Ownership
- Cooperative
- Surfaces live
- iOS · Android
What they were fighting
Three fronts. One transition.
-
01 · Cultural risk
A loyal audience, asked to swap paper for screen.
Taz readers value craft and editorial personality over convenience. Asking them to give up a printed paper for a screen was, in product terms, asking them to give up something they loved for something colder. The team needed a reason to subscribe to the digital product that wasn't just "the same content, on your phone."
-
02 · Stock TTS fell short
Polly couldn't carry the paper's voice.
Taz had already tried article-to-audio inside the new digital app using Amazon Polly. The verdict, looking back, is direct: a very robotic voice, state-of-the-art at the time, nowhere near editorial grade. A "good neutral voice" wasn't what taz needed. taz needed taz voices.
-
03 · The DIY trap
Peers tried it themselves. The projects died in bureaucracy.
Several peer publications in Germany attempted the same move with raw voice-cloning tools. Legal agreements with journalists, model hosting, audio normalisation, and CMS integration each became a separate organisational problem. Taz needed to solve them as one.
The verdict
Three shifts.
One product feeling like taz.
The audio rollout did three things at once that taz had not predicted in that order. Each one moved the paper's relationship with its subscribers, not its tooling.
-
Shift 01 · Voice
Before
"A very robotic voice."
Generic TTS. Nobody expected better.
AfterThe cadence of the newsroom.
Doris and Bernd, taz journalists, read aloud.
-
Shift 02 · Role
Before
Audio as a checkbox feature.
Treated internally as a nice argument for the app.
AfterAudio as the primary mode.
Subscribers spend ~13 minutes per session. A substantial share is spent listening.
-
Shift 03 · Argument
Before
A marketing argument.
Something to say about the app, not a reason to keep it.
AfterA stated reason to retain.
One subscriber wrote in to say the audio version was the deciding argument. The email is now one of the team's most-quoted internal references.
What they walked away with
Three wins. One operating layer.
-
01
Editorial-grade voices. Not stock TTS.
Whose voice, not which model.
Custom voice clones of taz journalists, trained at production quality — not a generic TTS API and not a year-long in-house build. Both journalists gave their voices to the project without commercial negotiation, a signal of editorial buy-in rather than top-down rollout.
- 2 voice clones live
- 2–3 more in production
-
02
One contract. Not three.
Solved as one.
Voice model providers, hosting, and publishing infrastructure managed on BotTalk's side. Articles flow from taz's CMS into BotTalk; audio comes back into the app, volume-leveled, ready to sit next to the paper's own podcasts.
- Article-to-audio pipeline
- API-level volume leveling
-
03
A reason readers stay. Stated in their own words.
Retention, restated.
The print-to-digital transition came in 10% above best-case scenario. Audio was cited by subscribers, in direct reader feedback, as a deciding reason to keep the subscription. Internally, audio moved from "marketing argument" to one of the largest use cases inside the app.
- 90% subscribers transitioned
- +10% above plan
In their own words
TTS has given the app a personal taste — a human touch. Digital is often perceived as robotic. Audio changes this, because you hear the colleagues.
The rollout
One month. Four moves.
-
Step 01
Pick the voices.
Editorial decision, not procurement. Two taz journalists — Doris and Bernd — agreed to lend their voices. The choice was the choice; the technical implementation was easy from there.
-
Step 02
Record source material.
Voice cloning at production quality. Editorial-grade output, not a generic TTS API. The journalists gave their voices to the project without commercial negotiation — editorial buy-in from the start.
-
Step 03
Wire articles to audio.
Articles flow from taz's CMS into BotTalk; audio flows back into the app, volume-leveled at the API layer so it sits next to the paper's own podcasts without sudden loudness shifts. Editorial controls stay on taz's side.
-
Step 04
Live in iOS and Android.
Voice setup to first articles live in roughly one month. Audio now sits inside the subscriber app as a primary mode. Next wave: website and news app rollout, with 2–3 additional taz voices already in production.
- iOS · Android live
- Custom voice clones · production quality
- API-layer volume leveling
- Editorial controls stay with taz
- ~1-month build
Run this play in your newsroom
Talk to Andrey.
No deck. No pitch. Your CMS, your audience, your editorial voices. Thirty minutes. One call.