How Danish transcription platform Good Tape grew from a newsroom hack to 2.5M users globally

When I learned about a transcription startup founded by a team inspired by the painpoints of journalists called Good Tape, my interest was piqued. I’ve been using various iterations of transcription software for over a decade. 

One of my most painful experiences was many years ago while covering the IoT beat for a Silicon Valley publication. The most up-to-date machine-learned embedded software at the time would translate every “IoT” spoken into “coyote”. A painful problem that persisted until I contacted the company and begged them for a solution. 

So I spoke to CEO Lasse Finderup to learn all about Good Tape, a pioneering Copenhagen-based scaleup company specialising in AI-powered transcriptions that convert speech to text. Its trusted by 2.5  million users worldwide while having transcribed over 10 million files. It has reached $2 million in ARR in less than 2 years.

Almost making a journalist cry

According to Finderup, the company was born from a real need. 

“The idea came from a guy who was in the IT department at Zetland, a Danish digital newspaper. He sat next to journalists during lunch breaks, and they would constantly complain: “I have to spend four hours transcribing this interview!”

I can relate. I once had a colleague who would pay me €100 to transcribe interviews. The software was so unreliable that it took ages. I was poor and needed the work, and it was definitely a hard slog!

According to Finderup, his colleague is a real open-source nerd, and around that time, the Whisper model had just come out. He asked Zetland’s CEO if he could try hacking something together. They said, “If you can solve this, go ahead.”

So he did. At first, it was just dragging your file into a folder, and tomorrow morning, it was ready. Finderup shared:

“One of the first journalists to use it almost cried — he couldn’t believe it was done so quickly.  He went to the CEO told him to try and, and build on and do something more with it.That’s when the realised there was something big here.” 

The privacy problem is underestimated

 However, accuracy is not the only translation challenge. There are also challenges related to security, privacy, and user experience. 

As journalists, we want transcription to be fast, accurate and private. Many users raise their concerns on how exactly their data is being used by transcription services.

For example, big transcription players like Rev utilise users’ data “perpetually” and “anonymously” to train its AI systems. Even if you delete your account, it will still train its AI on that information.

This is a huge concern if you are transcribing interviews with whistleblowers as a journalist, or in the case of business, workplace meetings that may include proprietary information. 

How do transcription services approach privacy? 

According to Finderup, there are two approaches in the industry: Open-source models, which rely on public data, and closed models trained on private data—which often don’t advertise this. 

“We chose the first route. We’re very clear: we don’t touch your data. That’s our biggest differentiator.

“We also don’t share, sell, rent, or trade your personal information with third parties for commercial purposes.”

Good Tape prioritises confidentiality when handling sensitive sources and materials, ensuring that customer transcription files are never used for AI learning; in addition, the platform employs industry-standard encryption, processes all files securely within the European Union, is fully GDPR compliant, and offers a Data Protection Agreement as part of the Premium account package.

“We’re the console, not the game” in the race for smarter transcription

Another common problem with transcription is accents. The performance of transcription models is directly tied to how much and what kind of data they’ve been trained on. So yes—English and major languages with lots of available data tend to perform much better, especially with different dialects and accents.

A group of people in a meeting may all be speaking English, and regional accents among Australians, Scots, and non-native speakers can result in a world of pain. And if you extend this globally, the question arises: Is there enough training data for smaller languages or diverse dialects?

Further, when it comes to smaller languages — like Estonian or Ukrainian — or even regional dialects, many models don’t perform as well because there’s simply not enough high-quality data available for training.

According to Finderup Good Tape’s models rely on the open-source community, but the company puts a lot of work into pre-processing audio: file formats, noise reduction, and silences. 

“The base model is the “PlayStation game,” and we’re the “console” that runs it.

English works best because there’s more data. But we see this as a moment of opportunity—especially in Europe, where being privacy-conscious is now a competitive advantage.”

According to Finderup, in terms of UX, ” the platform’s lack of features makes it stand out.

“ We keep things simple, like WeTransfer—just drag and drop. We don’t integrate with OpenAI or ChatGPT for summaries because that would compromise data privacy. We’ve just launched our own in-house language model to offer summaries and transcription chat while keeping everything secure.”

For example, you can generate summaries and chat with the transcription while maintaining access to sources.

“In the beginning of your user journey—let’s take your process as an example—the first thing you do when writing an article is what we’re doing now: having a conversation. Then, of course, you transcribe it.

Good Tape’s approach approach is to gradually support each step of that journey. The first focus was quality assurance: making sure the transcription is accurate, that you can listen back, make edits, and trust the output.

“Next, we added the ability to generate summaries. And now, you can actually chat with your transcription. What’s unique is that when you use our in-house language model, it provides clickable sources directly from the transcript.

So if the model says, ‘Lasse said he hates Microsoft,’ you can click and see the original quote in context—maybe it actually says, ‘I hate when Microsoft does this.’ It’s all about transparency and trust.”

Good Tape stands out for having its own LLM, which means it isn’t reliant on the decisions of big corporations like Open AI.

It also solves the common interviewer problem of interviewing a group of people: it’ll show you exactly which part of the audio a quote came from and who said what. 

“We solved that early with speaker labels. If you mention a name at the start—“Cate said…”—the system will follow that pattern. That integrates well with our summary and QA tools.”

When it comes to services with saturated markets I always wonder about churn, Finderup shared that while churn was high at the beginning, it’s improving. 

“Everyone in this space probably sees the same pattern—rapid user growth, but also rapid drop-offs.”

In terms of the future of transcription, Finderup contends that the tech is already really good and that future improvements will be small. He sees the big opportunity is around specific features for specific users — like journalists needing source citations. Transcription will become a commodity. The winners will be those who build useful things around it.

How do you grow in a saturated market? 

Among an abundance of stand alone and embedded solutions within other software subscriptions, the company has grown through trust. Finderup contends that “being based in Denmark — and the EU — helps.”

“A lot of companies sign on just because we’re GDPR-compliant. We also let people try the product without signing up. We trust our product to speak for itself.”

Further, big platforms like Zoom or Teams adding transcription features actually helps the company in that it educates users that transcription is possible, and then people look for better tools. One of Good Tape’s top markets is Taiwan, because a Chinese transcription tool was lacking — and they wanted a secure alternative.

The company has even partnered with Chile’s court system to transcribe criminal cases. 

“That was a big moment — we had to be absolutely sure we were secure.”

 What challenges have you faced as AI evolves?

Finderup admits, “honestly? It’s hard to go slow. When you see competitors adding flashy features by cutting corners, it’s tempting. But we keep things simple and proper. 

“We talk about this a lot internally—every time we add a new feature, there’s a cost to simplicity. That’s one of our core values. When we say, ‘Oh, now you can also do this,’ it might sound great, but from a user’s perspective, it can actually add confusion. Now they have to figure out what that thing is and how it works.

With AI moving so fast, there’s real value in not rushing. And Good Tape is seeing that payoff now.

For example, when Good Tape says everything is secure, it means it— as it hosts its own large language model. 

“But that also meant it took us four extra months to launch our summary feature, while competitors just plugged into ChatGPT and shipped immediately. We didn’t cut corners, and that’s intentional.”

At first it seems like OpenAI had all the power, but now it’s the “wrappers”—those building on top—that have the leverage. And companies like ours. The ground is shifting constantly.”

As transcription becomes a commodified layer of tech infrastructure, it’s not just about accuracy anymore. It’s about trust, and user experience. Good Tape’s commitment to privacy and clarity makes it more than just another app for journalists, researchers, and anyone handling sensitive information.

And maybe, just maybe, it means no one else will ever have to explain to an editor why their article quotes someone talking about “coyotes” in a piece about the Internet of Things.

Lead image: Good Tape. Photo: uncredited.