Earmark's Big Bet: Hiding the Transcript to Save the Product

The Product Pivot: From Transcript Central to Subtitle Support

Earmark’s initial product vision, as detailed by @ttorres on February 8, 2026, at 6:14 PM UTC, was rooted in radical transparency: the live transcript was given primary visual real estate, consuming a staggering 50% of the screen. This design choice reflected an early commitment to the raw, immediate output of their speech-to-text engine. However, this commitment soon illuminated an unforeseen user dynamic. Instead of skimming the video or simply relying on the audio, users were engaging in intense, almost surgical focus on the transcript detail itself. They treated the live feed not just as a supplement to the audio, but as the definitive source document in real-time, turning the product experience into an exercise in detailed textual scrutiny.

This intense focus quickly revealed a vulnerability in the user experience. When the tool was presented as definitive, users expected perfection, turning the interface into a proofreading station rather than a content consumption platform. The dedication to showcasing the live process inadvertently created friction when that process faltered, setting the stage for a necessary, if counterintuitive, strategic shift.

Identifying the User Friction Point

The problem became immediately apparent: users fixated relentlessly on minute inaccuracies within the sprawling transcript window. Every misplaced comma, every misspelled name of a lesser-known guest, and every predictable transcription error became a glaring obstacle. This intense scrutiny generated an organic, powerful demand from the user base: the need for manual correction. Users wanted the authority to edit the live feed as it generated, effectively demanding that the product serve dual roles—a content consumption tool and a real-time transcription editor. This desire to correct the source material indicated that the perceived value of the accuracy of the transcript was overshadowing the value of the final output the system was designed to create.

The Strategic Solution: Minimizing Distraction

The response to this user-driven friction was not to perfect the transcription engine overnight—a notoriously difficult feat—but to strategically re-engineer the presentation layer.

The Transcript Downgrade

The dramatic decision was made to execute what can only be described as a controlled Transcript Downgrade. The prominent 50% display was swiftly minimized, reduced to a far less intrusive subtitle bar tucked discreetly at the bottom of the view. This was a calculated acknowledgment of the limitations of current AI. Large Language Models (LLMs), while powerful, still produce noise, especially under the pressure of real-time inference. The product team decided on a philosophy of acceptance: Leveraging LLM capabilities for error tolerance and inference ("let them"). If the AI was robust enough to understand the context despite minor errors, the user shouldn't have to see those errors constantly.

The primary goal of this visual demotion was not merely aesthetic refinement; it was psychological engineering. By minimizing the visible transcript, the product actively directed user attention away from transcription errors and back towards the intended primary interaction points.

Refocusing on Core Value Generation

By successfully hiding the "noise"—the inevitable, yet distracting, errors inherent in real-time transcription—the user experience saw an immediate, measurable improvement. Users stopped editing and started absorbing. The successful shift allowed users to concentrate on the resulting artifacts generated by the system, which, presumably, were the derived summaries, key takeaways, or structured data that Earmark was ultimately supposed to produce. The signal finally rose above the noise, proving that sometimes, the best way to improve perceived accuracy is to reduce the visibility of imperfection.

Further Listening and Exploration

This strategic product pivot, which prioritized user flow over raw transparency, offers a compelling case study in modern SaaS design, particularly when integrating nascent AI capabilities. The context of this decision, shared publicly on February 8, 2026, provides insight into the rapid evolution of tool design in response to real-world user data. For those who wish to explore the full rationale behind this bold move and delve deeper into the lessons learned from centering or de-centering core AI outputs, the full episode provides comprehensive details.

👉 Find a link to the full episode here:

• Spotify: buff.ly/6vyu5jU
• Apple Podcast: buff.ly/0OKsd5v
• YouTube: buff.ly/Z4omRiH

Source Shared by @ttorres: https://x.com/ttorres/status/2020561817450295317