UnSilenced

UnSilenced: Seeing What the Algorithm Hides

Inspiration

We built UnSilenced after realizing that bias on platforms like YouTube is not just about politics or opinions. It is structural.

Recommendation algorithms reward very specific signals. High click-through thumbnails, fast early engagement, repetition, creator scale, and predictability. Over time, creators adapt to these incentives to survive. As a result, the algorithm does not just shape what people see. It shapes what gets created.

We noticed that many videos with real depth, effort, and value never surface. Not because they lack quality, but because they do not align with what the system optimizes for. That invisible layer of influence is rarely acknowledged, even though it quietly determines whose voices are heard.

UnSilenced exists to make that layer visible. It shows users what is being amplified, what is being buried, and why.

What It Does

UnSilenced is a browser extension that adds an optional bias lens on top of YouTube.

It does not replace YouTube’s recommendations. Instead, it gives users two parallel perspectives:

Noise: the existing recommended feed, annotated with explanations for why certain videos are being pushed

Silenced: high-quality videos on the same topics that receive less visibility due to algorithmic incentives

On individual videos and search results, UnSilenced also surfaces alternative perspectives and explains why some viewpoints dominate while others struggle to reach an audience.

The goal is not to tell users what to believe. The goal is to help them understand the forces shaping what they see.

How We Built It

UnSilenced was built as a Chrome extension that runs directly on YouTube pages.

The extension:

Reads live metadata from videos on the homepage, watch pages, and search results

Analyzes signals such as creator size, view velocity, topic repetition, format cues, and engagement patterns

Compares performance relative to creator scale rather than raw popularity

Uses structured AI explanations to translate metrics into clear, human-readable reasoning

For the MVP, we prioritized correctness and transparency over scale. Every explanation shown in the demo is grounded in real data from the videos currently being displayed.

Challenges

The biggest challenge was data integrity.

YouTube does not expose many metrics directly, so we had to carefully combine available signals with conservative assumptions and avoid inferred or fabricated values. At the same time, we had to balance depth with clarity. Too much technical detail can become noise itself.

Another challenge was language. Algorithmic bias here is not malicious. It is incentive-driven. Communicating that distinction clearly and responsibly was essential.

What We Learned

We learned that algorithmic bias is not a single flaw. It is a feedback loop.

When platforms reward certain behaviors, creators adapt. When creators adapt, content becomes more uniform. When content homogenizes, audiences lose access to a diversity of perspectives.

Making these dynamics visible changes how people interpret their feeds. Even small amounts of context can restore agency.