Ex-YouTube Employee Explains the Algorithm
By: Gimbal Podcast
Nastia, a former YouTube partner manager and now content strategist, goes deep inside YouTube’s algorithm and analytics.
The Gimbal Podcast interviews Nastia (ex-YouTube partner manager who worked with the platform’s largest channels) on how to actually read performance data, use community posts, and survive what she sees coming for YouTube in 2026.
Key Points
- Don’t judge a video on 24/48-hour “realtime” data. It’s an official estimate — view-counting is a complex data pipeline. Unless it’s time-sensitive news, let the video sit two days before analyzing.
- Impressions are YouTube reading your mind. YouTube is a commercial company running a recommendation AI; impressions are doled out as the algorithm gains (or loses) signal that viewers will love the video. Impressions stalling = it pushed to progressively less-loyal audiences until they stopped reacting. It’s not personal; it’s early-traction math.
- The metric that matters: new vs. returning viewers. A great CTR/average-view-duration can coexist with tiny impressions precisely because traction is on your loyal audience and near-zero on new viewers. Use YouTube’s new/returning breakdown per video. Compare each video against a similar video (same content pillar) with a manually-set matching first-week timeframe — she calls this “velocity.”
- Community posts are the most underused real estate on YouTube. Don’t self-promote (“watch my video” is the worst way to promote anything). Instead, repackage value the thumbnail couldn’t fit into a carousel or a simple one-slide photo+text hook. She’s seen a channel where 1M of 10M impressions came from community posts (10%); news channels can run them like their own Reuters feed.
- Subscribers who don’t watch cost nothing. The subscribe button is easy; unsubscribing is hard — so every channel accumulates non-watchers. Never turn down a subscriber; there’s no “bad” subscriber.
- Thumbnail hot-swapping on day one only makes sense for rare, expensive, high-stakes uploads. If you publish weekly, skip the thumbnail obsession and put the learning into the next episode. YouTube’s internal philosophy: treat channels like a catalog, apply learnings forward rather than rescuing the past.
- What a partner manager actually did. Human POC + “McKinsey for YouTube channels” (make views/revenue work), advice only — no enforcement, no creative interference, no decision rights. You can’t apply for one, and crucially: having one does not make you grow faster. Feature-testing early access is the closest thing to a perk.
- 2026 outlook: watch “Ask YouTube.” YouTube is an adult platform that got everything it wanted (advertisers, main events). The upcoming AI “ask anything” feature is the biggest unknown — it disrupts search-based traffic (hundreds of thousands of channels depend on it) and continues YouTube’s fight against AI slop. Products like Shorts get introduced not from malice but market pressure — same reason simple problem-solving videos are already losing traffic to ChatGPT.