Ex-YouTube Employee Explains the Algorithm

Summary

Former YouTube Partner Manager and content strategist Nastia Debowska explains that YouTube’s recommendation system is designed to match videos with viewers—not to reward creators for using particular features. Distribution expands when new groups of viewers respond well and slows when those signals weaken.

For creators, the practical task is to read channel-specific data, identify what is working, and allocate effort accordingly.

How to Diagnose an Underperforming Video

• Don’t overreact to the first 24–48 hours. Real-time view counts are estimates, and view calculations are technically complex. Unless a video is time-sensitive, let it gather data before making a judgment.
• Look beyond channel-wide averages. Compare a video with similar videos in the same content pillar—for example, one mascara review against another—rather than with unrelated uploads.
• Compare equivalent time windows. Use YouTube Analytics to compare each video’s first week, rather than comparing a new upload with an older video that has had much longer to accumulate views.
• Check audience mix as well as CTR and retention. A video may perform well with loyal viewers but gain little traction among new viewers. That can limit further recommendations even when the headline metrics look healthy.
• Review packaging before concluding the topic failed. If the video has promise, test whether its title and thumbnail communicate its value effectively.

Algorithm and Distribution

• Debowska describes recommendations as a series of increasingly broad audience tests. If viewers respond well, YouTube may show the video to less familiar audiences. Distribution can slow when those audiences respond less positively.
• Impressions are not guaranteed. Low impressions may indicate that a video found an audience among loyal viewers but did not gain enough traction with broader or newer audiences.
• YouTube’s business incentive is viewer satisfaction and continued use. Creators should assess performance in terms of audience response rather than assuming YouTube is withholding reach arbitrarily.

Practical Playbooks and Operating Principles

• Analyze channel-specific patterns. Identify what has worked on a particular channel and build on it. Debowska says there is no universal content blueprint.
• Allocate resources using a Pareto-style approach. Find the content areas that produce the most value and double down on them. Then decide whether weaker areas are worth fixing or whether that time and budget should be reinvested elsewhere.
• Prioritize the next upload over endless rescues. For frequent, lower-stakes publishing, Debowska generally favors applying lessons to the next episode rather than repeatedly changing old packaging. For expensive, infrequent, high-stakes videos, early title or thumbnail changes can make sense when the team has enough data and judgment.
• Repurpose the substance, not just the link. Rather than using a Community post simply to direct people to a video, turn useful ideas, examples, or facts from it into a standalone post, carousel, or image-and-text hook.
• Treat Community posts as their own content format. YouTube distributes them on surfaces such as the home page, so they can reach viewers beyond a video’s existing audience.
• Don’t over-segment subscribers into “good” and “bad.” According to Debowska, a subscriber who doesn’t watch every upload does not inherently harm the channel. Inactive subscribers are a natural result of how subscriptions work and of audiences’ interests changing over time.
• Don’t use features expecting an algorithmic bonus. YouTube does not automatically grant more impressions because a creator tries a new feature. Partner managers may help some creators access feature tests earlier and relay feedback, but this is not a growth guarantee.

Metrics and Examples Cited

• Client example: A channel received 10 million impressions in a given period, with 1 million from Community posts—about 10% of its impressions.
• Video-diagnosis example: The host described educational content with roughly 7% CTR, about 7 minutes’ average view duration on a 15–20-minute video, but only 10,000 impressions. Debowska recommended checking new versus returning viewers and comparing the upload with a similar video over the same first-week window.
• Debowska’s mother’s channel: The channel has close to 700,000 subscribers, mostly women over 50. Debowska used it to illustrate how a video can first perform with a loyal audience and then be tested with progressively broader groups.
• No explicit growth targets or timelines were set in the discussion.

Platform and Strategy Outlook

Debowska expects AI-driven features, including a conversational “Ask YouTube” experience discussed in the episode, to change discovery and search behavior. She notes that some straightforward problem-solving videos may already be losing traffic as users seek answers through AI tools. Her recommendation is to monitor how these product changes affect discovery and adapt the channel’s strategy rather than assume current search patterns will remain stable.

She also urges creators to view platform changes—including the introduction of Shorts—from YouTube’s business perspective: the company must respond to market shifts to retain viewers, even when those changes create friction for creators. A Partner Manager can offer data-informed advice and act as a platform contact, but cannot compel growth, control recommendations, or take over a creator’s production and creative decisions.

Presenters and Sources

Guest Nastia Debowska is a content strategist and former YouTube Partner Manager. The hosts are Devin Robbins and Olina Cavedoni. Examples and recommendations draw on Debowska’s client work, her former YouTube experience, and YouTube Analytics.