The 3-Layer YouTube Recommendation Model: A Practical Way to Diagnose Why Videos Stall

The majority of “algorithm” suggestions are more about your vibes than anything else: post consistently, post often, and use hashtags. Your video will generally do fine after one day of posting, then just stop.

I view YouTube recommendations as a three-layered process that can be measured with concrete data: packaging (impressions-to-click), session value (duration of watch time and continuation of viewing), and viewer satisfaction (return visits and indirect indicators of viewer happiness). For teams to benefit from YouTube promotion packages as a distribution tool, both packaging and session value need to be supported by data first; otherwise, paid distribution simply masks a lack of real audience signal strength.

It’s amusing that while YouTube Studio’s UI labels will frequently change, the mechanics behind them have consistently stayed the same throughout time. The impressions to clicks to view sessions to satisfaction to repeat viewing model has remained unchanged since inception my understanding is it remains unchanged today as well.

Layer 1: Packaging signals

In simple words; Packaging includes Cover image, Title and Clarity/Topic of Content. It sits in the relationship between how many people see your packaging and how many people click on your packaging. Additionally, the packaging will provide insight into how long people stay after they arrive (due to their expectations not being met within the first 15 seconds).

YouTube has always been transparent regarding how they optimize for viewer behavior rather than creator intent. The video from the Creator Liaison (Youtube) reiterates this concept, stating that recommendation systems track how and what types of videos are chosen by viewers and have a large following. This means that packaging is the “first yes.”

A typical sign that YouTube is not even trying out your material more widely is that your impressions are extremely low, meaning your impressions are far less than the number of views compared to that.

• Assess how well you have targeted a particular topic. If your title has the potential to match five other titles, then the system has lower confidence in where to put the video.
• This chart shows the Split of Buys versus Suggested Items. If the Suggested Item is almost zero, your catalog may not be compatible with your adjacent catalogs.
• Audit the thumbnail at phone size. If the focal point is unclear, CTR rarely recovers later.
• To evaluate this experiment: Show the video thumbnail and title to a friend for two seconds and ask them what they think the video is about. If they get it wrong, so will your audience on YouTube.

Layer 2: Session value

When discussing session value, many people will say that they care about session value, but when measuring it they do a poor job. Session value does not simply refer to the average view duration of a video, but instead refers to whether the session feels valuable and whether the session resulted in increased retention/usage of YouTube.

The most common indication of symptoms would be that you have very high impressions but only low views (clicks). Usually, this indicates the user sees your ad but isn’t clicking it; generally defined as a packaging problem. However, if you have a good amount of clicks but are seeing continued declining growth, then you are likely to be experiencing session problems.

I have noticed so many artists post music videos that have awesome intro shots, which take around twenty-five seconds to get spicy (the hook). Artists’ CTR (click-through rate) is fine because fans are already in. At the point that their “waiting to get there” energy is over, their retention crashes. So YouTube tests the video, “not enough watch time”, then they turn off the traffic.

Fixes here are rarely fancy:

• Re-edit the first half minute of film – it’s not about “speed” in a generic way, but it’s about clarity on what the benefit is and why I want to hang around.
• Create your chapters as a promise map. When breaking the content down, identify points where people are waiting.
• Consider comparing retention based on your last 5 uploads, not to someone else’s mythical creator. Just beat your baseline first.
• Construct an end screen with a bridge that will link the current video to the next logical video to watch instead of simply dumping a random playlist in the end screen.

In all honesty, most people screw up continuity because they choose their next video based on what they want to market, rather than based on what the viewer will want to watch next. If a viewer had just watched ‘How I did mix vocals,’ then they would most likely want to click the ‘My vocal chain template’ option, and not the ‘behind the scenes vlog’ option.

Layer 3: Satisfaction proxies

Satisfaction is what layer makers feel but can’t quantify. YouTube provides direct feedback methods (surveys) but you can still monitor the same things you’ve been monitoring: number of return viewers, number of likes and dislikes on a given video, and number of comments from users on a video. Also, you can monitor the number of users returning for another upload.

If you have a catchy title and also pace the video evenly, you could achieve a “win” in terms of CTR. However, if viewers feel tricked or annoyed after watching the video, the algorithm will remember this when it comes time to consider your video for future recommendations. More information on this topic has been provided by YouTube in the Recommendations Guide, specifically referencing Shaped. This means that YouTube will reward videos that produce positive viewer signals (such as likes, comments, and shares) and keep viewers engaged as well, not just those that receive high CTR.

Common symptom: A drop-off in views within 24 hours of a video’s posting. Your first 24 hours are your warm audience – subscribers, viewers who have come back to watch again, and people you may have driven from social media. After your first 24 hours have passed, YouTube will test a wider audience. If that wider audience is not satisfied, the video will receive decreased views/distribution.

• Check the comments section for any signs that people may be confused. If there are questions such as “Wait, what is this video actually about?” you are not fulfilling your promise based on what you actually delivered.
• If the number of returning viewers drops over a seven-day period, that’s an indicator of lost confidence while still having reach (viewership).
• Examine the velocity of likes instead of just the total number of likes. If a video stops receiving likes, it may be an indicator of audience dissatisfaction.
• Consistently checking audience retention drop-offs at a similar timestamp within your uploads will help identify repeated issues due to bad habits in editing and/or structure versus single events.

A simple diagnostic flow

I use this type of decision tree for working with small teams, and with individuals as well; it requires no dashboards or special tools other than the Studio app and the notes app(s) on your device.

1. If you are not receiving many impressions, it could mean that your packaging and/or clarity of content are off. YouTube needs to test you as broadly as possible before any other factors have any real impact on your impressions.
2. If your views remain stagnant despite having lots of impressions, it could mean there’s a problem with your click-through rate. Try changing your thumbnail and title, and ensure that the first 10 seconds of your video meet the expectations of your audience.
3. If you have a good CTR yet low retention, You have a session value issue that needs to be fixed by improving the opening, reducing the amount of dead space in your video, and increasing the continuity of your end cards.
4. If playback remains good while retention is high, check satisfaction proxies. Is the response of new users positive? Are viewers who receive the next upload returning?
5. When you see slow growth but everything else appears to remain stable, one explanation could be that your distribution is not adequate. Promotional support could make sense at this time, assuming that the signals have been in place for a period of time.

As someone who has an aversion to hype, I believe PromosoundGroup can be really helpful if your video is well packaged and has good viewer habits, but you just need some initial exposure to start testing it in the marketplace again. It is not a lifeboat; it is a tool.

Weekly review checklist

You should perform this review once weekly (for all channels) for a total time of approx 15 minutes per channel to find and identify which layer is having failure, so you don’t make the mistake of fixing an unrelated layer.

• Top 3 videos by impressions
• Top 3 by CTR
• Top 3 by average view duration
• Worst 3 retention dips (timestamp)
• Returning viewers trend (7-28 days)
• Traffic source mix (Browse, Suggested, Search)
• End screen click rate sanity check
• Comments: 10-second sentiment scan
• One change to test next upload

The secret is clear; select a single layer to improve upon every week. By doing this repeatedly, you will be consistently creating content that the algorithm feels confident recommending!