Enter your video metrics
to project the next 90 days
After publishing a video, the hardest part is not the wait โ it is not knowing whether the wait means anything. A video that sits at 3K views on day two might reach 200K by day 30, or it might end at 8K. Both outcomes start the same way. The difference is in signals that happen mostly in the first 48 to 96 hours, and most creators do not have a useful framework for reading them.
What makes view prediction genuinely difficult is not a lack of data โ you have CTR, retention, engagement, watch time, and channel history. It is that those signals interact with each other in non-linear ways, and the relationship between day-3 performance and day-30 performance changes depending on which phase of distribution the video is currently in.
This page explains how the model works, what it is actually measuring, and โ importantly โ where the model breaks down. Because it does break down in certain situations, and knowing those situations prevents misreading the output.
Methodology note: this predictor is built on observed view curve archetypes from creator-reported data, not on access to YouTube's internal systems. It models realistic probability ranges, not guaranteed outcomes. Individual video performance can deviate significantly from any prediction based on factors the model cannot observe.
The predictor does not have access to YouTube's internal data. What it has is a model built from observed patterns: how videos with similar algorithm signals tend to behave over 90 days, segmented by format, niche, language market, and channel authority. Here is what each input is actually doing in that model.
The algorithm score combines CTR, retention or completion rate, engagement signals, and channel authority into a single composite. This score then selects which of four curve archetypes โ weak, average, strong, or viral โ the prediction uses as its basis, with smooth blending between them. A score above roughly 1.5 typically corresponds to videos that receive extended browse and suggested distribution. Below 0.8, the curve tends to decay fast after the first day.
Every niche-language combination has a ceiling. An English-language finance video can theoretically reach a much larger audience than the same concept in a smaller language market, not because the content is better but because the addressable pool is larger. The model applies a realistic cap based on estimated market size and niche audience share.
Long-form and Shorts do not just have different RPM. They have structurally different view curves. Long-form typically builds a shoulder plateau around days 4โ14 as browse and suggested distribution kicks in after the algorithm test window. Shorts can spike earlier and harder, but tend to decay faster unless loop signals are strong.
High CTR without retention means the thumbnail and title are doing their job but the video is not. High retention without CTR means the video satisfies the people who watch it, but not enough people are clicking to find it. In the model, CTR and retention are multiplied together rather than added โ which means a weakness in either one creates a larger drag than a weakness in a less important signal.
Shares are weighted more heavily than likes in the model because they have a stronger relationship with extended distribution in practice. A video that gets shared at scale is reaching new audiences by definition. Subscriber gains are weighted similarly because they indicate the viewer found enough value to want more, which correlates with content that performs well in browse features.
The same CTR and retention numbers mean different things on a channel with 500K average views per video versus a channel with 2K average views. The model includes an authority multiplier that reflects this: consistent channels at scale get some distributional benefit from their track record, while newer or inconsistent channels are treated more conservatively.
A model built on historical patterns cannot anticipate structural changes. Here is where this one is most likely to be wrong.
Trend-driven videos violate the assumptions the model is built on. When a topic suddenly becomes culturally significant โ a news event, a viral moment, a policy change โ related videos can have view curves that look nothing like the observed archetypes. The model will usually underpredict these videos significantly.
Very new channels also create estimation problems. The channel authority component of the model is calibrated against channels with meaningful history. A channel with 5 videos has very little pattern to draw from.
Cross-platform spillover โ a video going viral on Twitter or Instagram and driving YouTube views from outside the platform's normal distribution โ is invisible to the model.
And seasonality affects both views and engagement in ways that vary by niche and market. The model does not apply a seasonal correction.
Most advice about YouTube growth treats the process as more predictable than it is. This is where most creators get it wrong โ not through lack of effort, but through applying frameworks that oversimplify a genuinely variable system.
The screenshot problem is significant. One viral video gets posted everywhere. The 40 average videos that came before and after it do not. The result is that most creators have a reference library of exceptional performances, not typical ones.
The same problem applies to 'views x $X' income estimates. The formula ignores niche, geography, retention, format, and ad market timing โ which is to say, it ignores the variables that actually determine the result.
These scenarios illustrate how the same starting-point metrics can lead to different 90-day outcomes depending on niche and audience profile. They are ranges, not predictions for specific channels.
Gaming can produce fast early spikes but has a harder time building the shoulder plateau that extends long-tail distribution. With strong algorithm signals, a 90-day total of 200Kโ280K is plausible. With average signals, the curve often flattens faster after the first week. The variance within 'gaming' is itself wide.
Finance videos with genuine search demand tend to build a longer tail than entertainment content because the topic remains relevant beyond the initial distribution window. A 90-day range of 350Kโ650K is plausible with strong signals and a US-heavy audience.
Tutorials for topics with consistent demand can build a distribution tail that extends well past 90 days. For tutorial channels, month 6 RPM is sometimes more informative than month 1 views.
A realistic 90-day projection might land between 380K and 720K views if CTR and engagement stay consistent with the early signals. The wide range reflects the genuine uncertainty in whether browse and suggested distribution maintains itself through the shoulder phase.
The higher competition density in gaming combined with typically lower watch time usually produces a more compressed 90-day range โ something like 190Kโ350K is plausible under similar signal conditions. Monetization per view is also lower.
Same starting views, same video age, very different probable trajectories. This is why comparing channels without matching niche, language market, retention profile, and audience geography is almost always a misleading exercise.
The most useful way to use this predictor is not to find 'the' forecast for a video โ it is to run several scenarios and understand what conditions would need to be true for each outcome. That kind of reasoning turns a prediction tool into a decision support tool.
Once you model real inputs instead of gut-feel estimates, patterns become visible that are hard to see in analytics alone. A channel that consistently gets strong CTR but weak retention has a different strategic problem than one with the reverse.
This is a structured approximation built from observed behavioral patterns and audience-size constraints. It is most useful for comparative analysis between channel configurations and for identifying which signal variables are limiting performance. It is not reliable for trend-driven content, very new channels with minimal history, or any situation where external traffic sources might significantly affect results.
Use the output as a planning range with honest uncertainty, not as a forecast to be held to.
These answers cover the variables that move the model most and explain where outcomes diverge between videos that look similar on the surface.
The YouTube Views Predictor uses aggregated creator-reported data and observed distribution patterns to model realistic 90-day view curves. Real performance can still shift significantly because of trend events, platform changes, external traffic sources, competition changes, or audience behavior that does not match historical patterns. Use this tool for planning and education, not guarantees. Terms of Use and Privacy Policy.
Enter your video metrics
to project the next 90 days