How We Hacked TikTok: Data Analysis of 31 Videos Reveals What Actually Drives Views

We reverse-engineered what makes TikTok videos go viral

Everyone has theories about the TikTok algorithm. Most are wrong.

Instead of guessing, we did something different: we pulled the raw data from 31 TikTok videos, tracked 14 performance metrics per video, and ran correlation analysis and multivariate regression to find out what actually predicts views and engagement.

The results surprised us. Some "common knowledge" turned out to be completely backwards. Here's the full research — including the original slide deck, the data visualizations, and exactly what we changed.

The full research deck

We published the complete analysis as an interactive presentation. You can browse every chart, every data table, and every insight below.

The headline numbers

Before diving into the methodology, here are the top-line findings:

What drives views on TikTok?

We calculated the Pearson correlation coefficient between each metric and total view count. A value of +1.0 means perfect positive correlation; -1.0 means perfect inverse correlation; 0.0 means no relationship at all.

The standout finding: comments have a correlation of exactly 0.00 with view count. Despite being one of the most-discussed "algorithm signals," our data shows no statistical relationship between comments and views. Videos went viral with zero comments. Heavily-commented videos sometimes flopped.

Saves (r = 0.93) and shares (r = 0.89) dominate. These are high-intent actions — a user who saves or shares a video is telling the algorithm the content has genuine value.

What drives engagement rate?

Views and engagement aren't the same thing. A video can get millions of views with a low engagement rate, or a modest view count with intense engagement. Here's what correlates with engagement rate specifically:

Here's where it gets interesting: the signals that drive views (saves, shares) are different from the signals that drive engagement rate (comments, followers, watch time). Comments — which had zero correlation with views — are the strongest predictor of engagement rate (r = 0.61).

This reveals a split in viewer behavior: silent watchers save and share (driving views), while active commenters engage deeply (driving engagement rate). They're different audiences doing different things.

Before and after: the impact

After identifying these patterns, we restructured our content strategy to optimize for saves and shares instead of comments. We also adjusted video length and hooks. Here's what happened to average views per video:

Multivariate regression: what predicts views?

Correlation tells you what moves together. Regression tells you what causes the movement when you control for everything else. We ran a standardized linear regression with views as the dependent variable:

Even after controlling for all other variables, saves remain the single strongest predictor of view count (coefficient: 0.52). Shares come second (0.34). Comments are effectively zero (-0.03).

The negative coefficient on engagement rate is counterintuitive but makes sense: viral videos dilute engagement rate. When a video hits millions of views, most of those viewers are passive — they watch but don't interact. High views and high engagement rate are somewhat at odds.

4 surprising findings

1. Comments don't drive views — at all

The "comment to boost the algorithm" advice is a myth, at least for views. Comments correlate strongly with engagement rate but have zero predictive power for view count. Stop optimizing for comments if your goal is reach.

2. Saves are the ultimate algorithm signal

Saves had the highest correlation (r = 0.93) and the highest regression coefficient (0.52). A save tells TikTok: "this content is so valuable I want to come back to it." It's the strongest signal a viewer can send.

3. Shorter videos get more views, longer videos get more engagement

Duration has a slightly negative relationship with views (-0.02 correlation, -0.07 regression coefficient) but a positive relationship with engagement rate (0.51 correlation). The right length depends on your goal.

4. Viral reach actually hurts engagement rate

The impressions-to-reach ratio has a -0.41 correlation with views but a +0.32 correlation with engagement rate. As a video goes viral, it reaches less-engaged audiences, diluting per-viewer engagement.

What we changed

Based on the data, we made five specific changes to our content strategy:

• Optimized for saves, not comments. We added "save this for later" CTAs and created content with reference value (cheat sheets, tip lists, frameworks).
• Made content shareable. We focused on "you'll want to send this to someone" moments — surprising stats, relatable situations, useful resources.
• Shortened average video length. We cut videos that were underperforming on views and tested tighter edits.
• Stopped chasing comments. We removed comment-bait hooks ("Comment if you agree!") and replaced them with value-first hooks.
• Used hooks backed by data. SocialGPT's hook generator draws from patterns proven to drive saves and shares, not just generic engagement.

How to run this analysis on your own account

You don't need a data science degree. SocialGPT's analytics dashboard tracks exactly these metrics — saves, shares, watch time, engagement rate — across all your posts. The Post Coach feature automatically identifies which of your videos are getting saved and shared versus just liked and commented on, and gives you specific recommendations to shift your content toward what the algorithm actually rewards.

The research is clear: create content worth saving. That's the algorithm hack that actually works.