Social Media Data & Algorithmic Manipulation
The feed does not show you what happened. It shows you what will make you stay.
The Feed That Shaped a View
Elena had not changed her political views in years. She had not sought out extreme content.
Over 14 months, the algorithm had learned she paused on outrage posts. It served more. She paused longer. It served more extreme versions. She followed a few accounts. The algorithm learned from that. By month 14, her feed contained content that would have shocked her in month one - but had arrived so gradually she did not notice the distance she had travelled.
She had not been hacked. She had not been deceived in any single moment. She had been optimised.
What Is Actually Happening
3.2x
more engagement generated by content that provokes moral outrage vs. neutral content.
Platforms optimise for engagement. Outrage is engagement. This is not a side effect.
Source: Yale Department of Psychology, Brady et al., 2017; replicated in 2023 meta-analysisEverything You Do - and Pause On
Social platforms track likes, shares, and comments. They also track dwell time - how long you pause on each post before scrolling. A 2-second pause on content counts as engagement even if you keep scrolling.
Used for Ad Targeting
Meta confirmed in a 2023 court filing that it scans the content of Messenger messages to improve ad targeting. End-to-end encryption was not default on Messenger until 2023, meaning years of messages were accessible.
Amplified by Design
A 2021 Facebook internal report found their own algorithm was "a significant contributor to political polarisation." The recommendation was not implemented. Engagement from divisive content was too valuable.
Suppressed Without Notice
Algorithmic suppression - reducing content reach without telling the creator - is used across platforms. TikTok's leaked 2019 moderation guidelines showed specific suppression criteria including disability, political content, and low "quality of life."
What Platforms Actually Collect
Engagement signals
Likes, shares, comments, saves, and click-throughs are the most visible. But dwell time, scroll speed, and return visits are equally important and entirely invisible to you.
Inferred attributes
From your engagement pattern, platforms infer: political lean, relationship status, income bracket, pregnancy, health conditions, and emotional state. These inferences are used for ad targeting even if you never declared them.
Off-platform activity
Meta's Off-Facebook Activity tool lets you see data sent to Meta from third-party sites and apps. Most users are surprised by how much activity is reported by sites they visited without logging into Facebook.
Private messages
Before end-to-end encryption became standard (Meta Messenger: late 2023, Instagram DMs: still rolling out), message content was accessible and used for ad inference.
Deleted content
On most platforms, deleted posts are not immediately removed from training data, ad profile inference, or internal analytics. Deletion removes the public post. It does not erase the record of what was in it.
Try It: The Engagement Trap
Scroll through 8 posts. Click what interests you. Skip what doesn't. See what the algorithm inferred.
What That Just Showed You
1. Pausing is engaging. You did not like or share anything. You paused. That was enough to build an inferred profile. Dwell time is one of the most powerful signals platforms use.
2. The algorithm infers, it does not ask. None of the attributes in your profile were declared. They were inferred from which posts you paused on and which you skipped. Those inferences are then used to shape what you see next.
3. The feed is not neutral. Content that generated strong reactions in the simulation was served to you not because it was accurate or important, but because engagement data predicted you would react to it.
Three Things Worth Doing
1. Check your off-platform activity on Meta. Facebook: Settings > Your Facebook Information > Off-Facebook Activity. This shows all the sites and apps that have sent data about you to Meta. You can clear it and disconnect future tracking.
2. Download your data from each platform. Instagram, Facebook, TikTok, and Twitter/X all allow full data downloads. The file shows everything the platform holds about you - including inferences.
3. Use chronological feeds where available. Instagram, X, and LinkedIn allow switching to chronological order. Chronological feeds are less engaging by design - and far less optimised to trigger reactions.
One Question Before You Continue
You scroll past a post without liking or commenting. Does the platform record this as a signal?