Twitter/X's algorithm amplifies low-credibility content and politically divisive posts, with 50% of 'For You' timeline content coming from accounts users don't follow

A peer-reviewed six-week audit of X's algorithmic content recommendations during the 2024 US Presidential Election, using 120 monitoring accounts, found that approximately 50% of tweets in users' 'For You' timelines are personalized recommendations from accounts they do not follow. The study found that tweets containing low-credibility URL domains perform better than tweets that do not, with high-engagement tweets more likely to receive amplified visibility when containing low-credibility content. High toxicity tweets and those with partisan bias see heightened amplification. A separate 10-day experiment with 1,256 volunteers provided causal evidence that exposure to algorithmically amplified divisive content alters political polarization. Why it matters: half of what users see in their primary feed is algorithmically selected from outside their chosen network, so users are disproportionately exposed to emotionally charged, outgroup-hostile content that the algorithm identifies as engagement-maximizing, so political polarization measurably increases as the algorithm creates filter bubbles that amplify extreme viewpoints, so users report that algorithmically selected political tweets make them feel worse about political opponents and do not match their stated content preferences, so democratic discourse degrades as the platform's engagement optimization systematically favors divisive misinformation over accurate, nuanced content. The structural root cause is that X's engagement-based ranking algorithm equates engagement (clicks, replies, shares) with value, but divisive and low-credibility content generates disproportionate engagement through outrage and controversy, creating a system that algorithmically rewards the most harmful content types.