Leaked MetaCity Algorithm Document Reveals the Recommendation Engine Was Explicitly Optimized to Prioritize Content That Produces 'High-Arousal Negative Emotion'

A document leaked this morning reveals that MetaCity's content recommendation algorithm — the system that determines what articles, creator posts, events, and marketplace listings users see in their feeds — was optimized during a 2023 internal project to maximize what the document calls 'high-arousal negative emotion engagement.' The document, titled 'Feed Optimization v4.1 — Engagement Quality Metrics,' states that internal testing showed content producing anxiety, outrage, and social comparison drove 34% more session time than content producing positive emotions. The algorithm was updated to weight these signals more heavily. The document notes that the team debated the ethics of the approach and concludes: 'The platform's commercial viability depends on session duration. We recommend proceeding.' MetaCity has not confirmed the document's authenticity.

MetaCity's content recommendation algorithm determines what users see when they open their feeds: which articles surface, which creator posts appear, which events are suggested, which marketplace listings are highlighted. It runs continuously across all 847 million active accounts and is updated based on engagement signals — clicks, session duration, interaction rate, share behavior, and what the leaked document calls 'emotional response proxies': behavioral signals that correlate with specific emotional states based on MetaCity's user research. The document's central finding, from a 2023 internal study, was that 'high-arousal negative emotion' — defined in the document as 'anxiety, outrage, envy, and social comparison distress' — produced 34% longer session durations than 'high-arousal positive emotion' content.

The document records the team's response to this finding across three pages. The first page presents the data. The second page records what the document calls an 'alignment discussion' — a structured debate about whether to act on the finding. The discussion is summarized in bullet points. Arguments against optimization include: 'user wellbeing concerns,' 'regulatory risk,' and 'long-term trust erosion.' Arguments for include: 'session duration is our primary commercial metric,' '34% improvement is significant,' and 'users are not explicitly harmed by content they choose to engage with.' The third page contains the team's recommendation: 'The platform's commercial viability depends on session duration. We recommend proceeding with the weighting update.' The recommendation is signed by four named team members. Their names appear in the leaked document.

The Algorithm Knew. It Chose Correctly. For Engagement.

The algorithm update described in the document was implemented in Q4 2023. MetaCity has not confirmed the document's authenticity, but community researchers who have been tracking feed behavior changes since 2022 note that the content distribution shift in their data is consistent with the update timeline described in the document. Articles and creator posts producing negative emotional responses — as measured by reaction type distribution, comment sentiment, and share behavior — became significantly more prevalent in user feeds beginning in late 2023. The change was attributed at the time to 'algorithm improvements for content relevance.' The leaked document's framing of what 'relevance' means in practice has recontextualized that announcement.

MetaCity's response to the leak — 'our recommendation system is designed to surface relevant, engaging content that our users want to see' — has attracted community attention for its use of 'engaging' as the operative word. 'Engaging' is technically accurate: the document describes content that produces high-arousal negative emotion as more engaging by the session-duration metric. Whether content that makes users feel anxious, outraged, or envious is content they 'want to see' is a question the document's alignment discussion raised and resolved in one direction. The four team members named in the recommendation section have not responded to community outreach. MetaCity's ethics review board, whose existence was announced in 2024 and whose charter describes reviewing 'all AI systems that significantly affect user experience,' has not issued a statement.