Deleting Posts Won't Fix Social Media. The Ranking Engine Will.

Based on the research ofMirko H. Benischke and Corinne Post, "From Platform Design to Attention Governance: Rethinking Social Media Externalities," Journal of Management Studies, 2026

That is the uncomfortable reframe at the center of Mirko Benischke and Corinne Post's introduction to a *Journal of Management Studies* Point-Counterpoint on social media and democracy. Their essay does not propose new data. It synthesizes a debate among four sets of scholars who disagree about the mechanism behind social media's societal harms, political capture, ecosystem incentive distortion, structural constraint, and shows that despite those disagreements, every account converges on the same underlying variable: attention governance, or "who controls attention allocation, by what rules, and with what accountability." Content moderation, in this frame, is downstream triage. The governance question that actually determines how much damage a piece of misinformation or a polarizing post can do is upstream, in the ranking logic and the business model that decides what gets amplified in the first place.

Two Frames, One Externality

Management and organization scholars have mostly treated platforms as a backdrop to other phenomena, a context in which firms operate, not an organizational actor whose own design choices produce externalities. Benischke and Post push back on that habit. They identify three threads running through the debate they introduce: attention is increasingly treated as a currency and an infrastructure in its own right; the rules that decide what gets seen have shifted from being collectively negotiated (editorial judgment, community norms) to being algorithmically determined; and the decision rules embedded in those algorithms have gotten thinner, collapsing complex social and political tradeoffs into a handful of engagement proxies like watch time, shares, and reaction counts.

That third thread is the one that does the most work. A moderation system can be arbitrarily sophisticated about what counts as harmful content and still be irrelevant to the externality, because the externality is produced before moderation ever gets a vote, at the moment the ranking system decides reach.

The Facebook Files and the Limits of the Content Frame

The clearest documented case of this gap comes from Facebook's own internal research, reported by the *Wall Street Journal* in its "Facebook Files" reporting. In 2018, Facebook rebuilt its News Feed ranking around "meaningful social interactions", a scoring change that weighted comments, shares, and strong reactions more heavily than passive likes. Internal researchers quickly found the change was amplifying angrier, more divisive content, and a 2018 presentation warned that the algorithm "exploits the human brain's attraction to divisiveness," pushing users toward more and more extreme material to hold their attention. According to the *Journal*'s reporting, senior executives, Mark Zuckerberg among them, largely shelved proposed fixes, partly over concerns the changes would reduce engagement and partly over fears they would be seen as targeting one side of the political spectrum.

Note what was and was not at stake in that internal fight. It was not a debate about which posts to take down. It was a debate about how to weight a ranking function, a pure attention-allocation decision, made by an engagement-optimized system whose incentives the company's own employees flagged as corrosive. A year later, the *Journal*'s 2021 reporting on Instagram showed the same pattern at the level of a single harm: internal research found that "thirty-two percent of teen girls said that when they felt bad about their bodies, Instagram made them feel worse." No individual post caused that effect. A feed architecture optimized to keep teenagers scrolling did.

YouTube and the Ambiguous Case for the Algorithm

The clean story would be that the ranking algorithm is always the villain and user preference is a red herring. The evidence is messier than that, which is itself instructive. Manoel Horta Ribeiro and colleagues' audit of YouTube found that users of "Intellectual Dark Web" and "Alt-lite" channels migrated toward more extreme "Alt-right" content over time, and that Alt-lite material was easily reachable from IDW channels through the recommendation system itself. But a later, larger study by Homa Hosseinmardi, Duncan Watts, and colleagues, tracking the on- and off-platform browsing behavior of more than 300,000 Americans from 2016 to 2019, found that far-right content consumption on YouTube tracked users' pre-existing off-platform tastes closely, and reported "no evidence that engagement with far-right content" was caused by the recommender steering people there against their preferences.

Benischke and Post's framing absorbs this disagreement rather than being undermined by it. Whether the mechanism is an algorithm actively pushing users toward extremity or an algorithm efficiently matching existing appetites to maximize engagement, the lever that matters is the same one: a system built to allocate attention toward whatever sustains the most of it. "Revealed preference" is not an argument that governance doesn't matter, it is a description of exactly the kind of engagement-maximizing allocation rule that attention governance is supposed to put under scrutiny.

The DSA's Quiet Bet on the Upstream Lever

The European Union's Digital Services Act is the first major regulatory instrument to treat the ranking system itself, rather than individual pieces of content, as a governance object. Article 27 requires platforms to set out, "in plain and intelligible language, the main parameters used in their recommender systems, as well as any options for the recipients of the service to modify or influence those main parameters." Article 38 goes further for the largest platforms: in addition to the Article 27 disclosure, "very large online platforms" must offer "at least one option for each of their recommender systems which is not based on profiling." And Article 34's systemic risk assessment explicitly requires the largest platforms to evaluate how "the design of their recommender systems and any other relevant algorithmic system" contributes to risks like disinformation and civic harm, folding ranking design into the same risk category as content moderation, rather than treating it as a separate, unregulated engineering choice.

The DSA does not ask platforms what posts they removed. It asks them to explain the rules that decided what got amplified before anyone had to remove anything.

This is attention governance as regulatory text, not just scholarly framing, and it is genuinely new. Whether disclosure of "main parameters" translates into different amplification outcomes, or whether platforms satisfy the letter of Article 27 with boilerplate that leaves the underlying engagement-maximizing logic untouched, is the open empirical question the law has created and that researchers have barely begun to answer.

Governing Attention, Not Content

Benischke and Post close their synthesis by calling for a governance research agenda built around attention allocation rather than content classification. The practical translation for executives and policymakers is the same insight restated as a decision rule: every time a harm shows up as a content problem, a viral falsehood, a radicalizing recommendation chain, a feed that makes teenagers feel worse about their bodies, ask what ranking and monetization decision made that specific piece of content worth more attention than its neighbors. The Facebook Files show internal teams that already knew the answer and were overruled by engagement concerns. The YouTube research shows that even when the algorithm is less culpable than assumed, the right question is still about the allocation system, not the individual video. And the DSA shows regulators starting, haltingly, to write that question into law.

Content moderation will keep absorbing political attention because it is visible and fast. But as long as the ranking and monetization engine beneath it keeps rewarding whatever content sustains the most attention, moderation is pruning leaves on a tree whose roots are getting watered every day.