Does higher cognitive load on social media increase engagement?
This reads the question as: does leaving people with something to do — a gap, a friction, an unresolved point — drive engagement more than handing them complete, effortless content? The corpus doesn't measure 'cognitive load' directly, but it has a sharp adjacent finding about the cost of removing it.
This explores whether friction — content that leaves the reader something to resolve — drives engagement, versus frictionless content that satisfies you and lets you scroll on. The corpus doesn't have a paper that measures cognitive load on a slider, but it converges hard on one surprising result: making content *easier* and *more complete* often suppresses the very engagement platforms chase. The clearest case is the informativeness paradox from Nextdoor's experiments Does better summary writing actually increase user engagement?: LLM-written notification summaries were objectively better and more informative, and precisely because of that, click-through dropped. When the summary already answers the question, there's no reason to open anything. The unmet need — the small cognitive gap — was doing the engagement work all along.
The same pattern shows up in how AI posts behave socially. Comprehensive, confident, fully-resolved posts rack up likes and visibility but kill the reply thread Why do AI posts get likes without inviting conversation?. A post that says everything invites no counter-argument and offers no opening to respond to — so you get one-directional recognition (a passive metric) without conversation (the effortful, high-engagement kind). This is the key wrinkle: 'engagement' isn't one thing. Frictionless completeness maximizes the cheap signals and starves the expensive ones.
That split matters because the expensive kind is what the platform actually runs on. AI-generated content displaces human voices precisely by being comprehensive enough to capture attention while building no reputation and provoking no exchange Does AI content displace human influencers on social media? — and the deeper threat is the loss of conversational *style*, the structure of genuine address that makes people want to talk back Does AI threaten social media's conversational function?. Remove the friction of a real interlocutor and you remove the reason to engage actively, even as the surface metrics hold up.
There's a cross-domain echo worth pulling in: more processing effort isn't monotonically good anywhere. In reasoning models, accuracy climbs then *falls* as you add thinking tokens — models overthink easy problems and the relationship inverts past a threshold Does more thinking time always improve reasoning accuracy?. So the honest answer to your question is an inverted-U, not a straight line: a little load (an unresolved gap, something to click into or argue with) drives engagement; too little gives you nothing to do and you leave; too much is just exhausting. The corpus's real contribution is naming the failure mode nobody optimizes for — content so good it gives you no reason to stay.
Sources 5 notes
Nextdoor experiments showed LLM-generated summaries were objectively more informative but decreased click-through rates. Users had no reason to open notifications when the summary already satisfied their information need, demonstrating how optimizing for informativeness can backfire on engagement metrics.
AI-generated posts achieve high engagement metrics through comprehensive, confident phrasing but suppress reply dynamics because they lack human authorship and invite no counter-argument. This creates one-sided recognition divorced from the conversational validation that historically legitimized social proof.
AI-generated posts capture engagement through comprehensiveness but accrue social proof without building any speaker's sustained reputation. This displacement compounds over time, eroding the platform's core function of promoting legitimate human voices while monetization continues.
AI-generated posts drain social media's function as a conversational medium because they lack the structure of genuine address and mutual orientation. This threat operates below the level where content moderation, fact-checking, and recommender adjustment can reach.
Increasing thinking tokens from ~1,100 to ~16K reduced benchmark accuracy from 87.3% to 70.3%, revealing a non-monotonic relationship where models overthink easy problems and underthink hard ones.