Does high knowledge density in text reduce user motivation to read more?
This explores whether packing more knowledge into fewer words (high knowledge density) actually discourages readers from going deeper — and the corpus suggests the relationship runs through engagement and trust mechanics, not reading effort per se.
This explores whether dense, efficient text dampens the urge to keep reading. The most direct evidence flips the question's framing: the corpus's clearest finding is that satisfying a reader's information need is exactly what kills the motivation to read more. In a real-world experiment, AI-generated summaries that were objectively *more* informative drove *down* click-through — people had no reason to open the notification once the summary already told them what they needed Does better summary writing actually increase user engagement?. So high density doesn't reduce motivation by exhausting the reader; it reduces it by closing the loop. The curiosity gap is the fuel, and density spends it.
But the corpus also lets us measure density itself. Knowledge Density — unique atomic knowledge units per token — turns 'reading efficiency' into a number, and notably finds that LLM writing scores *lower* than human writing because models pad and elaborate, inflating token count while holding actual content flat Can we measure reading efficiency as a quality metric?. That reframes the question: most machine-generated text isn't suffering from too much density, it's suffering from too little. The motivation problem and the density problem may point in opposite directions.
There's a second, sneakier channel. What readers respond to often isn't the knowledge at all but its surface signals. Users prefer answers with more citations even when those citations are irrelevant — citation count works as a decoupled trust heuristic, not a measure of substance Do users trust citations more when there are simply more of them?. And LLMs persuade in nearly every exchange by leaning on logical and quantitative framing, which makes their output feel authoritative regardless of whether the density is real Do LLMs persuade users more often than humans do?. So 'dense-feeling' text can hold attention through performance of rigor, while genuinely dense text can lose readers by leaving nothing to chase.
The constraint that actually limits how much people can absorb is cognitive, not motivational: reasoning accuracy drops sharply as input grows, falling from 92% to 68% with just a few thousand tokens of padding — well below any context limit, and unhelped by chain-of-thought Does reasoning ability actually degrade with longer inputs?. If dense text is also long text, comprehension degrades before motivation does. One promising counter-move is to give readers a map first: summarizing a document globally before diving in helps connect scattered evidence, suggesting structure — not minimal length — is what keeps long, dense material navigable Can building a document map first improve retrieval over long texts?.
The thing you didn't know you wanted to know: the engagement-killer isn't density, it's *completeness*. Text that fully answers leaves nowhere to go. The design lever isn't 'write less densely' but 'leave a deliberate edge of the unknown' — which is precisely what an Inquiring Line is built to do.
Sources 6 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.
Knowledge Density (KD) operationalizes reading efficiency by dividing unique atomic knowledge units by text length. LLM-generated text scores lower on KD than human writing because retrieval redundancy and the model's tendency to elaborate inflate token count while holding knowledge content constant.
Analysis of 24,000 Search Arena interactions shows irrelevant citations boost user preference (β=0.273) nearly as much as relevant citations (β=0.285), indicating citation count functions as a decoupled trust heuristic.
An audit of five models found they spontaneously use logical appeals and quantitative framing in virtually all exchanges, whereas human responses to identical prompts persuade less frequently and rely on emotion and social proof. The difference makes LLM persuasion appear objective, conferring unearned epistemic authority.
FLenQA shows reasoning accuracy drops from 92% to 68% at just 3000 tokens of padding, far below context window capacity. The degradation is task-agnostic, uncorrelated with language modeling performance, and persists even with chain-of-thought prompting.
MiA-RAG inverts standard RAG by summarizing documents first, then conditioning retrieval on that global view. This approach recovers discourse structure that bag-of-chunks retrieval destroys, making scattered evidence findable by their document role rather than surface similarity alone.