Reflexive bullet-listing
Turning continuous reasoning into disconnected bullets. Visible structure substituted for actual structure.
What it is
Prose that has internal logical flow gets chopped into disconnected bullets, losing the connective reasoning. Each bullet becomes a fragment; the relationships between them become invisible.
Why models do it (first principles)
Bullet-heavy responses score well in preference comparisons because they look 'organized' at a glance and are easy to skim. The model learned that markdown structure is a reliable reward signal. Once it commits to a list, every sub-thought becomes its own bullet because that's what it has seen rewarded.
How to think about it
Bullets are a tool for content that is genuinely list-shaped: parallel items, no flow between them. Applied to argumentative or narrative prose, they fragment the reasoning into atoms and force the reader to reconstruct the connections. The model performs 'organized' (visible structure) instead of being organized (clear logical flow). It optimizes for the look of clarity rather than clarity itself.
Examples
- React uses a virtual DOM - This makes updates fast - Because diffing is cheaper than re-rendering - Which is why it became popular
React uses a virtual DOM, which makes updates fast: diffing a JS tree is cheaper than re-rendering the real DOM. That performance is a big part of why it caught on.
Fix prompt
Match the shape of the format to the shape of the content. Bullets fit genuinely parallel items with no logic running between them; argument and narrative carry their meaning in the connectives, and stripping those connectives out doesn't make the reasoning clearer, it makes it the reader's job to reconstruct. Visible structure is not the same as actual structure. Choose the form that matches the thought.
Watch for
Concrete phrasings this pattern usually shows up as. These are not part of the copyable prompt. The prompt teaches the principle so the model can recognize the move even when the exact phrasing differs. Use this list to self-audit your own writing or to test a model.
- bulleted prose with conjunctions implied between items
- single-sentence bullets that depend on each other in sequence
- bullets used for cause-and-effect chains
- lists where removing the bullets and adding 'because' / 'so' / 'and' would read better