Bold-keyword spam
Bolding random nouns and verbs throughout body text. Emphasis applied uniformly conveys nothing.
What it is
Prose where seemingly arbitrary words are bolded. Not key terms being defined, not section headings, just words the model decided 'feel important'. Reading becomes choppy because the eye is constantly pulled to false signals.
Why models do it (first principles)
Markdown-formatted responses scored higher in preference data. The model learned 'bold = good' and over-applies it. Because there is no internal model of 'what should the reader scan for', bolding becomes uniform decoration rather than navigation.
How to think about it
Bold exists to direct attention to terms the reader needs to find again or remember. Applied uniformly, it conveys nothing. The same way text shouted everywhere conveys nothing. The model performs emphasis without modeling what emphasis is for: to mark what matters relative to what doesn't.
Examples
**React** uses a **virtual DOM** to make **updates** fast and **efficient** for **modern** web apps.
React uses a virtual DOM to make updates fast — useful for apps where the UI changes a lot.
Fix prompt
Emphasis is a contrast effect. It works only against an unemphasized background. Bolding to decorate rather than to direct attention erases the contrast and turns the visual cue into noise; the reader's eye is pulled to terms that don't repay the pull, and the actually-important terms become invisible. Use bold only when you would want the reader to find that exact phrase again later.
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.
- **bolded nouns** scattered through body prose
- first occurrence of every term bolded
- bolded phrases that aren't being defined or referenced again
- whole sentences in bold