AI tell · grade B
The bold-header list: an AI formatting tell
The bold-header list is prose flattened into bullets, each led by a bold word and a colon. Every model we logged produced it, and one model alone hit dozens in a single pass. Readers flag it because it mistakes formatting for thinking: the bold labels imply a structure the content has not earned, and the shape recurs on subjects that call for sentences. The fix is to write the paragraph the list was standing in for, and reserve real bullets for genuine enumeration.
What it is
A vertical list where each item opens with a boldfaced label, a colon, and then a fragment. It looks organized, but it usually breaks a single argument into disconnected shards, each dressed up with a heading it does not need.
How often it shows up
All 12 of 12 models we logged produced bold colon lead-ins (emp-002), and one model (emp-025) hit 35 of them in a single response. Because the pattern is so consistent and so consistently unnecessary, the checklist grades it B.
Why readers flag it
A reader sees a page of bold labels and reads effort-signaling, not structure. Real prose connects its points; the bold-header list severs them and hopes the formatting carries the logic. The uniform shape, applied to everything, is the giveaway.
The fix
Ask whether the items form an argument or a genuine list. If they connect, write them as a paragraph with real transitions. If they are truly a set of parallel options, keep plain bullets and drop the bold colon labels.
Examples
**Speed:** It is faster. **Quality:** It is better. **Cost:** It is cheaper.
It runs faster than the old process, the output needs fewer corrections, and it costs less to operate, mostly because it drops a manual review step.
Questions
Are bullet points themselves an AI tell?
No. A genuine list of parallel items is fine. The tell is turning connected prose into bullets that each carry a bold colon label, which severs an argument into fragments.
underslop marks this tell and the rest of the consensus set in any draft you paste, using the same open lint that scores it here. The free tier caps at 500 words a run, and your text is held for 0 seconds.