| name | humanize-ai-prose |
| description | Rewrite AI-generated text into genuinely human prose with seven ordered transform passes: add real named specifics, fix the structure, kill the tells (em-dashes, intensifiers, negative parallelism, meta-signposts), admit real limits, and read it aloud. |
Humanize AI Prose
A reusable, ordered editing system that turns model-generated prose into writing that
reads like an actual person wrote it. It's two layers: seven mechanical transform
passes (each tell is detectable and rewritable), then a taste checklist of judgment
calls no find-and-replace can make.
Why ordering matters
The passes run truth → structure → paragraph → sentence → word → friction → ear, and
the order is load-bearing. Inserting one real specific reshapes everything downstream (it
forces you to delete the vague sentence that was standing in for it), so specifics come
first. Word-level fixes come late, because there's no point polishing a sentence you're
about to cut. Read-aloud is last, after the text has stopped changing. Run them in order;
don't jump to the word pass because it feels productive.
The seven passes, in order
1. TRUTH: insert real specifics. AI prose names categories, never instances ("various
campuses," "a meaningful experience," "a typical user"). Give every section at least one
true, named, datable detail: a first name, a place, a month-and-year, a product, a number
that isn't round. This is the highest-leverage move in the whole system. If you don't know
the fact, do not invent it. Use the bracket rule below.
2. STRUCTURE: de-signpost, fix intro and ending. Delete sentences whose only job is
pointing at other sentences ("Here's the part that surprised me," "In this article we
will," "That's the whole flaw, stated cleanly"). Humans just say the thing. Rewrite the
intro and conclusion completely rather than patching them: they're the most
template-shaped parts of any draft. The ending should add something (an implication, an
open question, a last concrete image), never re-list the points. Cap dramatic one-line
closers ("mic drops") at two per piece; one is voice, five is a metronome.
