brand-voice

Gather 5 to 20 real writing samples such as posts, essays, or emails, then extract patterns for rhythm, compression, capitalization, and claim style into a structured voice profile. Reuse that profile across all downstream content instead of re-deriving style each time.

Define explicit banned moves in your voice profile, such as fake curiosity hooks, bait questions, "not X, just Y" phrasing, and LinkedIn thought-leader cadence. Ground every rule in observable patterns from real source material rather than generic platform exemplars.

Prioritize recent original posts and threads first, then articles, essays, and newsletters, followed by real outbound emails that worked, and finally product docs and site copy. Prefer recent material unless older writing is more canonical.

Yes, the profile includes channel notes for X, LinkedIn, and email so the same core style adapts per platform. If the source set splits between public launch voice and private working voice, the profile captures both instead of averaging them.

The profile quality depends entirely on the source set; thin or conflicting sources produce low-confidence results. It captures observable style patterns but cannot invent a voice that does not exist in the provided material.