The components: formality (casual ↔ formal), warmth (clinical ↔ personal), authority (modest ↔ confident), specificity (vague ↔ named), pace (verbose ↔ punchy). Each axis is a deliberate choice; combining them produces the voice fingerprint.
Common brand voice failures: copying competitor voice (every B2B SaaS sounding like the same Sequoia-funded startup), shifting voice mid-page (founder copy in the hero, agency copy in the body), or claiming a voice the brand hasn't earned (a stage-0 startup writing like a late-stage incumbent).
Strong brand voice is enforced at the editing layer. Every paragraph passes a "could this sentence appear on a competitor's page" test. If yes, it gets rewritten until it sounds like it could only have come from this brand.
How Lytms scores it
Lytms extracts a voice profile from the brand's scanned pages (founder LinkedIn, brand landing page, brand X account) and grades every generated draft against it. The variant engine enforces signature phrases + banned phrases + length patterns from the materialized voice profile.
See also
See brand voice on your own page.
One URL. About 2 minutes.