Brand extraction: how an engine wears your prospect's colors safely

Of all the slots a personalized page can fill, brand is the one prospects mention first. Copy personalization gets processed consciously; color gets processed before reading begins. A page in the prospect's palette produces a small, useful double-take — 'wait, is this ours?' — that buys the headline a genuine read.

What extraction actually pulls

• Palette: dominant and accent colors from the live site's CSS and rendered pixels, clustered and ranked.
• Type feel: serif vs. sans, weight, the general typographic temperature (an engine matches feel, not exact licensed fonts).
• Imagery temperature: photographic vs. illustrated, dark vs. light mode preferences.
• Tone markers: how the company writes about itself, which calibrates the generated copy's register.

The guardrails that matter

1. Confidence gating: when extraction is ambiguous (thin sites, conflicting palettes), fall back to neutral styling rather than guessing wrong — an off-brand page is worse than an unbranded one.
2. Identity honesty: the page references the prospect ('prepared for Acme') and may echo their palette; it must never present itself as being from the prospect. Sender identity stays unmistakably yours.
3. Logo discipline: showing a prospect's name in your headline is nominative use; rebuilding your page as their property is impersonation. Engines should enforce the line structurally.
4. Review surface: every extracted brand should be human-inspectable before a batch ships.

These rules aren't just etiquette — they're the difference between reciprocity ('they did homework on us') and alarm ('this is pretending to be us'). The flattery effect lives entirely on the right side of that line. Crutan's engine treats the line as a hard constraint, which is also why our own acceptable-use policy bans impersonation outright.