Brand voice, banned words, and rubric weights

• Voice doc— free-form description of your brand's tone, audience, do's and don'ts. Up to 1000 characters. Read by the analyzer before every Quick or Deep Bench.
• Banned words / phrases— up to 50 entries. Any match in the analyzed ad is flagged as a high-severity issue under "Brand violations."
• Compliance posture — standard / strict / FTC / TikTok Shop. Tilts how aggressively the analyzer flags regulatory and platform-policy issues.
• Rubric weights — six dimensions (hook, clarity, CTA, brand fit, native feel, pacing) on a 0–200 scale. 100 = normal weight, 200 = double emphasis, 0 = ignore.
• Default vertical— optional override so the analyzer doesn't re-infer "beauty" or "saas" on every run.

The prompt-injection budget is 1000 characters. Spend them on the things the analyzer can actually score against, not on aspirational positioning.

Good voice-doc content (the analyzer will use these to grade brand-fit + native-feel + every rewrite suggestion):

• Audience age + posture ("18–28 lifters, suspicious of supplements")
• Tone rules ("irreverent, no corporate speak, no 'we'")
• Format defaults ("UGC always, no studio polish")
• Specific lines or moves the brand owns ("always end with 'Stay strong'")
• Things the brand never does ("never references competitors by name")

Skip mission-statement language, abstract values, and anything that doesn't translate into "does this ad do X or not."

The banned-word list is a brute-force regex filter the analyzer applies before scoring. Any match is flagged as a high-severity issue with the matching quote called out. Use it for:

• Legal-restricted terms ("cure," "guaranteed," "FDA-approved" for supplements)
• Off-brand language ("literally," "crushing it" if your voice is corporate)
• Competitor names you'd rather not mention
• Internal codenames or working titles that shouldn't leak into copy

A flagged ad still runs — the issue lands in the Issues panel alongside whatever else the analyzer caught. Used aggressively, this is also a good QA gate for UGC submissions: pair with the Voice Match workbench at /draft/voice-match.

• Standard — flag obvious issues only. The default for most teams.
• Strict — flag any misleading claim, unsubstantiated stat, or dark-pattern CTA. Useful for regulated verticals (finance, health) or for QA passes where you want maximum scrutiny.
• FTC — enforce endorsement-guide requirements: visible disclosures, ambiguous affiliate signals, deceptive testimonials. Mandatory if you run paid creator content.
• TikTok Shop — platform policy: missing product-tag disclosures, prohibited health claims, fake urgency, price-guarantee policy violations.

Each scoring dimension can be weighted 0 (ignore) to 200 (double emphasis). Six named presets give sensible defaults — load one as a starting point, then nudge.

• Ecommerce / DTC — hook, clarity, CTA all heavy. Brand-fit lighter (DTC ads cap at mid-tier brand chrome).
• SaaS / B2B — clarity heaviest. Native-feel and pacing dialed back (a clear demo can beat a pattern-interrupt opener for B2B).
• UGC / creator-organic — native-feel maxed (200), pacing high. Brand-fit lighter (the ad is supposed to feel like a person, not a brand).
• Beauty / fashion — hook and brand-fit both heavy. Native-feel high.
• Affiliate / TikTok Shop — CTA maxed (180). Brand-fit lighter (affiliate creators don't care about your chrome).

Custom-tune from there. The injection only fires when weights deviate from default — if you stay on Balanced, no extra prompt tokens are spent.

/account/team/brand-kit — requires an Agency-tier team (5 included seats, $25/seat after). Owner-only edits; team members see a read-only view.

The Brand Kit applies to every analysis run by any member of the team. It also feeds the Voice Match workbench, which scores UGC submissions 0–100 against your kit before they ship.