The compliance scanner

What the scanner does

Every draft — whether AI-generated or human-edited — runs through the compliance scanner before it can move to Approved. The scanner is fast (sub-second on typical drafts) and runs:

1. At generation time. New drafts are scanned the moment they're produced.
2. At save time. Any edit triggers a re-scan.
3. At approval time. A final scan blocks Approved if anything failed.

Rule sources

The scanner combines four rule sources:

Source	Examples
Global rules	Unsupported "best in the world" claims, fabricated statistics, PII in body text
Industry pack	Pack-specific prohibited phrases (e.g. medical "miracle cure")
Workspace custom rules	Your team's banned phrases and required disclaimers
Compliance pack updates	Quarterly updates to global rules based on regulator guidance

What blocks vs. warns

Each rule has a severity:

• Block — the draft cannot move past in-review.
• Warn — the draft can be approved, but the reviewer sees the warning and signs off.
• Note — informational; visible in the review pane but not enforced.

Workspace admins can adjust workspace custom rules' severity. Global and pack rules are fixed.

What gets flagged (a representative sample)

Pattern	Severity
"Guaranteed [outcome]" without backing	Block
"Best in [category]" without source	Warn
Numbers like "47% growth" without a citation	Warn
"FDA approved" outside medical pack with verification	Block
Email addresses or phone numbers in body	Warn
Sensitive PII patterns (SSN, credit card)	Block
Disparaging competitor claims	Block
Misattributed quotes	Block

Reviewer experience

When a draft has flags, the review pane shows:

• A red or amber banner at the top.
• The exact span(s) highlighted.
• A one-line explanation of the rule.
• Action buttons: Auto-revise, Edit manually, Override (for warns only).

Auto-revise on compliance failure

Clicking Auto-revise treats every flag as a constraint and regenerates the draft. The new draft re-runs through the scanner. Most issues clear on the first revision; persistent issues usually mean the brief itself is asking for something the scanner won't accept (e.g. an unsupported superlative the source data doesn't justify).

Overriding a warn

For warn-severity flags, the reviewer can override with a justification. The override is audit-logged with:

• The draft id.
• The rule name.
• The flagged span.
• The reviewer's justification.

Overrides do not weaken the rule for future drafts.

Adding workspace custom rules

Settings → Compliance → Custom rules:

• Phrase rule — regex or exact-phrase match.
• Required disclaimer — every draft of a given content type must contain a specific block.
• Banned topic — keyword cluster the engine and scanner both avoid.

Custom rules apply to your workspace only. They take effect immediately on save.

What the scanner does NOT do

• Fact-check claims against the public web. That's a different system (research mode); the scanner only enforces patterns and configured rules.
• Detect plagiarism. Use a plagiarism tool if that's a concern. The content engine writes original text, but the scanner doesn't certify originality.
• Translate or adapt for non-English locales by default. Multi-locale support is on the roadmap.

Performance

Scans run in roughly 200–500ms per page of content. The cost is included in your plan — there's no per-scan charge.

False positives

If a flag is wrong (the regex matched something benign), open Help → Report compliance false positive. The submission goes to our compliance team and we tune the rule. Workspace overrides are still allowed for the specific draft in the meantime.

See also

• Regulated industries — pack-specific overlays.
• Industry packs — what the pack-level rules look like.
• Compliance guardrails — the original overview.