Brand Voice Critic
Customer-facingScores drafts generated from a journal entry against your brand voice, and refuses to publish one that failed.
What it does
When a draft is generated from a journal entry it is scored against your brand voice profile, checking for tone drift, banned phrases and AI tells and returning a score plus a list of fixes. A draft below threshold is revised once automatically and the better of the two is kept. The verdict is stored on the post, and publishing a post whose verdict was a failure is refused. Content written any other way — pasted in, imported, or written through the API by an agent — is not scored today, so nothing blocks it.
Customize this agent
You can layer additional instructions on top of this agent's built-in system prompt. Your text is appended below the base prompt — it stacks, it doesn't replace. Use it to tell the agent what to emphasise or avoid for your site specifically.
Open Agent settings →Triggers
- →A draft is generated from a journal entry
Tools it can touch
- •LLM scorer (gpt-5-mini)
- •Mechanical AI-tell and banned-phrase detectors
- •Brand voice profile (per-site, stored in DB)
Features it powers
Source
- src/lib/brand-voice-critic.ts