A city legal-aide chatbot drafts plain-language answers residents might treat as official counsel. What content-boundary design fits?

Legal-aide chatbots that paraphrase municipal rules can be mistaken for counsel if the product voice sounds official. Marking generated text as draft-only and not official legal advice works because clear boundary language reduces reliance risk, sets expectations that residents may need a human attorney or formal city channel, and aligns the assistant with appropriate-use policy. Cities can still offer plain-language help while refusing the posture of binding advice. Stamping drafts with the city seal to signal attorney review and approval fails because it implies official endorsement of AI output, blurring the draft-only boundary and inviting residents to rely on it as counsel. Disclosing AI origin only in the help menu while keeping answers authoritative fails because it buries the limitation where residents may not see it and still presents the output as reliable counsel. Shortening the disclaimer to a link labeled terms for cleaner mobile display fails because it tucks the material use limitation behind a tap, making the draft-only boundary easy to miss. Exam caveat: disclaimers do not license inaccurate law; content quality, citation to published codes, and escalation paths still matter, and some topics may require refusing advice entirely. Operational check: place persistent draft-not-counsel markers in UI and model instructions, link to official counsel or legal aid resources, and test resident-facing pages to confirm the disclaimer remains visible on mobile answer reveals.