The first signal of immediate danger (current threat, recent assault, child in the room, weapon present) triggers the safety-routing message: 911 reminder plus the National Domestic Violence Hotline at 1-800-799-SAFE (7233), available 24/7 with trained advocates. The bot does not push intake questions until the visitor confirms they are safe and chooses to continue. The system prompt enforces this order: safety first, intake second, never legal advice.

No. The system prompt declines opinions on evidentiary strategy, whether to call police, what to put in the affidavit, or the merits of the visitor's situation. The bot describes practice areas, books consultations, and captures structural intake only. Legal-advice refusal language is enumerated in the prompt and audited with adversarial test prompts every quarter. Bar associations treat unauthorised practice of law strictly; domestic-violence intake is one of the most sensitive contexts for getting this right.

Conversation logs live in the firm's WordPress database, not on the visitor's device, but messages remain in the browser session until the visitor closes it. The widget includes a "leave this page" link that exits to a neutral site and a one-click "delete this conversation" button. The bot reminds visitors of these controls at the start of the chat. For visitors on monitored devices, the bot can suggest calling the hotline from a different phone instead.

Yes. The bot captures contact preference (phone, encrypted email, in-office only) and a safe-to-call window. The booked consultation flows to the attorney with the preference flagged so subsequent contact respects the visitor's safety constraints. Phone consultations from a discreet number are common in domestic-violence practice and the intake supports that path explicitly. Visitors who cannot speak freely can also be redirected to the hotline for immediate confidential support.

Yes. GPT-4o and Claude both handle Spanish, Mandarin, Vietnamese, Tagalog, Arabic, and most common languages natively. Domestic-violence practices serve diverse communities and the bot replies in the language the visitor writes in. The safety-routing message includes the hotline number in the visitor's language where the hotline offers it. The booked consultation flags preferred language so the attorney prepares accordingly, and translation does not soften the safety-first or no-legal-advice guardrails.

BYO API key (OpenAI, Anthropic, Google, OpenRouter) with the standard terms means data is not used for training by default. Conversation logs live in your WordPress database and retention is your choice. For domestic-violence intake the typical posture is very short retention (24-48 hours) with structured handoff to a secure case-management system once an engagement starts. The widget can offer one-click deletion at any point. Document the bot in your privacy notice and align retention with the engagement-letter terms.

It can describe the firm's practice in custody modifications and divorce matters where abuse is a factor, and route to the right attorney. The bot declines to opine on whether evidence is sufficient, what the court will do, or how to handle a specific incident. Those are attorney calls. The intake captures jurisdiction and any current protective-order status so the attorney walks into the consultation with the structural context already in hand.

Configure the bot to ask clarifying questions early so the conversation routes to the right practice area. Firms that represent only victims should configure the bot to politely decline accused-party inquiries and refer to a defense firm or local bar referral service. Firms that handle both should scope multibot per practice with display conditions, so victim-facing pages run the safety-first bot and defense-facing pages run a separate intake. The bot must never blur this distinction.