How AI Responses Work

The AI behind ReviewAI

ReviewAI uses Claude by Anthropic to generate review responses. When a new review is detected, the AI reads the full review text along with your business context (tone, selling points) and generates a personalized response.

How responses are generated

1. Review detected — Our worker checks for new reviews every 30 minutes.
2. Context assembled — The review text, star rating, your business tone, and selling points are combined.
3. AI generates response — Claude creates a draft response (max 300 tokens, ~2-4 sentences).
4. Safety checks — The response is checked for prompt injection and inappropriate content.
5. Routing — Positive reviews auto-post (if enabled) or go to your approval queue. Negative reviews always go to the queue.

What makes responses good

The AI references specific details from each review — dish names, staff mentions, specific compliments or complaints. This makes responses feel personal rather than templated. The AI also matches the tone you set for your business.

Safety and guardrails

• Responses never offer discounts, refunds, or specific remedies.
• Prompt injection detection flags suspicious review content.
• If the AI fails to generate a response (timeout, error), the review is marked for manual handling.
• The AI retries up to 2 times on failure with a 30-second timeout per attempt.

Customizing tone

Each business has a tone field where you describe how responses should sound. Examples: “Professional and warm, like a family-owned restaurant” or “Concise and clinical, appropriate for a medical practice.” The AI adapts accordingly.