AI Answer Checker n8n Workflow: Google Search, GPT-4 Alerts

Catch AI Q&A hallucinations before they reach users—automatically

The AI Answer Checker n8n Workflow receives an AI-generated question and answer, searches the web for supporting evidence, and then uses OpenAI (GPT-4 Turbo) to classify the response as SAFE, SUSPICIOUS, or HALLUCINATED. When risk is detected, it alerts your team via Slack or sends a critical Gmail message—while returning the verdict back to the webhook caller.

What this workflow does

• Webhook intake: Exposes an n8n webhook endpoint that accepts a question and answer via POST at /hallucination-check.
• Google search sourcing: Calls your Google Search API endpoint to retrieve the top search results for the question.
• Compact source list: Formats the top search snippets into a compact list of sources to support validation.
• GPT-4 verification: Sends the question, answer, and sources to OpenAI (GPT-4 Turbo) to produce a JSON verdict with a confidence score, reasoning, and classification.
• Decision routing: Returns the result to the webhook caller (especially for SAFE outcomes).
• Alerts: If SUSPICIOUS, posts a message to your chosen Slack channel; if HALLUCINATED, sends a critical Gmail alert email.

Use cases

• Validate answers in an AI support assistant before publishing to customers.
• Reduce risk in a SaaS knowledge bot by detecting fabricated claims.
• Quality-control internal AI workflows for compliance and trust.

Technical details

• Nodes/stack: webhook, set, code, OpenAI (GPT-4 Turbo), switch, slack, gmail.
• Search integration: Replace the placeholder Google/Serp Search API URL and X-API-KEY header with your provider details.
• Routing logic: switch node routes by classification (SAFE, SUSPICIOUS, HALLUCINATED).