How to fix n8n webhook not delivering messages to the LLM node?

The webhook usually doesn’t deliver data to an LLM node in n8n because the Webhook node isn’t set to respond immediately, or the incoming JSON isn’t where the LLM node expects it, or the workflow runs in production mode but you're testing with the wrong webhook URL. In most real cases, fixing the webhook’s response mode, checking the incoming payload with a Set or Function node, and confirming the execution path going into the LLM node resolves it.

 

Root causes and how to fix them

 

Below are the reasons that consistently break real webhook-to-LLM flows, and exactly what to change.

• The webhook is still in "Test" mode. In n8n, every Webhook node has two URLs: a test URL (works only when the workflow is open in the editor and "waiting") and a production URL. If you're hitting the test URL while the workflow is closed, nothing will run. Use the production URL when running normally.
• The Webhook node is set to "Respond to Webhook" and the workflow pauses before the LLM node. When "Respond Immediately" is not chosen, n8n waits for the workflow to finish before answering the webhook. If your LLM node takes a long time, the requester may time out, and n8n stops the flow. Fix: In the Webhook node → Response section → choose "Respond immediately".
• The incoming JSON structure doesn't match what the LLM node expects. LLM nodes typically expect a text input field like:
  {{$json.body.message}} or {{$json.text}} If your webhook sends {"query":"Hello"} but the LLM node reads {{$json.message}}, it will send literally nothing to the LLM. Fix: Use a Set node or Function node right after the Webhook to normalize the input fields.
• The Webhook node uses POST but your service sends GET (or vice versa). The Webhook node is strict about HTTP method. If the incoming method doesn’t match, the workflow does not start at all. Fix: Ensure Webhook node → HTTP Method matches the sender (often POST).
• LLM node has an expression error. If you see “No data was received” inside the LLM node, hover the expression. If it shows “undefined”, that means the JSON path is wrong. Fix: Use the "JSON" view in the Webhook node → Output panel to copy the correct paths.
• Binary vs JSON confusion. Some integrations send binary payloads. LLM nodes only accept text/JSON. Fix: Ensure the Webhook node → "Binary Data" toggle is OFF (unless needed), or convert binary to text with a Function node.
• Missing credentials or model not configured. The workflow may run, but the LLM node fails silently in preview mode. Fix: Open the LLM node and ensure the Credentials and Model fields are properly selected.

 

Practical step-by-step fix

 

Here is a clean, production-safe setup:

• Webhook node → set to POST, Production URL, Response Mode → Respond Immediately.
• Add a Set node right after the Webhook with something simple like:
  text: {{$json.query || $json.message || $json.body.message}} This normalizes the input so the LLM node always has a clean "text" field.
• Feed LLM node with:
  {{$json.text}}
• Optional: Add an Error Workflow so that if the LLM node times out, you still get diagnostics.

 

Example minimal working flow

 

// This Function node simply maps the incoming webhook payload
// into a single field "text" that the LLM node can safely consume

return [
  {
    json: {
      text: $json.query || $json.message || $json.body?.message || "No input"
    }
  }
];

 

After that, in the LLM node → Prompt field, use:

{{$json.text}}

 

Why this works

 

The key is that n8n passes data from node to node as JSON, and every node only sees what the previous one outputs. Webhooks often produce deeply nested or inconsistent JSON. LLM nodes require a clean string input. Ensuring the Webhook responds immediately, the JSON is cleaned up, and the model credentials are valid eliminates 95% of the real-world issues with webhook → LLM delivery.