How to Fix n8n Webhook Not Delivering Messages to the LLM Node

Send a test request to your webhook and open the execution in n8n. Click on the Webhook node and examine the output panel. Switch to JSON view to see the raw data structure. Pay attention to how the request body is nested. Depending on your Webhook node configuration, the body might be at $json.body, directly at $json, or nested under $json.body.data. The HTTP Method setting and the Response Mode setting also affect the output structure. If the webhook is set to receive form data vs JSON, the output structure differs. Note the exact field names and paths that contain the user's message and any other data you need to pass to the LLM node.

1// Common Webhook output structures:
2
3// JSON body (Content-Type: application/json):
4// $json.body.message, $json.body.userId, etc.
5
6// When 'Raw Body' is disabled (default):
7// $json = { headers: {...}, params: {...}, query: {...}, body: {...} }
8
9// When 'Raw Body' is enabled:
10// $json = { body: "raw string", headers: {...} }
11
12// Form data (Content-Type: application/x-www-form-urlencoded):
13// $json.body.field1, $json.body.field2

Click on the LLM node (AI Agent, Basic LLM Chain, etc.) and examine what input fields it expects. The AI Agent node has a Prompt field that takes the user message. The Basic LLM Chain node has a Prompt field as well. If you are using a sub-node configuration, the AI Agent expects the input to contain a chatInput or similar field in the incoming data. Open the node settings and check each input field to see what expression is used. If the field contains an expression like {{ $json.chatInput }} but the webhook outputs the message at $json.body.message, there is a mismatch. The expression will resolve to undefined and the LLM will receive no prompt, producing empty or generic output.

Insert a Code node between the Webhook and the LLM node that transforms the webhook output into the format the LLM node expects. This is the most reliable fix because it explicitly maps fields regardless of how the webhook structures its output. Set the Code node to Run Once for All Items. Extract the user message from wherever it exists in the webhook data and output it as the field the LLM node expects. Also pass through any other fields needed downstream, like session IDs or metadata. This transformation layer decouples the webhook format from the LLM format, making the workflow resilient to changes on either side.

1// Code node: Webhook to LLM Data Mapper
2// Mode: Run Once for All Items
3
4const items = $input.all();
5const results = [];
6
7for (const item of items) {
8  const data = item.json;
9  const body = data.body || data;
10
11  // Extract user message from common locations
12  const userMessage = body.message
13    || body.text
14    || body.content
15    || body.prompt
16    || body.query
17    || body.input
18    || '';
19
20  if (!userMessage) {
21    console.log('WARNING: No user message found. Available keys:', Object.keys(body));
22  }
23
24  // Output in the format the AI Agent expects
25  results.push({
26    json: {
27      chatInput: userMessage,
28      sessionId: body.sessionId || body.userId || 'default',
29      metadata: {
30        source: 'webhook',
31        timestamp: new Date().toISOString(),
32        ip: (data.headers || {})['x-forwarded-for'] || 'unknown'
33      }
34    }
35  });
36}
37
38return results;

Check that nodes are connected with the correct output-input mapping. In n8n, each node connection carries all output items to the next node's input. However, some nodes have multiple outputs (like the IF node or Switch node), and connecting the wrong output port means no data flows through. Click on the connection line between nodes to verify it is the correct output. Also verify that the LLM node is not configured to receive input from a different source. In the AI Agent node, the main input connector expects the user prompt. If you accidentally connected to a tool input or memory input, the user message will not reach the agent's prompt. Right-click the connection and select Delete, then re-connect from the correct output to the correct input.

Use n8n's data pinning feature to isolate and test each section of the pipeline. First, send a test webhook request and run the workflow. Then pin the output of the Webhook node by clicking the pin icon. Now you can run just the downstream nodes without needing to send another webhook request. Pin the Code node's output too, then test the LLM node in isolation. This step-by-step approach reveals exactly where data gets lost or transformed incorrectly. Unpinning automatically when you need fresh data is done by clicking the pin icon again.

Not all webhook requests will have a properly formatted body. Add validation in your data mapping Code node to handle missing fields, empty bodies, wrong content types, and unexpected data formats. When validation fails, return a clear error message via the Respond to Webhook node instead of passing empty data to the LLM. This prevents wasted LLM API calls and gives the webhook caller actionable feedback about what went wrong with their request.

1// Code node: Validate Webhook Input
2const items = $input.all();
3
4for (const item of items) {
5  const body = item.json.body || item.json;
6  const message = body.message || body.text || body.content || '';
7
8  if (!message || message.trim() === '') {
9    // Return error response
10    item.json.isValid = false;
11    item.json.error = 'Missing required field: message. Send a JSON body with a "message" field.';
12    item.json.chatInput = '';
13  } else if (message.length > 10000) {
14    item.json.isValid = false;
15    item.json.error = 'Message too long. Maximum 10,000 characters.';
16    item.json.chatInput = '';
17  } else {
18    item.json.isValid = true;
19    item.json.error = null;
20    item.json.chatInput = message.trim();
21  }
22}
23
24return items;
25
26// Follow with IF node: {{ $json.isValid }} === true
27// True → AI Agent node
28// False → Respond to Webhook with error