How to Debug Why Model Responses Are Not Reaching the Next Node in n8n

The first step is confirming that the LLM node actually produced output. Click on the LLM node after a test execution and look at the Output tab. Switch to JSON view (not Table view) to see the raw data structure. If the output panel is empty, the LLM call itself failed or returned nothing. If the output panel shows data, the issue is in how downstream nodes reference it. Pay attention to the exact property names and nesting — the response text might be under .text, .content, .message.content, or .output depending on which LLM node you use.

The most common cause of missing data is a mismatched JSON path. Different LLM nodes in n8n output data under different property names. The OpenAI Chat Model node outputs under a different structure than the HTTP Request node calling OpenAI directly. Click on the downstream node that appears to receive no data, find the expression field that should reference the LLM output, and check that the path matches exactly. Use the expression editor's autocomplete feature — click the curly brace icon and browse the available fields from the previous node.

1// Common output paths for different LLM setups:
2
3// AI Agent node or LLM Chain node:
4{{ $json.output }}
5// or
6{{ $json.text }}
7
8// OpenAI Chat Model connected via AI Agent:
9{{ $json.output }}
10
11// HTTP Request node calling OpenAI API directly:
12{{ $json.choices[0].message.content }}
13
14// HTTP Request node calling Anthropic API directly:
15{{ $json.content[0].text }}
16
17// Basic LLM Chain node:
18{{ $json.response.text }}

Insert a Code node between the LLM node and the downstream node to log the exact data structure being passed. This node outputs the full item structure so you can see every available property. After running the workflow, inspect this debug node's output to find the correct path to the response text. Once you identify the right path, update the downstream node's expression and remove the debug node.

1// Debug Code node — Run Once for All Items
2const items = $input.all();
3const debugOutput = [];
4
5for (const item of items) {
6  debugOutput.push({
7    json: {
8      _debug_keys: Object.keys(item.json),
9      _debug_full_data: item.json,
10      _debug_type: typeof item.json,
11      _debug_has_output: 'output' in item.json,
12      _debug_has_text: 'text' in item.json,
13      _debug_has_content: 'content' in item.json,
14      _debug_has_choices: 'choices' in item.json,
15      _debug_has_message: 'message' in item.json
16    }
17  });
18}
19
20return debugOutput;

If the LLM node shows output but the response text is empty, the model may have returned a blank response due to content filtering, token limit exhaustion, or an invalid prompt. Check the node output for any error fields or status indicators. Also check the node's On Error setting — if it is set to Continue (Using Error Output), errors go to a separate branch and the main output receives nothing. If On Error is set to Continue (Regular Output), the error data might overwrite the expected response structure.

If your workflow has branches or multiple paths, the downstream node might be pulling data from the wrong upstream node. Instead of relying on implicit data flow (which uses the directly connected node), use the explicit $('NodeName') syntax to reference the exact LLM node by name. This eliminates ambiguity about which node's output is being used, especially in workflows with multiple merge points or branches.

1// Instead of:
2{{ $json.output }}
3
4// Use explicit node reference:
5{{ $('OpenAI Chat Model').item.json.output }}
6
7// Or for the AI Agent node:
8{{ $('AI Agent').item.json.output }}
9
10// Access all items from a node:
11{{ $('OpenAI Chat Model').all()[0].json.output }}