Why does the OpenAI Chat Model node return a 400 error with 'content' null in n8n?

Below is a checklist based on known issues with the AI Agent and Chat Model nodes in n8n:

1. Check Your Prompt Settings

Even though you’re sending plain text, double-check that:

• The Prompt field isn’t set to “Take from previous node automatically” if the upstream data might sometimes be missing or incorrectly named.
• If you use expressions, verify they resolve to a valid (non-null) string by testing with a dedicated Set node before the Chat Model node.

2. Confirm Data Consistency in Batches

n8n sub-nodes (like the Chat Model node) always resolve expressions against the first item of the input. When handling over 3,000 items:

• Some items may be missing content because of batch or data-mapping issues.
• Consider using a Split In Batches node so that you process a smaller, more manageable number of items at a time. This also reduces any potential issues related to synchronously processing a large load.

3. Validate Field Mappings

Ensure that the field carrying your prompt (often named chatInput for sub-node processing) is:

• Properly defined and not null for any of your items.
• If needed, use an Edit Fields (Set) node to standardize your incoming data fields before handing them off to the Chat Model node.

4. Consider Workaround Options

If the error persists:

• Try changing the prompt configuration from Take from previous node automatically to Define below, then explicitly map the field containing your text. This ensures that the prompt is always pulling a clear, defined value.
• Alternatively, you can experiment with using the HTTP Request node (with batch-limit options) as a substitute for the OpenAI node. This might offer more insight or stability under high-load conditions.

Summary

Even though you’re providing plain text, the error message indicates that for some items the prompt field is evaluated to null (most likely due to data mapping or the sub-node’s behavior with expressions). Making sure that every item processed has a valid string in the expected field and breaking up the workflow into smaller batches should help mitigate this issue.

By verifying your input data and ensuring the proper node configuration, you should be able to resolve the “400 Invalid value for ‘content’” error.