Hello,
I have a question regarding the AI Web Research node and its Confidence Scores output. According to the documentation, the field [field_name]_confidence (Accuracy confidence) is described as a numeric score between 0 and 100. However, in practice, the node only returns categorical values: high, medium, or low. Could you clarify why the output is categorical instead of numerical as documented? Is this expected behavior, a recent change, or a configuration option I may be missing?
I also noticed that the Reasoning output sometimes appears in French and sometimes in English. Is this behavior expected? Is it automatically inferred from the prompt, the variant names, or the detected language of the inputs? Is there a way to explicitly control or enforce the language of the Reasoning output in order to keep it consistent across runs?
Thank you in advance for the clarification.
Hey @Sebastien – The behavior you’re seeing is expected. The AI Web Research node currently returns categorical confidence levels (high, medium, low). The reference to a numeric 0–100 scale in the documentation is outdated and will be fixed soon. Thanks for flagging this.
The language used in the Reasoning output is automatically inferred by the underlying research API. It can depend on the language of your prompt, field descriptions, and the web sources being analyzed. At the moment, there isn’t a configuration option to explicitly enforce a single language for this output. To improve consistency, we recommend keeping all prompts and schema descriptions in your preferred language. You can also add a clear instruction to your prompt, for example:
IMPORTANT: Toutes les réponses, y compris les champs “Reasoning”, doivent être rédigées en français.
If you have any additional questions or if this behavior causes issues in your workflow, feel free to let me know.
Hello Wasay,
Thank you for the clear and detailed clarification and for the quick support. This answers my questions perfectly.
Have a nice day !
1 Like