CathInfo.com

  When I asked "What does the Church teach about the Novus Ordo Mass?" the AI   
  returned raw JSON instead of an actual answer — you can see it printed {"follow_ups": [...]} directly in the chat window. The follow-up   
  suggestion buttons at the bottom did render correctly, so that part is working, but the JSON is leaking into the visible response when it
  shouldn't be.                                                                                                                             
                 
  The bigger issue though is that there was no actual answer to the question — just the JSON block and nothing else. So the AI skipped the
  content entirely and only returned the follow-up suggestions.                                                                             
 
  Probably a tweak needed in the system prompt to tell the model to keep the JSON separate from the main response, and to always answer the 
  question first before generating follow-ups.

The AI was specifically instructed to NOT ever return JSON blobs. Sometimes it just ignores my commands.

You can ask it to drop the follow up questions altogether. on such a small model size it's probably best.