AI tasks much more inaccurate than using same model and prompt directly through Gemini

[grizzled lake]

I'm using an AI Task to look up local sports game information. the AI task continuously reports the wrong information, while copy pasting the same prompt into Gemini directly (both are using the 2.5 flash model) gives me the correct information. What the heck am I doing wrong?

[digital bramble]

I've never actually used Gemini with home assistant, but I have used the API directly. Not sure if Home Assistant supports this, but you likely need to enable Google Search Grounding.

https://ai.google.dev/gemini-api/docs/google-search