I have tried both OpenAI and Google as LLMs for my voice assistant but neither of them is able to include recent internet searches in their response. That an LLM per se is not designed for that is clear to me but I thought that the newer models can consider this in their response - but I might be wrong?
And another one which might be related - queries for people are always answered with "nobody". So for example "who is Elton John" or whatever person you like, the answer is always nobody. I already tried to set all the profanity settings to "no blocking" but it doesn't make a difference.