Not necessarily, one aspect of the LLM arms race is to have the most up-to-date records or to use a search engine to find stuff.

If the LLM thinks to consult a search engine. If the next token predicted is the start of a link rather than the start of a tool call, it's going to be spitting out a link. Getting them to reliably use tools rather than freeball seems to be quite a difficult problem to solve.