Just because AI Overview doesn't cite All About Berlin specifically, that doesn't disprove that All About Berlin contributed to the generated answer.
I don't know the details of how Google AI Overview tracks data provenance, but I don't think they have an accurate way of tracing the provenance of data in their model weights.
If the value fluctuates much, then the value baked into the model will be unreliable, though? I don't think that the answer here comes from the model's intuition, but rather from the search results it got from a tool call. Especially because it's a clear paraphrase of those cited sources.