A key thing that's driving this is the fact that, currently, Gemini is kind of shit compared to the real frontier models from OpenAI and Anthropic. Add on to that the need for Google to serve an absolute insane number of these summaries and have low latency for each one - this means no real thinking time and a comparatively small models.

So it's partially AI, sure, but it's also because you're getting the summaries from what is an extremely stupid AI compared to the "good stuff" that's out there.