This might just be the anti-recency bias that exists in LLMs because they will place far more weight on their training set for things that are well-established (like historian's opinions on President Harding) versus current events.
If you really want to find out for sure, fire up GLM-5.2, MiMo-V2.5-Pro, and K3 and ask them to perform this analysis, and ask them not to use any external web searches. External web searches means your LLM is basically just giving you a summary of the current zeitgeist of the equivalent of a Google search.
In order to avoid the "...
Let's add another one:
This is a feature that Claude now keeps almost completely hidden, and GPT-5.6 hides a lot more than it used to. The masses using these tools should be able to see the reasoning that goes into generating the answer.