When people are trying variations on a search it's a clear sign of failure. They should do something about it, like use a verbose natural language interface or select a different strategy for ranking or enable the exact in-depth mode. Apparently the NLP community can do natural language Q&A in papers, but Google can't do it in search.
Other pain points: searching in depth all results on a query, not just the top skim and remembering context between searches.
And Google Assistant itself is too poorly integrated and dumb. Where's the GPT-3 like language skill? So many TPUs what are they doing all day long? They have more text and images than any other entity on the internet, Google bot has been sucking it up for so long (not to mention links, keywords and clicks). It should show in the quality of their AI. It's a shame to have OpenAI steal their thunder like this.