I'd give it real questions first. Search Console queries, anonymized customer questions, public threads. Neither model gives you other users' private prompt logs. Try "group these by intent, keep each original question and its source ID, put anything you add under hypotheses. Don't invent volumes." A generated list doesn't measure demand, so keep any real query counts attached to their source too. Otherwise you can't tell an actual question from a plausible guess.
If you are planning to rank on overviews and visible to GPTs then your best friend is backlinks cause with an active backlink profile which is often updated/refreshed gives more crawling signals to these bots. It's the same algo that news algo uses for sure. No matter what you need unique content which gives the straight answer.
