That's a very narrow definition that might have been accurate a few years ago. But of course people mostly use things like chat gpt or claude which can use tools these days, including any search tools you make available as well as web searches. Which means they typically may do multiple searches that don't just send exactly what you type but instead interpret it, adapt to what comes back and then iterate to formulate refined queries, query different search tools you might have, etc. You can still do RAG in the narrow sense you suggest but it's probably not that optimal for answering real world questions.

It’s the type of rag the article is about however, it’s still very relevant.