That is because it bases its responses on web searches with hits like comments in threads like these or blog articles that were generated from systems trained based on threads like these from the previous iteration of scrapers and generative language models.