they used pupeteer to scrape all the content and then feed that into openai as its training set, and they then periodically rescrape and refresh the integration