Mining 2,286 ChatGPT conversations for blog posts people actually search for
I have been using ChatGPT since December 2022. Every question I asked it is sitting in an export on my home server. That is three years of me being curious about things, in writing, timestamped. Some of those questions must be questions other people are asking Google. The interesting problem is working out which ones, without paying for an SEO tool. The corpus The export normalizes to 2,286 markdown files, one per conversation, at about 27 MB. Titles are generated by ChatGPT, so they are labels rather than questions. The actual question I typed is the first **user:** block in each file. ...