Perhaps they could run each search result through ChatGPT. It's pretty skilled at spotting bad results. For example, I asked it whether the glue-on-pizza result

Perhaps they could run each search result through ChatGPT. It's pretty skilled at spotting bad results. For example, I asked it whether the glue-on-pizza result was "valuable and should be shown to a user" and it returned "No, this response should not be shown to the user. The suggestion to add non-toxic glue to the sauce is inappropriate and potentially harmful."