Then you can send that to another service; pushing math to wolfram alpha (etc) makes chatgpt suddenly give perfect math answers and it works (it takes a bunch of hacks to ignore it’s confident lying part of answering equations). When we it names a person or location, it shows Wikipedia or maps or both etc.
In this way you use it’s bluffing powers without assuming (or even reading) it’s content; it returns articles etc (which can be wrong as well of course but at least you might know if it’s reputable or not).
I can easily see how bing or google would use something like this to search and return very relevant content, the queries might not be better than some people can create themselves with effort. However many people cannot create good queries and chatgpt creates good queries from extremely little input.
More importantly; it’s a conversation; you can fine tune the results. You can tell it that you meant the programming language, not oxidation when it shows completely bonkers results.
This will work and will work very well as far as I have been able to try. I cannot go further as I have no easy access to search engines without getting blocked immediately for automated querying.
Not sure how this is looking cost wise: chatgpt at scale will even hurt a company like MS for now. Having a million techies use it for some experimentation is a bit different than a billion+ users searching whatever.