Someone on Mastodon gave me a great idea for a search tool, but I’m having to think my way around some CORS issues so it’s stewing in the back of my brain. In the meantime I’m working on integrating the Stract search API with Super Edu Search.

I made Super Edu Search because Google’s site: search is both very powerful and so general as to seriously grind my nerves. This is especially true in the higher education Web space. Sure, site:edu is great, but site:all-the-universities-in-California would be even better, or site:all-the-private-Catholic-universities-in-the-US, for example.

The original version of Super Edu Search is at SearchGizmos . That version uses the College Scorecard to find institutions of higher education (and a few other affiliated organizations) that match the criteria you choose via dropdown menus. The institutions which match your criteria have their domains bundled into a series of Google search URLs (they’re divided into sets of 15 bc of Google’s search limits) along with your original query.

With Stract’s API I can’t do giant OR search URLs like I can with Google, but I can get the search results directly (which I *can’t* with Google.) So the Stract version of Super Edu Search generates a dropdown list of institutions which match your criteria. Select one and a Stract search is run for your query in that institution’s Web space and the results displayed.

The disadvantage is I won’t be looking at several institutional results at a time like I would with Google search results. The advantage is since I’ll have the result objects from the Stract API I can do more direct analysis and display more from initial searches. Furthermore, this looks like an interesting use case for that ChatGPT.

It would be interesting, since we’re applying the same query to each institution’s Web space, to have a function which collected the Web pages for a given set of results, shoved the body text into a big ol’ string, and asked a ChatGPT API call to summarize the content using a standardized question, like “Using the content from ${bigOlString}, provide a general overview on the topic of ${originalQuery}.”

I’ve been getting more lately into the idea of slicing search results into discrete chunks using the persistent metadata of time or place (in this case place) and then using AI to summarize each chunk in a way that is so standard as to be comparable. It gives me an output with a clearly defined context of time or place but it is not so specific as to have limited use.