SearchResearch (6/20/26): Celebrating AI search / Specialty search tools

It's easy to complain...

P/C DS Studio at Pexels.com 

... about the quality of AI-powered search tools.  (I've done my fair share!) But when you're searching for something that's fairly difficult, I've often found the answers to be incredibly helpful, especially when the research question is vague or difficult.  They really help you find the right puzzle piece, even in a massive soup of pieces that all look pretty-much the same. 

As you know, this blog is about trying to give you tips and methods to be better at your online research.  But it's also about the best ways to think about what our online research tools are doing.  In other words, what's the most effective mental model you can have. 

If you think about Google as a database search, that's the wrong mental model.  A database search implies that the query will find every thing that matches.  If your query is something like "magic trick" then I'd expect the database to give me back a complete and accurate list of all the hits.  Those are key ideas: "accurate" and "complete." 

But that's not the way any search engine works. Instead of database records, a search engine indexes all kinds of documents--text files, Word documents, PDFs, videos, spreadsheets, images, etc etc etc.  Your search engine finds the most probable hits and then rank orders them by what it thinks is best.  Usually, that means sorting the hits by relevancy.  (What makes something relevant is a topic of long debate and discussion, quickly approaching the zenith of technical discussion.  Here's an article with more details, should you wish to learn more.)  

With the additional AI overviews, the search engines now have another tool that tries to answer your question.  A good old-fashioned Google query (short and to the point) isn't as helpful to the AI as an extended question: for AI questions, longer (with more detail) is often better.  

My point is that people often complain about the changes to search engines.  You're right to complain about inaccuracies and errors, but it's also worth taking a moment to celebrate how truly magical some of the AI-augmented search experience really is.  

There's been a huge improvement in finding difficult-to-find things.  

When I needed some quark (a kind of a fresh, unaged, and spreadable dairy product from Central Europe; halfway between yogurt and cream cheese) for a recipe I'm making, I just asked and got a truly helpful response: 

These are really decent suggestions, although there are a couple of errors. Kalinka isn't at that address anymore and neither is the Slavic Shop at that address. But they're both plausible places to buy quark. (And it was simple to find their current addresses with a quick Google search.)  

Best of all, the answer suggests contacting the stores via their online stock-checker. That just saved me several pointless trips.  And this was an incredibly useful suggestion--I didn't KNOW you could do online stock checking!  

My point is that this was a useful AI-augmented result.  It didn't quite give me the answer, but it told me useful information that I could carry foward to get to the answer.  

As members of the SearchResearch Rancho have noted, it's often true that just putting in old SRS challenges works pretty well.  Current AI search technology just answers them. 

For instance, if you remember the Carolina Parakeet Challenge (find an image drawn from life), copy-pasting the Challenge with the default current Google search gives more-or-less the same answer that we worked out by hand.  

Other AI engines also do pretty well.  Here's Claude's lovely answer… 

But wait, there's more!  Specialty Search Tools!

You might have also noticed that there are an increasing number of other kinds of search engines.  We've talked before about music identification systems.  But there are more: like Shazam for music, or Starwalk2 for things in the night sky, or Vinvino for wine identification. 

I want to mention another special-purpose search app that I've been using recently.  

Merlin Bird ID is incredibly accurate—with roughly 98% accuracy for photo identification and 70-80% for sound recognition. Developed by the Cornell Lab of Ornithology, it relies on massive, crowdsourced databases to deliver reliable results in your immediate area.  

Identifiying birds just by their songs requires human verification.  A part of good search practice is double checking.  For instance, there's the problem of the Northern Mockingbird, which imitates other species (hence, "mocking bird").  Merlin will often identify the mockingbird as the original bird... that is, the one being mocked.  Ah well.  (Pro tip: Listen for consistency: If Merlin flags an unusual bird but you only hear a split-second snippet of it once, it is likely a misidentification. If the song repeats continuously, it's generally accurate.)  

AND, when Merlin hears a bird, it will show a picture of that bird and give you additional spotting information.  (Such as "look in the top branches of a nearby tree; they love to perch there...")  

Here's an example from this morning's birds: 

The Merlin Bird ID interface.

When Merlin hears a particular birdsong, it will highlight that bird in the list (or add it to your list if it hasn't heard it yet today).  That way you can quickly learn which bit of bird song you're hearing is actually that bird.  

SearchResearch Lessons 

1. AI search engines sometimes make mistakes... but they're often useful.  For what it's worth, *I* sometimes make mistakes as well.  Learn from the errors and try to figure out what happens, why, and how to work around the issue.  (Big tip:  CHECK EVERYTHING!)  

2. What used to be hard SearchResearch Challenges are now (mostly) straightforward.  This is a huge shift! And I'm celebrating the increase in our ability to find the answers to complex questions.  And, as always, be sure you understand the answer.

3. Consider other kinds of special purpose search tools. There are a large number of speciality search tools.  It's good to learn which ones are useful for the tasks you do.  (I'll try to collect a list of them in a future post. It will go out of date quickly, but it will show us the range of possibilities!)  

And... 

        ... keep searching.