SearchResearch (1/2/26): (UPDATE) A Review of the year at the SearchResearch Rancho

[Dan Russell]

Friday, January 2, 2026

SearchResearch (1/2/26): (UPDATE) A Review of the year at the SearchResearch Rancho

 Thank heavens for friends.  

After yesterday's post, a friend emailed me to ask "why didn't you use the advanced models of the LLMs?" 

Fair question, but in my defense, there are about a zillion models out there and I can't test them all.  Besides, what I was asking the AIs to do wasn't that hard.  (At least to me, a human.)  So I assumed they could all do it.  

But as we found yesterday, only Claude (Opus 4.5) was able to do it "out of the box."  

So I tried using the more advanced models.  Guess what?  They all worked and did a much better job. 

However, they're not perfect.  

In my previous post I used Gemini 3 (Thinking) and ChatGPT (5.2 Auto).  They both failed in interesting and distinctive ways.  

This morning I tried Gemini 3 (Pro) next to ChatGPT (5.2 Thinking /Extended) and got much better answers.  

The input data was the same (here's the data SRS blog posts from 2025) which I uploaded to each.  

I then prompted them with:  

[This is a list of text links to SearchResearch blog posts for 2025. For each text link here, please create a spreadsheet with each text link in Column A, then please extract the URL from that link into Column B, and then in Column C, please write a 100 word summary of the content on the URL.]

Gemini did this, then stopped after a few and asked "Would you like me to continue generating this table for the remaining older items in your list (e.g., from September and August 2025)?" 

Yeah, I would.  But I replied with:

 [Please generate all 55 entries]

Gemini and ChatGPT both created new spreadsheets with CORRECT URLs and summaries.  

Here's the side-by-side sheet of each link to the blog post and the Summary by each of the 3 systems.  

And here's what one entry looks like: 

As you can see, the ChatGPT summary is really more of a condensed version of the blog post--not a summary, per se.  Claude's summary is really short, but pretty decent.  I find Gemini's the closest to what I'm looking for in a summary--succinct, and a synthesis / restatement of what was written in the post.  

SearchResearch Lesson 

1. When trying to do something complicated, be sure to test with the most advanced model offered.  It might cost you some money, but you'll get a huge quality improvement.  

2. The same prompt will cause very different outputs from different models.  Try several!  

Keep searching.  

--

Daniel M. Russell 

Check out my book:  The Joy of Search (2019) 

Available in fine bookstores (and online) everywhere.