Search further, but check who publishes it

In today's lesson we'll use Gemini Notebook again and keep getting more out of the tool. Today's idea is simple: you'll see that finding information on a topic takes seconds. What really takes time is deciding which of that information is actually useful. In this lesson we'll expand the notebook by searching the web, and we'll see how to keep only what we can trust. If you've read the story, the clinic wants the whole team to handle a frightened cat consistently, from reception to the consulting room. And if you don't know Gemini Notebook, I invite you to revisit the previous lesson, and we'll put in a link to the previous lesson so you can revisit the concept behind the tool.

The idea is to create a new notebook. Here we've simply created the starting point and named it 'Cat-friendly handling in the consulting room'. We have nothing else, and the first step in any notebook is to add a source. In this case, the first thing I want to show you is a 2022 guideline from two feline medicine societies, the American and the international one. We have it as a PDF, here it is. It's 40 pages, a substantial guideline with plenty of information and so on, and it's very interesting. But if we only uploaded this document, we'd still be roughly at the stage of the previous lesson, where we took a document and used it to analyse it. Let's go a little further.

Today's idea is that we can add that source and keep creating things from there. But let's start with a source. As I said, the idea is to upload the PDF directly, and here it starts loading. When you upload a file, you can rest assured that it normally contains all the information you can see, and the tool will analyse it and read it perfectly. The problem comes later, when we search, because it gives us website links that can be a little problematic. I wanted to show you something. For example, if you give me a second, if I add this source from a website, which is where I downloaded the PDF, it will load it too, and let's see what we get.

Since we already have a reference, while this is loading, which I wanted to show you later, we can do an initial search for sources. We simply come here and ask it to search for supplementary sources. While I'm showing you a detail, I'll load another link to the same article so you can compare them. Well, when we have a source, you can always click it, and you should, to see what it contains. First, is it complete? This is a PDF, as I said, so it's usually fairly reliable, and we can see that it gives us a short summary and then all the content.

Although the images appear and the layout is a little disjointed, the first thing you can see is that it's very long, so you can be sure the complete PDF, with all its images and so on, is in the sources. And I've also added this one, because although, as you'll see, it's another link to the same guideline, again with an abstract, an introduction and lots of information, let's see how this one is recognised differently, and what appears is many more links, without the PDF in the same form. We need to be careful, and you must always check this whenever you use a source: validate it, check the content is correct and that it isn't just a cover page, a header or something similar. Thank you.

What do you think? We'll delete these two sources and keep only the PDF, just to keep one. And now we'll continue searching for new sources to supplement the information. Right, let's search for sources. You search for sources directly in this box. We give it the instruction. Here we'll say we want cat-friendly handling in the veterinary consulting room, and we have it in fast mode, with web searches selected. If we had documents or other material on a drive, we could use that. Here we want web sources. We click, and it runs the search. Well, we have the sources. It's given us all of them here. We could go through them, validate them and so on.

For now we'll import them directly, and then see how to filter them. Let's make the same request in English. While it's loading them, it runs another search. Here we could select them all, pick just a few, see what we want, then click import, and that's it. Here they are in English, and we simply do what we said. We import them, and we have all our sources. Right, once it's loaded, we can see two shown in red, and here it already says they couldn't be uploaded due to a temporary error. Usually 'temporary' isn't temporary: it's permanent. What we can do is, well, try opening them, right, or open the URL. It's usually this sort of situation: we have a PDF, but there's a verification step first.

We can download it, and we could put the study here. We can have problems with downloads, but anyway, that's done. And now we would simply delete the links, by going here to delete the source. We could delete all the ones with errors. We have 19 sources now, and what we could do is upload the PDFs again. Here we are. And that's it. Now let's filter them. Right, the next step, if you agree, is to classify them, using this prompt, which you'll find in the instructions. We'll say: make a table of all the sources in this notebook that we can read at a glance. We have all 22 sources selected. We'll send it, and it will give us a table.

Right, here is the table with, as we requested, each of the sources, who wrote it, what type of document it is, the year, which brands are present, and references to them too. And whether money is involved: it says 'authors without funding'. Well, they haven't received any kind of payment for this. Right, you have all the information. And with this, the idea is that a table saves us time. In a couple of minutes we have the table, and we can look through it. And in a minute or 30 seconds, we can assess each article and check whether it will be useful or not. And if we did this by hand, it would obviously take hours of work.

What matters is that it writes 'not stated' when it doesn't know, rather than making it up. That's why we must always specify these things. We don't need to check everything in depth. We simply need to decide which are most relevant, or which interest us most. For example, we've seen a source from 2025 here. And we've also seen a source even from 2026 that could be interesting, with the latest developments. Well, we could take a look. Just because documents have sponsors doesn't automatically make that document or source a brochure or a sales pitch.

For example, the feline guidelines from the American and the worldwide association have many sponsors, but that doesn't directly determine quality. It doesn't mean it's a poor-quality source, of course. Right, as I said, we have the references, and here we could start filtering manually. We could visit the sources and assess them directly using some of these criteria. For example, here's a study about pregabalin to reduce anxiety in cats. It may be useful, but it isn't really what we're looking for, even if it's a scientific article from 2023 and so on. It says here that it's sponsored by Orion Corporation, and here we have some data confirming it.

And we can click through to the source, where it will tell us. And in the same way, here, OK, we have certain data that we'd need to assess. As this data doesn't... Well, this study won't interest us much either, so we could go to pharmacology and pheromones and say, well, this one... We don't want it, the pregabalin one. Right, not this one either. These three pharmacology sources aren't what we want: we want handling, right. We can also... In this case I've labelled them too, as you saw. There's a way to label them. Directly, and that also makes assessment easier, so we can use it. And simply, as I said, we won't look at these, so we'll delete them. And that's it.

Besides this approach, where the factors we've classified can help us do it, another interesting approach would be to try a prompt directly, asking it to use this table to decide which sources can stay in this notebook. We send it and see what it gives us. Right, here's the table, and it identifies several, several sources that it actually suggests keeping. Why? Although it mentions one with corporate sponsorship, we know it's the official guideline, as we'd already seen, so we can keep it. And we also have a few to review, because they're undated or, well, for whatever reason, due to different factors. We may or may not use them. We'd check them manually here and look them up.

What matters is that you can decide. Here, for example, we have these two: the website and the PDF, so we could remove the website. Right, you choose, and we can somehow remove what doesn't interest us. And, as I said, from here we now have a set of sources we could use. Here we have 11. We can narrow it down further, or as we see fit. Finally, let's try the other type of search, by clicking here on Fast Research; here we switch to Deep Research directly. We'll give it a slightly more in-depth task, where we'll tell it... As you can see, we've first written a more carefully crafted prompt, which we could also have used earlier to find better sources.

This matters too; it's an important lesson. But now let's see the result of Deep Research, rather than a quick search for sources. Right, about four minutes later, Deep Research has finished here, and it gives us a report we can read. It's a report produced with artificial intelligence, so it shouldn't be part of our set of sources. But it's interesting: we can study it, review it and take a look. The point of this part was to show you how the search type changes: it gives us far more sources. Before it gave us 10; now it's 20, plus another 27 it doesn't cite. And it also gives us this fairly long report, which can provide a general summary.

But what matters is basically how a quick search differs from an in-depth search, and that the brief matters: the more tightly we define it, the more it narrows down the information. And that's basically it for today's lesson. We started with a notebook that was empty, added a guideline, tried different searches, filtered the results, and in the end the idea is to keep just a few sources, so that, in the next lesson, we'll use this notebook to work with, and we'll limit it to 15 sources at most, and that's what we'll work with. So, I'll say goodbye, and see you next week.