How to summarize and categorize news articles with Zapier Agents

I’m going to teach you how to create a News Story Categorizer agent in Zapier Agents, and we’re going to use this News Story Categorizer template.

This page has a little bit of information about how it works, so review that if you’d like to, but we’re going to go over all those details when we set up the agent.

So let’s go ahead and click Use Template. Again. And this page is going to be mostly review. These are actually the instructions verbatim that the agent is going to be using, but you can’t change them here yet.

So just pointing that out. And then also you can review that the trigger is currently set to On Demand. One of the last little bits of detail here is in this right-hand corner, the little check mark with Zapier.

That just tells us that it’s an official Zapier template. So once you’re done with your review, you’ll go ahead and click Use This Template.

With this particular template, one of the capabilities is that it connects to Google Docs to share the output of the agent.

And so that’s what it’s asking you to do here. So, in my case, I already have my Google Docs account set up.

So I’m that it’s going to use by using it with other Zapier automations. I don’t necessarily have to reconnect it here as a new connection in Zapier Agents, but if you were going to connect a new account, you could click here and click to connect a new account.

You could switch between connected accounts, and then you can also click this widget here to either connect a new account or manage all of your connected accounts.

And also note that this isn’t the only opportunity you have to change which Google Docs account you’re connecting with or other apps you want to use.

It’s just for this particular template, one of the steps that we need to set up. So since I have my Google Docs account that I want to use, I’ve selected that, and I’m going to click create.

All right, so now we’re interacting with Google with the actual agent itself. And like I said previously, these are all the instructions.

We had previously discussed the trigger. Here’s where we can kind of start to interact with it. So essentially, there are different types of triggers that different Zapier agents use.

This one happens to be on demand, and a quirk of using Zapier agents is when you use the on-demand trigger, then the agent is automatically enabled with this toggle here.

If you were to click here and change to another trigger, such as scheduled by Zapier or triggering from another app, then you would have to manually enable the agent when you’re done setting it up.

So just keep that in mind if you mess with any of the default settings. And, uh, we’re going to go ahead and keep that as is.

We’re going to use on-demand for the trigger because that’s how this one works. So I’m not going to fundamentally change anything in this template here, but I might give you a couple of ideas for how you could customize it yourself as we go through.

So let’s just read the instructions. It says, When you ask me to categorize a list of news articles, I will, one, probably go ahead and the article links if they haven’t been given already.

You’ll see how this works when we test the agent. Two, once I receive the links, I will split the links so each URL is an item in a list.

Just makes it a little bit easier to grab the details one by one. And then, here’s kind of where the magic happens.

For each link, I will first access the URL to retrieve the title and the content of the news article. Two, summarize the content.

And then, finally, categorize the article into one of the following categories here. And, as you’ll notice, it’s kind of set up for, I would say, kind of personal news consumption.

You know, you see lifestyle, opinion, arts, U.S. world, and then we do have business as well. My suggestion to you would be to customize this according to the types of news articles.

That you want to consume on a daily basis, for your industry, maybe specific subtopics. So you can really customize this how you want.

And then, the next step is, after all articles have been summarized and categorized, I will compile a final report and create a Google Doc using that Google Docs account that we just connected, and this specific tool, Google Docs, Create Docs, Create Document from Text.

If you click this widget, you can see some places where you could customize it. I’m going to go ahead and leave all of these options to let your agent generate a value for this field, but you could set specific values if you wanted to see it, uhm, say, for example, have a specific document name, or live

in a specific folder, but we’re not going to change anything, so I’ll just click Cancel instead of Save. Bye. I’ll also point out to you this button up here that says Insert Tools, so that’s where something like this Google Docs Create Document from Text tool would have come from, and then also you’ll

notice things like Visit Website is a tool that’s enabled for this agent in order for it to be able to access the URL and complete these other steps.

All right so our agent is all set up because we’re not going to change the But I want to show you how it works.

So let’s click Test Agent. And what we’re doing here is almost exactly what it looks like when you’re interacting with the agent outside of test mode.

So when you’re actually using the agent, and I’ll point out kind of what those differences would be. All right, so the trigger is when I message the assistant on demand.

And if you And per the first step, it is prompting me to provide the article links that it wants me to categorize.

So I put a couple together from a couple of different publications. We’ll see what it does with that. So you notice that this is on its own line, this is on its own line.

And then these two are on the same line, but they’re kind of separated by formatting. There’s a space in between them, and then the links themselves have, like, a little background color.

They’re underlined so that you can physically see how they’re separating them. So that was the next step here. And then now we’re on step three.

It is accessing the URLs and getting that initial content, and then it’ll go through and summarize and categorize the article after it goes through them.

So as you can see, you can see the tool actions Thank you happening in this test, and if you were using the agent, and you can see that it grabbed this first one.

Looks like it already parsed the second one, too. It’s on the third one now. And one thing that I’ll note, too, while it’s going through this test is that certain articles, you may have trouble with the Zapier agent accessing them, and that’s not always unique to Zapier agents.

It’s, um, an issue with web scraping in general, just depending on the website you’re visiting. It may have, uh, certain security measures set up that block access for the scraper to go in and access the URL and summarize the content and retrieve the relevant details.

I don’t think it’s going to happen with my tests here, but I just wanted to point it out that if that happens, the end result report of this workflow will let you know.

It won’t just ignore the article, it’ll tell you that it couldn’t access it. And that’s just something to look out for, because a lot of, especially news publications, which we’re talking about here, typically block access to web scrapers depending on their business model.

So, alright, so we have the third article and we have the fourth. And so, it’s saying, next I’ll summarize and categorize it.

Because we’re in test mode, I have to kind of keep it moving. So I’m going to say, proceed. And it will do those next steps.

You don’t necessarily have to do that when the agent’s actually running. It’s just kind of in the test mode going through the different tools one by one, and the steps one by one.

Alright, so now it is summarizing and categorizing each article. And it’s also preparing the action that comes with this tool here to create that document from text.

And so another thing I was going to mention about personalizing it, and I’m going to go ahead and click approve just so it completes this step, and then I’ll tell you what happened.

Um, another way that you could personalize this instruction in the agent is if you wanted the report to be kind of be structured or formatted in a certain way, um, the reason why you might want to do that is because this report is always going to look different if you don’t.

That’s kind of a quirk of using AI in general, that, like, the same prompt will get different results. So if you want to standardize it, do that in the instructions.

So anyway, as you can see here, it took that action. It gives us a preview in agent. Whereas, er, sorry, in the test mode, whereas if you were using the agent, it would just complete the task and you wouldn’t have to click approve.

But, again, since we’re in test mode, you can see the logic, the tasks, how it gets there, and then I approved it.

So we should be able to now see the document that it created. So let’s click that and check how it turned out.

And so, like, I’ve, I’ve run tests with this before. And, again, it’s always slightly different the way that the report gets together, but it always has the same fundamental features.

It typically adds a date for when the report was generated. And then for each news article shared, it has the heading, the title, the category, and the summary based on the content that it ingested.

And you can see that it did it for all of the articles that it included. There are no errors about websites it couldn’t access to generate this report.

So that’s great. Let me just show you one more thing. Alright, so we’re, we’re done with testing. We have it enabled because it’s on demand, so it was enabled by default.

So in order to use this agent outside of test mode, One way that you can do that is by clicking into the activity.

And then clicking run. There are different ways that you can access this run button, but immediately after building it and testing it, this is the quickest way to then use the agent.

So I’m just going to say hi, kind of to wake it up and to see if it’s going to prompt me with those URLs, since that is part of its instruction.

So that’s how we know if it’s working as intended. So I click run. I said hi, it’s running this behavior and it says, hello, please provide the links to the news articles you’d like me to categorize.

And as you can imagine, we could then go through the rest of this process for a new set of articles and continue doing that moving forward because now we have our agent built.

So you are ready to create your own news story category. Categorize your agent with whatever customizations or tools you want to use.

Go out there and build.