How To Use ChatGPT to Create Presentations (powerpoint, google slides)
Let's build with Zapier and AI (100+ videos)📅 2025-05-07
Transcript
[00:00] You see that right there? I want to do the same thing, but I want to do it automatically. What happened right here is Peter right here went ahead and leveraged the 03 model to manually look at the financial reports of Nvidia and they created really, really cool slide deck. I can go ahead and click through. Boom, boom, and boom. Let's do this automatically with artificial intelligence and automation. So therefore, by the end of this video, you're going to learn everything you need to know when it comes to automatically using AI for financial reports, but you'll be able to apply this skill to anything. Sound good? Let's jump in. Welcome back. Today's video is sponsored by Zapier. You
[00:31] already know me and Zapier have teamed up and they want to do a bunch of cool videos when showcasing you how to use artificial intelligence 03 Gemini Cloud, whatever it may be. In this video, I'm going to set up a very simple automation here that will work around the clock automatically to start creating these kind of presentations based off financial reports. But as discussed earlier in the video and as you saw that cool little camera angle that I've added, there's going to be a lot of cool stuff on this channel. You should check it out. We are going to be doing it specifically for financial reports. But the logic and automation you're about to see could be applied to a bunch of cool stuff. To automatically do it, we'll use Zap year. The artificial intelligence we'll be using is chat GBT. And the
[01:02] example data we'll use today will be Nvidia. Let's go ahead and proceed. We're going to come up here to create new zap. Now, there's a couple ways we can approach this trigger. In today's video, I'm going to do the trigger based off an RCS feed where essentially when Nvidia presents a new press release, eg all that stuff, we're going to be able to do some logic here where it's going to check whether or not this press release is a financial report. If it is, we'll continue with the workflow. If it's not, then we'll exit the workflow. Therefore, we're going to grab this press release RSS feed right here. And
[01:32] for more context, all the RSS feed really does is simply watch that little blog that we saw earlier, grab when a recent new PR comes out and says, "Hey, we got a new PR." And we kind of go from there. So, coming over to our new zeb here, we're going to rename this to auto slides because we're going to be creating auto presentations here. So, we're going to have to create a template here in Google Slides as well. For the trigger, let's go to add it though. We are going to come here to RSS feed and the alternative trigger that could be possible in your context could be a trigger of a scheduling event or auler
[02:02] essentially whatever company you care about you could see what specific dates their quarterly reports come out and kind of just mimic that eg change the app schedule and then set it to specific dates. You can go as far as basically setting up four different zaps that follow Q1, Q2, Q3 and Q4. Apply that logic how you may want. RSS feed we have the choice of new item in feed or alternatively new items in multiple feeds. So if you want to create an RSS feed that not only follows Nvidia but also follows Apple also follows Tesla
[02:33] also follows XYZ company then you can proceed with the exact same logic but just add multiple feed URLs. For now though we'll do new item in feed continue. The feed URL is going to be that URL that we saw earlier all this crazy stuff. Just come up here to the URL command C or control C. Copy paste and we're good to go. We're going to continue here. We're going to test this trigger and we're going to get some data. So, what this will do is this is going to load up data of past PRs, right? So, we got Nvidia, Blackwell, GeForce, RTX or Rise for every gamer starting at 299. But, as you know, we're
[03:03] setting this up specifically for financial reports. Therefore, information like this we don't care about. So, let's set up some logic so we can kind of filter this. We're going to set up a new action here called path. What path is, and if you code simply is just an if else statement. So, if this is true, proceed. If it's not true, then do the alternative of whatever the else is. Before we actually make the conditional statement here, we need to add a GBT block here. Chat GBT. We're going to do an action event of conversation conversation. And we have a ton of different options here, but all we really care about is identifying whether or not this is a financial
[03:34] report or not. So, I'm going to come over here and we're actually going to opt for a very low model. The model we're going to opt for is GBT 4.1 Nano. And if you were here since the beginning, you already know this model's kind of like the GBT 3.5, okay? But now it's 4.1 Nano. Don't ask me why OpenAI names stuff how it names stuff. You just do it. Okay, let's keep going. We're going to say based on this headline semicolon. And what I like to do is put stuff in parenthesis. Going to add it here. And we're going to say title. And we will also include and content semicolon parenthesis. As you already
[04:05] know, the name of the game when it comes to artificial intelligence is give as much context as possible so it doesn't hallucinate. We're going to simply ask for a yes or no question here. So we're going to generate quotation marks. Yes. If this is a financial report in quotation marks, no. If it's not, I'm going to add context block here. In theory, I could laser in a little bit harder here and say this is a press release for Nvidia, but I want to leave this kind of more vague and ambiguous just in case you add multiple lines to your RSS feed or multiple feed URLs. This is a press release from a company. Next, most important thing for you to add here is going to be a memory key. So, this is a random string of 32
[04:36] characters. Uh, when it comes to the model, app for nano, cheapest. Also, you can go a little higher if you're having issues with consistency, etc. Hit continue, and we're going to test the step. Now, I'm expecting this to say no. Just no. So, it actually looks like we need to add the temperature. So, for advanced options, we're going to say true. Coming down here, this should all be good. Higher the number, the more creative it'll be. Lower the number, the more consistent it'll be. And now, we'll add a little creativity. It's fine. Scrolling down here from that previous article that we saw about a new gaming setup. We got the reply of no. That's exactly what we want because now we can
[05:06] leverage this. This logic I'm about to show you. I mean, this is going to open your brain. You're going to be like, whoa, I can do a ton of cool stuff I didn't think about otherwise. So the first thing we can do here is that only continue if we're going to choose conversation by JGBT and we're going to go for reply. Boom. We're going to say exactly matches yes continue. Now here's the situation. We'll say path A means this is a financial report and we are greenlighted and good to go. Click these three ellipses. Rename is report. Okay. Enter. Which means that for path B, this is what happens when it's not a report. So not report. And typically if it's not
[05:38] the report for this specific context, we can do nothing. Or alternatively, you could do stuff if you choose to do so. Same situation. Reply. We say only continue if reply exactly matches. No. Continue. And there we go. As you can see from the previous data, it would continue because it is not a report. So in this context, I can proceed and build out an automation for what to do if it was not a report. Now, if you don't care about that and you're like, Corbin, I don't care. I only want to make this for financial reports. Then we can just take a step back here and we don't necessarily even have to do a path. So I'm going to delete this and we can just
[06:08] do a filter. I just wanted to show you that conditional logic as you may want to interact with each type of report depending on the company differently. Filter same idea only continue if the reply exactly matches yes. Boom. And as you can see it would not continue. Now what's cool about this is just for an example let's go ahead and fill this in with some little dummy data. We come back to our prompt here uh and we just do like Nvidia financial or actually let me just grab the actual headline. Scrolling down here I can go to the financial report here. So, Nvidia
[06:39] announces financial results for fourth quarter of fiscal 2045. Press release. I'm going to copy this title as this is how the data would be received. And then for the content, we can just copy the content. I'm not going to copy it all, but just for the sake of presentation, I'll copy the paragraph. So, we should see yes. Continue. And we should see a green light here. Let's see. There we go. Corbin, why don't you just use some test data from the RSS feed? That's because the last time a financial report came out was February, and we're in April. So, I'm going to show you through this way. Okay. Now, in today's video, I'm going to show you how to do the
[07:09] title slide and the first slide. The steps and processes I'm about to show you can be repeated, repeated, repeated, and repeated for whatever your use case might be. So, coming over to Google Slides, we're going to zoom in a little. Actually, I can't zoom in, so it'll be kind of small up there. We're going to do financial template. Now, the way Google Slides works, especially with Zapier, which is cool, is we can set up variable based templates. So what I mean by that is that I can put a variable here like so name and we'll be able to reference these through this workflow. So I'll go and just put quarter title
[07:41] and the subtitle can be date. With that done, we can go ahead and jump to the next slide here. This next slide is the executive summary. So we will structure the same executive summary. So as you can notice here, some of it can just be fixed text, right? In theory, if we're going to build this template here, we kind of already know the structuring of a financial report. Therefore, for titles like this, we can keep fixed text, executive summary. Although the actual executive summary itself, we obviously need the artificial intelligence to analyze and give us context. So, I'm going to say exact bullet points or I say exec bullet. And
[08:12] there we go. So, they will be able to put data here. We have our title here. Now, let's actually take this current quarterly report and compare it to the previous quarterly report and really see how AI could take it to the next level. So, it's very simple. Uh we're going to go into command C command V and we'll say compare quarters and then we'll say quarter compare. Obviously you probably want to name the variable to something you remember or understand. This can be manipulated changed different color text to add a little bit more flare. Right now we just have a very very clean presentation. Now with this done and
[08:43] Google Slides connected to our Zampier let's do the artificial intelligence. First things first we will be able to you know obviously proceed because the data is correct. So to do that, we're going to hit add step here and we are going to do chat GBT. The event we're going to do is going to be API request. Don't worry, you're going to see some code here, but I'm going to make it super simple, like really simple. Select your relevant account. And we're going to be leveraging one of OpenAI's cool new features here that allows us to actually do web search with an API request. So if you see up here, web search, we're going to be able to add
[09:14] this little tool here and get real information. That is the status quo. With the HTTP method, we'll do post. This is just another way of saying we're sending data. The URL is going to be the responses URL that we saw in our documentation here. Additional header request. This is standard stuff that you typically find in payloads. For OpenAI, we just need two. So, this is going to be the content type, how we're going to send the data, which is application JSON. Then, we're going to add one more here. And the last one here is authorization, essentially our API key, which I'll show you how to place here. But first, let's do our body of the
[09:44] payload. Here it is right here. I should leave this as the first comment in this video. If I don't, remind me. If not, just take a screenshot and you'll be like, "Okay, this is how you do it." Choose the model GBT40. The tool is going to be web search preview and the input is what you're asking for. Now, I'm going to test if this works in 03 and see if they've upgraded by now. It might not. So, we might have to opt for GBT40. You should take the initiative though on your end to see if 03 works because this would be a cool example with 03. As now from Peter here, that is what he did, but he did it all manually. We want this automatic. So, coming over
[10:15] here, we're going to try our input. I'm going to go ahead and use part of Peter's prompt here. So, thank you for our prompt. So, because of the fact that we know previous in this workflow, it's getting triggered on RSS as you see right there. Therefore, we know intuitively this report exists and we can start leveraging the data that exists. What's cool about web search preview is that we can simply say pull the most recent Nvidia financial report. And since we know RSS feed that report exists and it's the most recent one, we don't really have to identify the report itself. So, we're going to go ahead and
[10:45] do this first prompt here, and then we're going to do something I like to call a layering effect, where we're going to be able to output all of the information we care about with AI. Then, we're going to format it correctly, and push it to that slide deck. Here is our first prompt. Pull the most recent Nvidian financial report, extract the AI related revenue figures, run throughout a trend analysis, Q Y, percent of total, TTM, and share your calculations. I want you to go beyond the basic idea and really come up with insightful narratives about Nvidia. I'm going to try L3 here. I don't think they have upgraded it for it yet when it comes to
[11:17] specifically web search preview. Let's try test for context. 03 is the higher level model does deeper analysis and contextualizes information better than the 40. But 40 is still really good, way better than it used to be with that 3 of the five. As expected, right now 03 can't do web search preview. So we'll up 40. Also, I realized we need to add our key. To do that, simply go to your OpenAI profile API keys and paste it over. Here's my key right there. You can try to use it. I'm going to delete it. So, what is super cool as you can see right here is that we got the relevant
[11:47] information that we're looking for. Nvidia's recent financial performance underscores its pivotal role in artificial intelligence sector and it starts grabbing recent data, right? So, we're looking at Q1 FY2025 and a bunch of other information that's really cool. Also, I noticed that it seemed to opt for markdown. Now, what's cool about this output is that it actually already inherently did the comparing between quarters in past information such as data centers revenue increased by 154% year-over-year reaching approximately 22.6 billion out of the total revenue of 30 billion. That's a lot of billions. Therefore, let's leverage that. We're
[12:18] going to do the layering effect that I talked about earlier, which is going to be CHBT again conversation. The user message is going to be so we're going to say here is the most recent company financial report. We're going to add the response body output the following. Now, here is the situation. We are using a very high level model here. We're going to be using 03. As you can probably assume, the higher level the model is, the better it is at following directions. So, let's be very specific of our directions here. Coming back to our slide presentation here, the first variable we need to identify is the quarter and then the associated date. Current quarter semicolon date of
[12:50] release of report semicolon. Now, in theory, we don't necessarily have to get this to be honest with you because we could just grab that from the RSS feed. But let's keep proceeding here. The next data point we need is the exact bullet. So let's get that three bullet points for the executive summary. And finally quarter compare. So next I'll say summary of how this quarter did to the past and what stands out. I'll put the data as each new line. This will make more sense. This is going to allow us to format it using a formatter block with Zapier here to then extract it and put into our slides. With this done though, lock in a memory key. We're going to test this step. Not bad. It did work.
[13:22] But notice how it's still providing the, you know, original data points that we're requesting in the payload. We don't want that. So, we're going to go and simplify this and simply do output the data as each new line and just the output. No text before or after. And sometimes you can give an example like eg don't include the text current quarter semicolon in the output. I'm going to continue and test the step. And we're just going to grab the date from the RSS feed. Anyways, and also I want to show you a really cool thing that you could do to get even more information for this logic. And here we go. We got
[13:53] the title. We got the relevant bullet points. And finally, we got the summary. Now, one thing you could do because we're grabbing this from an RSS feed, so we can actually leverage something else in Zapier called web parser. With web parser, this allows us to take different links found on the internet and grab all the information relevant to them. So, parse web page, configure, URL to parse. We're going to go to our trigger here and we should find it right here with link. So, we can do that. Continue, test step, and this is going to grab all that data and output it for us. This payload was requested as HTML as you can see
[14:24] right there. I'm going to go and delete that and we'll keep going here. So next up is we got to format this data. So we're do formatter. We're going to do the action event text because we're dealing with a text output. The transform value we're going to do split line or split text. Continue. The input is going to be the previous output which is the previous AI block. The L3 we'll select reply right there. And the separator which you can look into Zapier's documentation but one of them is called new line which if I remember correctly I believe it is like this. Okay, it is like that. So it's new line segment index. We're going to do all as
[14:55] separate fields. This will make more sense once I show you what this does. We're going to continue here and test this step. So, it does see with our current output, it's only outputting the data as one block. So, I'm going to make a slight change to our 03 prompt. So, what I'm going to do is after each data point, we're going to simply add a comma. I'm going to say add one comma after each data output. Now that in our output, we get a comma. We're going to be able to leverage that with just the simple app formattered by Zapier text event. In the config gear, we'll split text again, provide the input here. And our separator is simply just going to be that comma. Pretty nice. And make sure
[15:25] you choose all as separate fields. This will output the data. So, we can actually grab between it rather than just having one big chunk of data. Let's go to slides here. Now, if we didn't do that previous step when trying to mess with the data found in that 03 block, it would just be one big chunk, which wouldn't be really good for the template we're about to do right now. Choose action event, create presentation from template. Continue. The title of the presentation, we could just say Nvidia report is shared, your discretion. And the presentation itself will just show up here. It'll be financial template. And boom, we get all that variable data ready to be plugged into. So for the
[15:57] date, we'll add from the RCS feed. Boom. We're going to go to quarter title, hit add. We will simply select our relevant quarter title and fill in the rest of this information. So our exact bullets will be the exact bullets we got on our output, and our quarter compare will be the summary that we got from our output. Let me go and add that. With all that added, we can simply hit test app here and this should automatically create a presentation. But here's a couple things. Due to the limitations of artificial intelligence when actually get through API, we are kind of constrained when it comes to making specific charts for this financial report, but in theory, we could load up
[16:27] an OpenAI prop that is specific to image generation through DLE, etc., etc. So, there's ways to work around this, but for now, these financial reports would need to be much more textbased with a little bit of maybe like some fancy images here and there. Let's check it out. And here we go. All done automatically. I could have been asleep and this would have happened. We got our title here that obviously we can format to how we want it to look in the automation itself. We got our specific date. We got our executive summary with the three major bullet points here. And we got a summary here of comparing the quarters all done through artificial
[16:58] intelligence. Now we've successfully set up an automation that will listen using that RSS feed say hey does Nvidia have a new PR that's a financial report and output a slide all textbased all good to go and what you would want to do in this context if you wanted to add graphs and maybe more specific images to that underlying report is probably a hybrid approach where simply you set up this automation all for the text based information you'd want to show up in a slide deck and then you fill in the gaps with maybe using the 03 model within chatgbt.com but without further ado using Zapier Open AAI all automatically.
[17:30] Make sure you leave a like if you feel like you learned something. Those are two random videos.