Transcription
If your job involves tedious, repetitive tasks, then stay tuned, because today I'll show you how to use ChatGPT Projects to do those tasks with consistently high quality, like making reports, manipulating data, tuning up your email quality, and much more. Let's dive in.
Let's start our journey on ChatGPT.com. You can click the card in the top right or use the link in the description. Once you're here, make sure that you've logged in and that you see the projects section in the left sidebar. If you don't see a project section, it means that you're on the free plan. And even though OpenAI has promised projects for the free plan, they aren't available yet. So your screen will have an upgrade plan button in the bottom left-hand corner, which you'll need to follow this tutorial. Enough talk. Let's make our first project by clicking on the new project button. I like making riddles, so I'm going to call this project "riddles." Then click the create project button.
Projects are a great way to organize your past chats. For example, I can take my past chats and drag them in here one by one. I can also add them by clicking on the three dots next to the chat's name and then clicking "add to project" and selecting "riddles." And if I want to add a new chat, I just make sure that this project is selected. And then I add a prompt to the prompt box. We can submit the prompt, and it becomes its own chat. Now if projects were just about organization, we could end the video here. But there are two other curious features.
Let's first talk about this "add instructions" button. When you click it, you're asked a very important question: How can ChatGPT best help you with every chat in this project? Everything you put in this text box will be respected and considered by ChatGPT in all of its responses. But just for chats in this project, here's a simple example. We'll ask for responses formatted with bullet points, only suitable for a sixth-grade reading level. And to make it obvious this is working, we'll ask ChatGPT to start each response with a fun emoji. Click "save" to commit those instructions. Now when we add a new chat, like this riddle, the answer is formatted according to our instructions. These custom instructions are quite powerful. So I'm going to give you five more examples to inspire how you use them.
First, I've just created a project for mentorship chats. I'll use the chats here to ask advice and remind me of the high standards I want to uphold. Fortunately, I have these standards written down in a few documents. This document, for example, contains principles on how to build great products. To supercharge ChatGPT with this knowledge, I'll use the "add files" button in this project. Then I'll drag and drop these two files so they become part of the context for every chat in this project. Once the files are finished uploading, we can click the X in the top right corner to go back to the main project page. Now we'll go back into the custom instructions. The instructions are simple: Use the principles in the attached documents to guide your response. Make sure the advice you give is tied back to the original principle from which it originates. Let's click "save" and ask for some advice. Here's a scenario where some of you might feel this hits too close to home: We have to build an AI-powered feature, but we're stretched thin. Plus, we've got a whole bunch of other things that we need to do. How can we navigate this? Let's ask ChatGPT to find out. Here's the result. We've got a clear set of steps we need to accomplish. And when these steps align with the principles, they're called out explicitly. You can continue adding files and custom instructions to your mentorship project as you learn more about what you need in a mentor.
For our next project, let's automate a repetitive task, building charts from data. Imagine every week we get a PDF report of store sales and we need to turn this raw data into a set of charts for a dashboard. Let's see how projects can help us. Click on the "add instructions" button. This time our instructions look a little different. We're telling ChatGPT to expect an input PDF report from which we want to know the total revenue for all stores for the week. We want a horizontal bar graph of the total revenue by store and a vertical bar graph of the total revenue by product line. I got these other descriptors through trial and error. In a nutshell, I asked ChatGPT for these charts. And if it did things I didn't like, I added instructions to not do those things. Finally, ChatGPT sometimes likes to add lib. So I've added this final instruction to keep it on track. Once again, we'll commit these instructions. And to start the chart generation process, I'll drag in the report that we just looked at. ChatGPT already has the instructions, so I'll just submit the report. And without any additional prompting, we've got the total revenue and the two charts we wanted: total revenue by product and total revenue by store, which means that the next time we have a report to build, all we need to do is upload the PDF. And once again, the charts are created. You notice that the bar widths are different this time. So if you want consistency, put it in the custom instructions.
Next up is a project to help you with data entry. See, I used to work in insurance. And toward the end of the year, we needed a catalog of all the health insurance plans offered in the United States. But we only got PDFs like this one, which had data we had to type up by hand. Now it's 2025. So let's let ChatGPT do this for us. Well, once again, go to the "add instructions" button. And we'll break these instructions into two parts. Part one tells ChatGPT that we're going from PDF to CSV. Part two specifies the format of that CSV. Then we'll save these. And just like last time, we're going to attach a few files and submit these without a prompt. We get a flawless response from ChatGPT in table format, which includes exactly the columns we specified in our custom instructions. And you can download this data into a CSV file by clicking the download button in the top right.
For our next example, we'll tackle another tedious process. And that is addressing customer feedback. Every week we get feedback from our customers that look like this. Instead of us trying to look for themes in this document, let's just ask ChatGPT. Click on "add instructions," and we'll break up this instruction set into two pieces. Piece number one will tell it that we'll provide a CSV filled with customer feedback. And we want to identify the most important customer feedback based on what's the most common negative feedback. Once that feedback has been identified, create an action plan with the following format. Then piece number two is the format, which we'll paste below. This is what it looks like. This we use markdown syntax like pound signs for headers and asterisks and hyphens for bullets. Using square brackets, we also instruct ChatGPT to fill in the results of its analysis. Let's see how it does. Click the "save" button. We'll upload the feedback through the CSV file and once again, submit it directly without a prompt. And after about a minute, you get an action plan. It calls out staff attitude problems as a result of this feedback, beat of service based on two other customers' feedback. And it looks like we need an upgrade to our donut flavors.
For our last example, we'll try something simple yet powerful, a negotiation email strategy. Say you're in a customer service role and you get customers asking for discounts all the time. You want to hold firm on not giving everyone a discount, but you also want to adopt an attitude that we're in this together. So instead of struggling to write every email from scratch, we can start with these custom instructions. Basically, write an email response in the style of Chris Boss, negotiate, but don't agree. Be solution-oriented. You know the drill. Let's save these instructions. So now when you receive emails asking for discounts, you can just paste the email in directly, and ChatGPT will write your response. And while it might get some of the details incorrect, it should start you off strong. So editing is a breeze.
Now if you've used custom GPTs before, you might be wondering what's the difference between projects and making your own GPT. After all, GPTs allow you to provide instructions just like projects. So yes, in that sense, they are the same, but there are a couple of key differences. First off, it's simpler to create a project; fewer options, fewer clicks; you click the add button, give it a name, you're ready to go. But custom GPTs give you more power. Not only can you provide custom instructions, but you can also integrate them with external services using actions. Let's check out a number of other differences back on the homepage. All your chats with custom GPTs appear in the main feed. For example, this conversation is with the Planty GPT. You can't move GPT chats into projects, even though you can organize other chats into projects. With custom GPTs, you're limited to the GPT-4-0 model. But with projects, you have many more options, including the 01 reasoning model and some legacy models. Finally, a big disadvantage of projects is that you can't share them. So if you have a set of custom instructions that you want to share with your team, it's best not to make a project, but instead to make a GPT. Then you can share the GPT using its link. And on a final note, be aware that chats within projects do not share contexts. In other words, every chat is independent, and you can't reference data from other chats.
What are you going to use projects for? If you've got a cool use case, let us know in the comments below. This is David, and I'll see you in the next video.