OpenAI Doesn't Ask ChatGPT to Summarize. It Asks This.

0:00Every week people upload a file to chat GPT and type the same word, summarize. They get a tiny report, they nod, and then they change nothing. The summary was never the problem. The question was. OpenAI just published the real prompts their finance team runs every single week. And their team almost never asks what the data says. They ask where the

0:22next dollar should go. I'm Dylan. I run an AI consultancy. And I've typed summarize for years before this made me stop. So I'm going to show you the questions their whole workflow is built on and how to run it in files that you likely already have right now. So, let's get into it. But the key thing here is these 16 workflows are sure being applied to finance, but many of the things you can learn from these can be applied to really any other team that relies on data to make decisions. And that's almost every

0:51single team inside of a company. Customer support, marketing, operations, HR, etc. So from these 16 workflows, I want to show you the things that actually matter. And the first thing here is the primary use case that they wanted to flaunt through those 16 workflows. Their finance team was in marketing. They struggled to make decisions quickly. It wasn't in the issue that they didn't have data and they didn't have money to spend, but it was figuring out where that money was best suited to be spent. And if this were the week where they're spending the money to publish ads in regards to selling a product, by the time they came

1:20to an answer for making a decision, that week already passed. So the data and the decision was outdated. So they needed a faster way to make decisions. Because if you've done any type of marketing or spending on anything in regards to advertising your services or products, you know that there's a form of diminishing returns. So if the vertical axis here is return and the horizontal axis here is money spent over time, you can see the first dollar that's spent on this channel. So this could be Reddit, Instagram, YouTube, doing some sort of ads. You realize that the return here is much higher. So you put in $1 and maybe

1:49you get back three. But over time as you spend more and you scale, you realize that one specific channel, you put in $1 and you likely get $1 out. So the return on that is very low. And the first primary key we can learn from this example that they talked about in multiple workflows that they shared is that the question that they asked when utilizing AI in their workflows wasn't what does this data say? Cuz that's just a summary ask. Instead, they were saying, "Where should the next dollar go?" And that's the key here. Is they're not simply asking the AI to summarize the data, but instead to take the data

2:19itself and help them make decisions based on it quickly. And by asking this type of question to the AI, you fundamentally have the AI think and do different things. And it's not to say that summaries are a bad thing. Summaries are okay. I think it's a good place to start. But fundamentally, when you ask an AI to summarize something, it's simply shrinking down the data and distilling it into something that's digestible for you. But it's retroactive. It's on the previous month, the previous week. It's something that's already happened. But oftentimes, when we're asking an AI to summarize something for us, we're really trying to

2:47make a decision based on that summary. And that's often where comparison comes into play. When we're asking the AI, based on this data set, this is where we've been spending our time, either our time or money or whatever else, should we shift it over here or vice versa? Quick pause. If you're enjoying this,

3:02you're going to enjoy two other things. First off, below is a 30-day AI insight series, completely free. You'll get 30 insights in your inbox as to how you can apply AI to your business and your work. The second thing is if you'd like to work with me, below are a series of offerings to see if there's a good fit for the two of us. Now, let's get back into the video. And there are three key questions that the finance team kept on asking through these different workflows they were sharing. So, the first one is what everybody's already doing. So, we're simply asking the AI, "What

3:28happened?" So, this is a summary ask. But remember, this is retroactive, and this is just the base. This is where you should start. But quickly after that, you want to climb the ladder of value. So, asking more valuable questions, getting more valuable answers back from the AI. The next one here is asking, "Where am I getting less back?" So, this goes back to the diminishing returns point that I made previously. A common example people like to talk about is pizza. So, if I give you one slice of pizza and you're really hungry, you'll probably really enjoy that piece of pizza. But if I give you five or 10

3:54slices, and after that 10th piece of pizza that you've eaten, it's likely not going to be as good as the first piece that you've eaten. And that's the thing here, is that the pizza has become diminishing over time. The same thing applies to time spent or money spent on different areas. And this question to

4:08the AI helps you answer that quickly. And then finally, we have the last question, which is usually the highest value, and this is what OpenAI asked very commonly in the 16 workflows they provided, was what should I move, meaning time or money, where should I move from one place to the next. And this question directly is connected to the previous. Now, these last two questions, I'm going to give you exact prompts you can copy and paste and use for yourself. So, the first one here for the second question is a prompt that you can copy and paste. So, the prompt here is simply telling the AI, I want you to

4:33look at this data, assuming you've given it data, where am I getting less and less back for the same time or money that I'm spending. And importantly, the second part of this prompt is I want you to show me the numbers that made you think so. And the reason this is so important, I'll talk about this in the second prompt as well, is that we're asking the AI to provide proof to us. We

4:49don't want to take its word for it. Every single time an AI gives you advice or a decision on data, it should always provide proof of where it got that data from and how it inferred that specific decision. That's what the second part of the prompt is doing. Now, our prompt for the third question, and this is honestly the prompt that you definitely want to copy and paste from this video. And on that note, you're going to get the presentation below, so you can easily copy and paste this. This is the prompt that OpenAI actually showed in their blog posts. So here, we're asking the AI, I have to decide where my time and

5:16money goes next week. Write your answer for a busy owner who can only change one thing this week and needs to pick the right one. Now, I'm going to stop here for a second and emphasize the importance of this part of the prompt. Here, we're giving the AI an audience, not a persona. In the past, a lot of prompt engineering best practices, people would tell you to tell your AI that it's a world-class writer or world-class business advisor or something like that. Giving an AI a persona actually gives you worse outputs now with newer models. So, if you're using GPT-5.6 or Opus-5 or Fable-5 or anything like that, these newer models

5:46get hyper-fixated on trying to be that persona, and they actually degrade the output over time. So, instead of giving it a persona, you want to give it an audience, and that's what this prompt is doing. We're saying you're serving a busy business owner. So, that's the first thing I'm going to call out. The next part of this prompt are the three questions that the AI needs to answer based on this data, which is, "What should I do more of and why?" So, where should I double down my resources, either time or money? "What should I pull back from and why?" So, where should I take resources away from? And

6:13the final question here that's connected to the first two is asking the AI, "If I make the one trade, what do I gain?" So, either if I'm taking something away or doubling down on something, what's the benefit there? And in the end, we're telling the AI to tell me what this data can't see that might change your recommendation. And this is again a really key point that I'll call out is that the data doesn't have access to everything. There's some data you might not share with it or some data that you just can't share because it's not something that's explicitly written

6:36down. Those are the AI's blind spots. So, the AI's going to try to guess what they could be that would determine the recommendation and how that would change. That's going to help you trust or distrust the AI's output. So, now that you understand the questions and the associated prompts of how you can think about this, you already have data sets you can use to do this activity now. So, either do it tonight, today, this weekend, whatever. So, there's three data sets that came to mind for me. The first one is your calendar. So, everybody has a calendar where they're spending their time. So, you can export

7:01last month's time of how you spent it, give it to the AI, and ask it the associated questions. Often times, there's some surprises and harsh truths where we're wasting time. So, the AI can help there. So, the next data set is card statements. So, this can either be expenses you're spending money on personally or professionally or ad spend similar to OpenAI that I use case I just walked you through. You can do the same exact thing they did just for your own data. On the subscription side, obviously if you just export all your expenses, ask the AI, "Am I wasting money on anything? Am I using something

7:29to its fullest potential?" If not, then maybe I can save myself on expenses there. And then the ad spend is self-explanatory based on what I explained earlier. And then our third one is the client list. So, a client list is basically covering who your clients are and how much time you spend with each one of those clients. And then we want to ask the AI, based on these clients, should I should build relationships with certain clients or let go of relationships with others. So, if somebody's paying you very little, and they're a huge hassle, then maybe you should let go of that one. Somebody that pays you a lot, and they ask for

7:56very little, then maybe you should spend more time with them. This is just another way to use data sets to actually ask these types of questions. Now, one important thing that OpenAI talked a lot about through all of the workflows is the importance of having the human or maintaining the human in the loop. So, we can't simply just take the AI's recommendation at face value and just do what it says. That means you're outsourcing your judgment to the AI, and that's not okay. What we're doing here is instead we're augmenting our judgment with the AI. So, we get the output from the AI, but then we scrutinize it in two

8:21ways. First, we're asking the AI to validate. For every single number or decision you provide, there has to be an associated number and source and citation associated to that, so I can validate that it's true. And then also, we've already talked about the blind spots. There's a lot of data the AI might not necessarily have access to. For example, maybe your business is seasonal, so it spikes in Q1 and drops in Q2. The AI needs to understand that,

8:42so it doesn't make a bad determination. Another one that might be relevant for a lot of you is that maybe you have a low-value client, {quote} {unquote} low-value, that gives you a lot of referrals. So, the AI needs to understand that context to make its determination. There are many of these examples, those are just two. So, the point here is not to simply accept what the AI gives you, but

8:57scrutinize it in many different ways. These are just two really good ways to do it. And then you decide what you're going to do based off of that data and that recommendation. Here are the three primary things we can take away when it comes to those 16 workflows OpenAI provided. The first one is when we ask questions of AI, especially in regards to data. Instead of asking what the summary is, we should ask, what should I move? I.e., what decisions should I make based on this data, not simply the summary for it. I then shared that there are three obvious data sets you probably have access to right now where you can

9:24practice this activity to understand how the AI performs over time. That's going to be your calendar, your card statements, and your client list. And then the final thing here is that we should always check before we act. Meaning that if the AI's going to give me a recommendation, we need to validate that the data is good because the AI's given us proof and receipts and citations, and we've double-checked the blind spots. We've asked the AI to tell us what they are and how they could impact the output. So, we're applying our human judgment and being augmented by the AI, not outsourcing it completely to AI. And that's it. So, as a reminder,

9:53two quick things. First off, below is a 30-day AI insight series, completely free. You'll get 30 insights in your inbox if I can apply AI to your business and your work. The second thing is if you'd like to work with me, lower series of offerings to see if there's a good fit for the two of us. And if you've enjoyed this video, you're going to love this video because the YouTube gods say so. So, give it a click, and I'll see

10:15you next time, internet.