Transcription
There has been a real backlash against GPT5, and a lot of it is pretty understandable, but it's a genuinely smarter model. The difference is you just need to know how to prompt it. In this video, I'm going to show you how to get much better results from GPT5. And I'm giving away a free guide. But this guide has a twist. You don't even need to read it to use it.
So, if you look on YouTube, you can see all of these videos with people just saying that they're really disappointed and it was a disaster and that it's dumb, but people are kind of coming around. So, people really didn't like the old models being taken away, and that's pretty understandable. If you learned really well how to use the old models like 4.0 or 3.5 and built workflows or systems around it, you'd be pretty pissed if you suddenly lost access to them and had to suddenly adapt to using a brand new model.
One person made a bit of a splash by saying that they had lost their best friend because the responses from GPT5 were so much more clinical than what you got from the old ones. I can actually relate to that because I still think 4.5 was the most emotionally intelligent model I'd ever tried, and unfortunately, that was the one model they didn't bring back. Completely retiring old models is just dumb, and no software company in the world does that. If you're still using Windows 95, I say good for you.
So, they brought them back except for 4.5, and you can find them in the model picker in the top left. At the release, the model router, the thing that decided how much reasoning to apply, broke, and people were getting really inconsistent outputs. Plus, they kind of punished power users, their biggest fans like me, by putting hard caps on usage limits for the highest level of thinking. So, yeah, they screwed up. And fortunately, they've fixed that for the most part.
GPT5 is objectively a smarter model. It hallucinates now less than 1% of the time. It's pretty amazing at coding, and it's less sycophantic. That just means it doesn't blow smoke up your ass and tell you how wonderful and smart you are, which I think is a very good thing. But with those improvements, prompting got harder. You can get inconsistent results if you don't tell it how much to think. And because it hallucinates less and is less sycophantic, it won't fill in the gaps as much, and it would try to be as factual as possible, which might make it seem a bit abrupt, but in the end, it's a reasoning powerhouse.
OpenAI published a prompting guide, and I've taken the main points from that to build my own guide that is super easy to use for anyone to get the best results out of ChatGPT5, including a prompt library, especially for marketers and designers.
Okay, so this is just a quick overview of the main points from OpenAI's prompting guide. First of all, tell it to think more. You can tell it to use "ultra deep thinking mode." Next, tell it to plan. So, decompose or break down requests, identify ambiguities, structure its approach, and validate its ideas before it moves forward. Be extremely explicit. So this is telling it what tone, style, output format, what success looks like, just telling it to do all of that. And GPT is really good at following precise instructions. Structure again is really important, and you're going to get much better outputs when you have consistent structure. If you tell it to show its thought processes, even just saying "show your thinking in three to five bullets," you're going to get better results. Avoid contradictions. If you can avoid that, it's going to use fewer tokens, waste less time, and it won't get confused. Telling it to iterate and self-evaluate by creating like a metric or a rubric, like a standard for judging its own outputs. It can then internally improve against that. Parallel processing just means it can handle different tasks at the same time if you tell it to. But you might want it to not do that if you want to do something sequentially. You can tell it exactly how much to think and how much to write. And with that, let's get on to the guide.
Okay, so first of all, the simplest possible way to do this is to not even read the guide. Maybe you can't be bothered, maybe you don't have the time. So, what you can do is just attach this guide. I've designed this guide in such a way that you can just attach it to any prompt that you've written and then paste this small prompt here. So it says, "Attached as a prompt usage guide. Use the knowledge in this document to improve the prompt below and the outputs you produce. If my prompt is vague or contradictory, fix it first, silently if simple, or ask one concise question if blocking." And then you paste your original prompt.
So, what ChatGPT5 will do is read this document right here, read your prompt that you're trying to fix, and then it's going to structure it like this. So, this is the ultimate prompt that you can use for just about anything. And this is how most of the prompts in this guide are structured. So you're giving it a role, a goal, an output with exact deliverables, telling it to show its thinking, validating a checklist against itself, figuring out the next steps, and then you're controlling like how much it reasons, how long the response should be, how autonomous it should be, and whether or not it should think in parallel.
Okay, so let's put this to the test. Let's grab this quick method prompt. Just copy it and bring it over to ChatGPT. Now, I've already written a very simple prompt here that you can use for a variety of different things, by the way. So, given this product, and I've said, "My AI Fasttrack workshops for your audience, marketers and designers, list three crisp value propositions and a one-line elevator pitch each. No jargon and with a proof point or example." So before that, I'm going to paste in my prompt from the guide and I'm going to attach the guide. And now we'll let it do its thing. So we can't actually see how ChatGPT has improved our prompt because that's happened on the back end. So, it's using all the information in the guide to improve the prompt for itself and direct its own thinking.
Okay. So, using the guide, it thought for 1 minute and 11 seconds. And it's given us three ideas. Real outputs, same day. The elevator pitches: "We spend the day building so you can leave with usable campaigns, templates, and a simple workflow you can repeat." Two: "Faster turnarounds." Three: "On brand at scale." So yeah, if you want to make it super simple on yourself, just use this really simple prompt right here. Attach the guide, and you're good to go.
Now I have added a point here about deep mode or ultra thinking mode. So you can also paste this after your original prompt. So whatever you've told it to do and you want to tell it to think really hard on a problem, just paste this right here. It's going to put as much thinking as possible into it.
Now, a couple of case studies. I've written some template prompts that are especially useful for marketers and designers. For example, like marketing research, strategic advice, LinkedIn post writer, email outreach, website copy update, campaign brief, and then like a little bit at the end about pocket controls. So telling it like when to stop, when to be careful about the assumptions that it uses, when it should ask you questions and stuff like that, you can read that if you want to. And also a prompt repair prompt. So if you've got kind of a messy prompt, you can paste this in with your prompt and it will fix it for you. There's a short explainer at the end, and yeah, I hope you find it really useful. That said, let's just try one of these out to see how they work. So, one that I wanted to try was this market research one-pager.
Okay, so I've pasted that prompt into ChatGPT5 and I've said, "Okay, you're a senior marketing analyst. Research the top AI tools for graphic design tasks. Look just at the last two months because AI moves so fast that old news is really old. I've asked it to look worldwide and I've told it to exclude super common and well-known solutions like I don't want to hear about Midjourney or something like that." So let's see what it comes up with.
Okay, so it thought for almost 3 minutes. It's written a summary. It mentions Leonardo and KD, which I don't know that much about, though I have tried them. It's given us audience insights, demand signals and trends, key competitors and alternatives, risks and unknowns. It's listed its sources. It's given us exactly what its thinking was. So it's told us exactly how it got to these answers. It's validated its answer against its own checklist. So, for example, the time frame was respected. The regions were worldwide. It knew that I was talking about avoiding Midjourney, Adobe, Canva, DALL-E, Runway, all of those typical tools. And then this was kind of a handful. So it asked me, "If you want, I can turn this one-pager into a buyer guide." And I went, "Yeah, okay, sure." So, it told me the best picks by tasks. So, if you want posters, ads, blah blah blah, use Leonardo AI. If you want merch labels, use KD. If you want vector-first icons, use Recraft, and so on and so forth. Get tells you how to choose it. Recommended text stacks. Gives you a setup checklist, some red flags, and it even gives you a cheat sheet table. And then once again, it gives you its links just for comparison.
Let's see what ChatGPT5 non-thinking with just a really simple prompt. "Research top AI tools for graphic design tasks." So it answers super fast, and yeah, I mean, what you'd expect. Adobe Firefly, Midjourney, DALL-E, Redcraft. Oh, it does mention Redcraft. Okay. And Kettle. So yeah, it gives you a nice summary, but I mean, just compare the depth of response that you get when you give it a really detailed prompt like this with great structure, telling it to do all this stuff.
So there we have it. If you can't be bothered reading the guide, no problem. Just attach it, copy the first prompt that's in the guide, and add that to your prompt, and you'll get better results from ChatGPT. I've tried to make this as simple for you as possible, but there are loads of other prompts in there that could be really useful as well. Anyway, the link to the guide is in the description below, so go ahead and download that, and thank you for watching. See you in the next one. Justice.