Transcription
AI models keep getting smarter every day, but there's one thing that never changes. They only give you great answers if you ask them great questions. To do this, you don't need to become technical or learn some advanced logic. You just need one simple framework. It's called Riceco. Six prompt ingredients that when used correctly won't just improve your AI results. This can change how you work, create, think, and communicate. I'll break down each step with examples. You won't need all six every time. So, I've ranked them by importance and frequency. Then, at the end, I'll give you the condensed three-step version that covers you 80% of the time. And everything I'm covering works in every large language model. It doesn't matter if you're using ChatgPT, Gemini, Claude, Grock, or any other. This formula improves your outputs no matter what.
First is roll. Assigning the model a role is like flipping a switch. It instantly changes the tone, perspective, and depth of the response just by telling the model who it is. This won't be necessary for every prompt. That's why it has a lower importance ranking. But when it does apply, it's one of the most powerful shortcuts in prompting. But here's an example. I'll use the exact same simple prompt three times. The only thing that changes is the role. Starting with no role at all. Give advice on how to get better sleep. So this comes back with generic standard advice. Not bad, but nothing special. Now let's try. You are a board-certified sleep doctor. Give me advice on how to get better sleep. This gives a concise evidence-based approach to better sleep. Now, how about you are a sleep-deprived parent writing to other parents, give advice on how to get better sleep. So, this gives a much more conversational and emotional relatable advice. The type of thing you could use in a blog post. Even though the task stayed the same, each version responded completely differently in tone, priorities, and level of detail.
Let's do one more example. So, some options if you were pitching a startup idea or a business. I'll use our own product, Skill Leap AI, the learning platform we have on Futureedia. Explain what Skill Leap AI is and why it matters. This one with no role gives the generic overview and kind of feature dump type of answer like right here. This is often called the Netflix of AI learning that comes straight from one of the old landing pages. The why it matters is pretty generic. So now I'll change it to you are the founder of Futuredia pitching to potential investors. Explain what Skillip AI is and why it matters. So this starts with the vision and down here on the why it matters. That's really good. The skill gap isn't a looming threat. It's here now. Halflife of skills in tech and many industries is under five years. Companies lose billions from outdated skills. That's great. Shows like a big pain point and then offers the solution. Then it focuses on market potential. So yeah, just completely different perspective and approach. Now how about you are an AI YouTuber recommending Skillpap AI to your audience. Explain what Skillip AI is and why it matters. This YouTuber version is more casual, personal, and audience focused. I mean, this sounds like something a creator might actually say. Not me. I would tone it down for sure. I could never pitch something so hard like this, but it's directionally right. Each one was given the same product and task, but depending on the role, it framed the product in a totally different way. This is exactly why RO is such a powerful shortcut. You're telling the model how to think, not just what to say. It works for serious prompts, creative prompts, and everything in between.
Next is instruction. The most essential part of any prompt. This is the core task you're giving the AI. What exactly do you want it to do? It's also the most obvious part, like you're going to include this every time. But there are a few really common mistakes here that are easy to fix. Most important is to be specific. Sometimes you think you're being clear, but the instruction is actually too vague. And whenever something is vague, the AI has to fill in the gaps, and it usually does that with generic middle-of-the-road content. A weak instruction would be write me an engaging YouTube short about prompting tips. A stronger instruction is write a 60-second YouTube short script about prompting tips using a curiosity gap hook and a scroll-stopping visual anchor. The first one leaves too much to interpretation. The second gives the AI a real job to do. Instead of using a vague term like engaging, which leaves it open-ended, I define what engaging looks like more precisely. Same idea with words like cool or interesting. Like instead of make it interesting, try start with a surprising stat and end with a thought-provoking question. Something like that. Be direct, be specific, tell the AI what job it's doing, the rest, like tone, audience, and purpose comes next in context.
Once you've given the AI a clear instruction, the next thing it needs is context. The background that makes the output relevant, useful, and actually aligned with what you're trying to do. This is the part most people skip or rush, and it's the biggest reason why so many outputs feel generic, surface-level, or just off. Some types of context to include are audience, who is this for? Background, what's the business platform or scenario? Purpose, what are you trying to accomplish and why does it matter? Tone or perspective, what kind of voice or framing should it use? So, here's an example. With no context, write a 500-word blog post about AI video tools. And with context, write a 500-word blog post about AI video tools for small business owners who want to turn podcast episodes into YouTube shorts. The goal is to explain the value in simple, practical terms. Same instruction, but now the model knows who it's writing to, what the scenario is, and how it should shape the tone and focus. Sometimes context gets blended into the instruction, and that's totally fine. For example, write a LinkedIn post that explains how AI is helping HR teams write better job descriptions. Do people still use LinkedIn? I don't know. But that prompt does include context. It's got the audience of HR teams, the platform, LinkedIn, and a topic AI for job descriptions. But it's all written as one line mixed in with the instruction. Instruction and context often overlap, but thinking about them separately helps you catch what might be missing. Now, a little context goes a long way, but more context is usually better, just as long as your instructions stay clear. You can even give the model a full context dump. Especially for complex or strategic tasks, it can read multiple pages of context instantaneously. Just make sure the core instruction doesn't get buried or lost in the middle. Clarity still comes first.
If the Rice Go method is clicking so far, I have a free resource from HubSpot that pairs perfectly with this. It's a 7-day playbook called Advanced Chat GPT Prompt Engineering. It incorporates many of the ideas I'm covering here, then expands into a lot of different directions and systems you can develop around them. This is structured as a 7-day system to walk you through the more advanced levels of prompt engineering. That includes prompt templates, example scenarios, and guided exercises each day. My favorite part is the modular systems development section. That's where you learn to break prompts into reusable building blocks and link them together for different use cases. Then it moves on to building your own custom frameworks, complete with testing and validation protocols, so you know it works before you rely on it. There's a ton of value here. It's the perfect next step after this video to actually implement these concepts into your workflows. You can download it for free using the link in the description and thank you to HubSpot for sponsoring this video and providing valuable resources like this one.
The next step is examples. In specific scenarios, this has the biggest impact. Examples show the model exactly what kind of output you're looking for, whether that's structure, tone, formatting, or even logic. You're not just telling it what to do, though. You're showing it how to do it. This is most commonly used in writing prompts and for good reason. Let's say you're asking the model to help draft your newsletter. You could include a real one you've written before and say, "Use this as a reference for the tone and format." That one example gives the model an anchor, something it can imitate or remix rather than guessing from scratch. And this result is really close to the example I gave. It has a very similar use of emojis, has all the same main sections, and it actually went out and found new updates to include in these other sections. All I instructed it about was GPT5, but the example had all these other sections and it knew it needed to fill them, so it figured out how. I don't know how accurate all of this is. I'd have to double-check it all. But that demonstrates how impactful adding an example can be. And this is what's called few-shot prompting. If you ever see that term, it just means including examples. And it works really well. Even just one or two strong examples can make a big difference. Some scenarios, you might want more. For the newsletter example, three or four might be helpful. But for something like short-form hooks, I have a list of around 20 that I reuse. If it's longer writing, like a full 20-minute YouTube script, one strong example is usually enough.
But this isn't just for writing. Examples are incredibly powerful in less obvious use cases, too. For instance, the JSON prompting method for the V3 video model has been all over lately. I don't actually use it. If you already have good structure in your prompts, I don't think it really makes a difference, but that's a separate topic. I'm just using this as an example because the prompts look intimidating at first, but it's super easy to get these if you use a prompt with a few examples. All it takes is explaining the video you want in natural language, like a woman riding a snow leopard through the Himalayas. Then ask the model to format it like the example and paste in a few JSON prompt examples you find on Twitter. And it works right away just from that. I'll actually go ahead and give this prompt a test in V3. And that looks really good. And you can do the same thing with any type of image or video generation prompts, but also data formatting or even multi-art logic. So if you're ever struggling to get the tone, the structure or formatting right, try giving it a clear example. It's like handing it a cheat sheet.
Next is constraints. This is where you tighten things up. This step defines any limits, rules, or must-haves the model should follow when generating your output. Also for helping eliminate bad habits the models often fall into like being too wordy, too vague, too repetitive, or too safe. So, here's what this might look like. Keep it under 100 words. Avoid buzzwords like innovative or cutting-edge. Don't use hashtags. They're lame. Use a warm conversational tone, not overly corporate. Include a quote and a stat in the first paragraph. The output should feel like it came from a founder, not a marketer. I could go on and on with those, but this is different from context. Context gives the AI the background. Constraints give it the boundaries. You don't need a long list every time. And if you're collaborating with AI long-term, like using it for content workflows or automation, setting clear constraints early saves you a lot of revision time later. And you can add to them and touch them up over time. You can even add them into custom instructions or I use projects for pretty much everything in Chat GPT. And that lets you add custom instructions that will apply to every new chat you start in a specific folder. That's a great place to put constraints and examples.
The final step is output format. This is where you specify how you want the response structured. It's not necessarily about making your outputs smarter, but making them cleaner and more usable. You won't need this every time, but this is things like bullet points, a table, JSON, markdown with headers, a tweet thread, a three-act script, or write this as a set of carousel slides, or as a flowchart. There's tons of ways this applies, but it can be really helpful. Like, say you're asking for a comparison between three tools. You may end up with just a wall of text. Instead, you could say, "Give this to me as a three-column table with columns for tool name, key features, and ideal use cases." That one line makes it way more readable. And you can push this into more advanced formats like create a mind map structure using nested markdown lists. Format this as a code snippet I can paste into my app or a whole range of options for data analysis like bar graph, box plot, heat map, histogram, pie chart. The more you define the format, the less time you'll spend reformatting, rewriting, and trying to figure out what to do with a block of unstructured text.
Now that we've seen how each part of the Riceco framework works on its own, let's put it all together. I'll use a prompt that utilizes all six steps. Then I'll show the condensed must-have version next. I'll also cover what to do once you get your output back. But here's using the full framework. Starting with a bad prompt. How can I implement AI and automations into my real estate business? Pretty vague instruction. It has almost no context to go off. It's way too open-ended. Just a bad prompt. So now let's use the Riceco framework. The role is you are a business growth strategist specializing in AI adoption for businesses. I don't put headers for each step normally. That's just here to be helpful for the demonstration. The instruction identify the top AI and automation opportunities to help streamline my business and produce a prioritized action plan that maximizes time savings and revenue growth. For the context, this is basically a context dump about the business where leads come from, areas the most time is spent on. Under examples, I did some potential areas to automate. Then constraints has the $400 budget must be less than 3 hours per week to maintain and not be overly technical. The output format present as a prioritized AI and automation playbook with quick wins, core systems, and long-term growth.
Now, for the result that came back, let's look at what happened when we actually used all six steps. Instead of a vague templated response, this is completely customized to the location and market I said I'm in. It's within the budget and time limits I gave it. And it's all broken down into a very actionable plan that targets the specific pain points I mentioned, plus additional suggestions common to real estate needs. It gives a projected time savings of 8 to 12 hours per week while closing more deals. Of course, you need to test and validate some of this stuff, but most businesses have a ton of areas they can streamline like this and see this type of result. Not to be the AI hype guy, but taking a response like this and following through on it could actually change someone's life in business. I mean, I do have this YouTube channel and we have the website and the full course platform dedicated to this. So, maybe I am the AI hype guy, but I have seen results like this so many times, it's hard not to be sometimes, but I'll move on.
Now, for quick prompts, you can get amazing results by just using the top three steps. These are the steps I find myself using on almost every single prompt. Whether I'm writing scripts, researching topics, or outlining course content, IC covers the three most important ingredients: instruction, context, and constraints. If you only include three things in your prompt, make it these. The instruction, what exactly you want the model to do. Be specific, not vague. Context, who is it for? What's the background? Why does this matter? Then constraints, any must-haves or boundaries around tone, structure, or content. And that's it. So, here's a super simple ICC example. You've got the vague prompt. Write a Twitter post about chat GPT tips. Already you can see what it will have to guess, like who's the audience, what kind of tone, how long should it be. Now, we'll compare it to this. Write a Twitter thread with five lessons I've learned from using ChachiBT every day. The audience is indie creators and solo entrepreneurs trying to use AI more practically. Make it clear, non-technical, and casually insightful. Avoid hype or buzzwords. No hashtags. They're lame. That's just a clean, clear prompt with the three elements that matter most. And for 80% of use cases, that's all you need.
Now, over time, you'll start thinking this way naturally. And with the longer or condensed method, you won't always follow a strict order. It may be mixed in together naturally. The key is to just think clearly about what you want, who it's for, and how it should sound. That mindset will help no matter what you're trying to create. Even with a framework, your output won't always be perfect, especially for longer workflows or complex problems. So from here, you'll evaluate, iterate, and optimize. So it's Rissco EIO. This is the part that turns a good output into a great one, and it helps you create reusable and reliable prompts over time. Well, start by reviewing what the AI gave you. But don't just read it, interrogate it. Ask the model questions like, "What assumptions did you make in this response? What might be missing based on the prompt? Is there anything that feels off or mismatched with the goal? Steelman the opposing view. You can even say, "Now critique your own output. Where did you miss the mark?" That's a fast way to surface blind spots or get a second take without needing to start from scratch. Asking it to critique itself can be super helpful.
Next, treat the AI like a creative partner. Don't accept the first draft unless it's perfect, and it usually isn't. Prompt it again with instructions like, "Now rewrite this to be more concise. Add a bit of humor. Try a more emotional tone. Give me three variations with different pacing." This helps you explore different angles without reinventing the prompt every time. Once you've got an output that works, the final step is to refine the prompt itself. Most people will have prompts they're reusing regularly for work or personal uses. Spending a little extra time optimizing that prompt can save you a lot of time down the road. So, cut any unnecessary words. Make the instruction as clear as possible. Turn background info into tighter context. Ask the model. Rewrite this prompt to be more concise and reliable while keeping the same intent. You'll end up with a leaner, sharper version that's easier to reuse and less likely to break when the topic shifts. This phase doesn't have to take long. Sometimes it's one to two quick tweaks, but this is what turns prompting into a repeatable system.
There's Riceco EIO. You don't need to become a prompt engineer. Just use a simple framework that helps you be clear about what you want and how you communicate it. The way you prompt shapes everything that comes after. Strong prompts mean faster workflows, fewer rewrites, more creative momentum, and better results consistently. And this is a system you can rely on to produce those results. Now, if you want to go deeper into learning AI, we have built a full course platform at Futuredia with over 500 lessons across 20 AI courses. You'll find full learning paths on ChatGpt, prompt engineering, automation, custom GPTs, video generation, coding with AI, and a lot more. It's all included in one subscription. You can get a 7-day free trial using the link in the description or check out this video with an entire road map on how to learn AI.