WEBVTT 00:00:00.000 --> 00:00:02.580 You ever ask your super smart AI for something 00:00:02.580 --> 00:00:05.700 amazing and what you get back sounds like, I 00:00:05.700 --> 00:00:08.339 don't know, a screeching cat? Uh -huh. It feels 00:00:08.339 --> 00:00:10.320 like someone handed you this incredible instrument, 00:00:10.439 --> 00:00:12.900 a Stradivarius, but you only know how to play, 00:00:13.000 --> 00:00:15.740 like, beginner scale. Yeah. All that power, but 00:00:15.740 --> 00:00:18.660 the results. Kind of disappointing. Yeah, and 00:00:18.660 --> 00:00:21.320 that AI sameness you see everywhere online, where 00:00:21.320 --> 00:00:23.980 everything just sounds flat. Exactly. We're going 00:00:23.980 --> 00:00:26.239 to dive into why that happens and, more importantly, 00:00:26.460 --> 00:00:30.620 how we can fix it. Welcome back to the Deep Dive. 00:00:30.719 --> 00:00:33.399 Today, we are unpacking a framework that could 00:00:33.399 --> 00:00:36.140 really change how you talk to AI. It's called 00:00:36.140 --> 00:00:39.679 Race Eco. Think of it like the sheet music for 00:00:39.679 --> 00:00:42.000 your AI. It helps turn that generic noise into 00:00:42.000 --> 00:00:45.200 something, well, symphonic. No more bland cookie 00:00:45.200 --> 00:00:47.460 cutter outputs. We'll go through each part. Role, 00:00:47.460 --> 00:00:49.820 instruction, context, examples, constraints, 00:00:50.000 --> 00:00:53.000 and output format. Six key things. And then we'll 00:00:53.000 --> 00:00:55.079 touch on a simpler version for everyday stuff 00:00:55.079 --> 00:00:57.100 and a way to really refine your prompts over 00:00:57.100 --> 00:00:59.350 time. Yeah, the goal here is to help you shift 00:00:59.350 --> 00:01:01.170 from just, you know, throwing prompts out there 00:01:01.170 --> 00:01:03.369 to actually being an architect of the AI's response. 00:01:03.810 --> 00:01:06.390 Crafting stuff that's genuinely exceptional, 00:01:06.810 --> 00:01:10.629 tailored every time. Let's make some music. So 00:01:10.629 --> 00:01:13.109 a lot of us, myself included sometimes, we just 00:01:13.109 --> 00:01:16.390 talk to AI like we're chatting, right? Expecting 00:01:16.390 --> 00:01:19.129 magic. But that often just leads to that, that 00:01:19.129 --> 00:01:21.909 sameness, that flood of content that sounds like 00:01:21.909 --> 00:01:24.659 a robot wrote it. Right. If your prompt is just 00:01:24.659 --> 00:01:27.459 write an article about plants, well... You're 00:01:27.459 --> 00:01:29.519 basically just adding to the noise floor online. 00:01:29.920 --> 00:01:32.579 Yeah. The AI is powerful, yeah, but it's definitely 00:01:32.579 --> 00:01:34.560 not reading your mind. And that's where this 00:01:34.560 --> 00:01:37.700 RACIQ framework comes in. It's a more structured 00:01:37.700 --> 00:01:40.799 way, a systematic set of instructions. Like that 00:01:40.799 --> 00:01:43.480 sheet music idea. It guides the AI past the generic 00:01:43.480 --> 00:01:45.739 stuff. Helps it produce something, you know, 00:01:45.739 --> 00:01:48.280 actually beautiful, exceptional. It's a six -step 00:01:48.280 --> 00:01:50.760 formula. Kind of like a chef getting everything 00:01:50.760 --> 00:01:53.420 prepped, the mise en place before cooking. For 00:01:53.420 --> 00:01:55.819 perfect AI output. You might not need all six 00:01:55.819 --> 00:01:58.219 steps for every single prompt. That's true. But 00:01:58.219 --> 00:02:01.159 knowing them, that's the key to really leveling 00:02:01.159 --> 00:02:03.319 up what you create. Okay. So let's start at the 00:02:03.319 --> 00:02:05.239 beginning. What's that very first ingredient 00:02:05.239 --> 00:02:07.260 if you want a unique perspective from the AI? 00:02:07.500 --> 00:02:10.419 It really comes down to giving the AI a specific 00:02:10.419 --> 00:02:13.939 persona, a character or point of view to adopt. 00:02:14.219 --> 00:02:17.539 Right. R for role. So assigning a role. It's 00:02:17.539 --> 00:02:19.740 like casting a really good method actor in a 00:02:19.740 --> 00:02:22.719 film. Exactly. instantly changes the whole vibe. 00:02:22.800 --> 00:02:25.800 The tone, the words it chooses, its perspective, 00:02:26.120 --> 00:02:29.039 everything shifts. Imagine telling your AI, okay, 00:02:29.120 --> 00:02:31.840 now you're a board -certified sleep doctor. Right. 00:02:31.939 --> 00:02:34.259 And then for the next prompt, telling it, okay, 00:02:34.340 --> 00:02:36.900 switch gears, now you're an absolutely exhausted 00:02:36.900 --> 00:02:39.319 parent trying to give advice to other parents. 00:02:39.479 --> 00:02:42.080 The difference is going to be massive. So let's 00:02:42.080 --> 00:02:44.139 take that sleep advice example. If you just ask, 00:02:44.159 --> 00:02:46.780 give me advice on how to get better sleep, what 00:02:46.780 --> 00:02:48.580 do you get? Probably something pretty generic, 00:02:48.699 --> 00:02:50.879 like a basic Google search result. Technically 00:02:50.879 --> 00:02:53.819 okay, but totally uninspired. But if you add 00:02:53.819 --> 00:02:56.379 the role, you are a board -certified sleep doctor 00:02:56.379 --> 00:02:59.159 with 20 years of clinical experience. Give me 00:02:59.159 --> 00:03:02.400 advice. Then, suddenly, the output changes. It 00:03:02.400 --> 00:03:04.539 gets concise. It's evidence -based. It probably 00:03:04.539 --> 00:03:07.219 uses more clinical terms. Feels like a professional 00:03:07.219 --> 00:03:10.240 consult. Or the other way. You are a sleep -deprived 00:03:10.240 --> 00:03:12.439 parent writing a blog post for other exhausted 00:03:12.439 --> 00:03:15.479 parents. And boom. It's conversational, it's 00:03:15.479 --> 00:03:18.219 empathetic, maybe even a little funny, full of 00:03:18.219 --> 00:03:20.259 practical tips that actually resonate with parents 00:03:20.259 --> 00:03:22.740 who are in the trenches. Wow. So the basic task, 00:03:23.000 --> 00:03:25.659 give sleep advice, didn't change at all. Nope. 00:03:25.979 --> 00:03:28.740 But the role fundamentally shifted how the AI 00:03:28.740 --> 00:03:31.580 approached it, how it thought, how it spoke, 00:03:31.780 --> 00:03:34.020 not just what it said. That's pretty powerful. 00:03:34.400 --> 00:03:37.199 Just defining who the AI should be changes the 00:03:37.199 --> 00:03:40.039 nuance that much. And here's a pro tip. Try the 00:03:40.039 --> 00:03:42.539 expert panel prompt. You actually ask the AI 00:03:42.539 --> 00:03:46.430 to embody. Multiple roles at the same time. Maybe 00:03:46.430 --> 00:03:48.889 conflicting roles. Like a skeptical professor, 00:03:49.289 --> 00:03:52.110 a super optimistic VC, and maybe a pragmatic 00:03:52.110 --> 00:03:54.729 engineer all arguing about AI consciousness. 00:03:55.189 --> 00:03:57.849 Exactly. It forces the AI to explore different 00:03:57.849 --> 00:04:00.830 angles, generate a more nuanced discussion, moves 00:04:00.830 --> 00:04:03.650 beyond just one simple answer. Great for brainstorming 00:04:03.650 --> 00:04:05.949 complex topics. Okay, so we've cast our actor 00:04:05.949 --> 00:04:07.849 with the role. Now, how do we make sure they 00:04:07.849 --> 00:04:09.689 know exactly what scene they're in, what to do? 00:04:09.810 --> 00:04:11.969 That comes down to giving it really clear, precise, 00:04:12.189 --> 00:04:16.660 unambiguous commands. Got it. I is for instruction. 00:04:16.920 --> 00:04:19.899 The core action you want the AI to take. And 00:04:19.899 --> 00:04:23.500 vagueness here is your enemy. Vague instructions. 00:04:23.660 --> 00:04:26.920 You get vague, generic results every time. You 00:04:26.920 --> 00:04:29.120 got to think like a, well, like a good commander. 00:04:29.160 --> 00:04:32.120 You said, go take that hill isn't enough. Right. 00:04:32.180 --> 00:04:34.680 It needs to be. Take that hill. Flank from the 00:04:34.680 --> 00:04:37.040 east. Use available cover. Establish defense 00:04:37.040 --> 00:04:41.180 by sundown. See the difference. Precision. This 00:04:41.180 --> 00:04:43.240 is where the specificity principle comes in. 00:04:43.319 --> 00:04:45.779 You need to ditch those fuzzy words like engaging 00:04:45.779 --> 00:04:48.720 or cool or professional. Yeah, those are like 00:04:48.720 --> 00:04:50.899 empty calories for an AI. It doesn't know what 00:04:50.899 --> 00:04:53.259 you really mean by engaging. It's just guessing. 00:04:53.459 --> 00:04:55.879 Let's say you want a YouTube short script. A 00:04:55.879 --> 00:04:58.459 weak instruction is, write me an engaging YouTube 00:04:58.459 --> 00:05:01.420 short about haunting tips. The AI just throws 00:05:01.420 --> 00:05:03.360 darts in the dark hoping to hit engaging. But 00:05:03.360 --> 00:05:05.240 a strong instruction, maybe something like... 00:05:05.389 --> 00:05:08.189 Write a 60 -second YouTube short script on prompting 00:05:08.189 --> 00:05:10.870 tips. Start with a strong curiosity gap hook. 00:05:11.129 --> 00:05:13.610 Include a scroll -stopping visual anchor in the 00:05:13.610 --> 00:05:16.050 first three seconds. Now you've given it concrete, 00:05:16.209 --> 00:05:19.339 objective things to aim for. You translated your 00:05:19.339 --> 00:05:22.240 subjective desire engaging into specific parameters. 00:05:22.620 --> 00:05:25.639 So instead of just asking for creative, you might 00:05:25.639 --> 00:05:28.720 say, write in the style of a 1940s hard -boiled 00:05:28.720 --> 00:05:31.740 detective novel. Exactly. Precision is totally 00:05:31.740 --> 00:05:33.680 power when you're talking to AI. And there's 00:05:33.680 --> 00:05:36.019 this technique, chain of thought prompting. Ah, 00:05:36.180 --> 00:05:38.160 yeah. That's a really useful one. You basically 00:05:38.160 --> 00:05:40.639 tell the AI, hey, before you give me the answer, 00:05:40.759 --> 00:05:43.480 show me your work. Think step by step. Okay. 00:05:44.560 --> 00:05:47.980 So for anything involving complex logic or math 00:05:47.980 --> 00:05:51.100 problems or reasoning, telling it to outline 00:05:51.100 --> 00:05:53.980 your entire chain of thought step by step makes 00:05:53.980 --> 00:05:56.180 a huge difference in accuracy. And you can see 00:05:56.180 --> 00:05:58.480 how it got the answer. More transparent. Precisely. 00:05:58.519 --> 00:06:01.839 All right. So the AI knows who to be, role, and 00:06:01.839 --> 00:06:04.819 what to do. Instruction. What else does it need 00:06:04.819 --> 00:06:07.279 to really understand the situation, the world 00:06:07.279 --> 00:06:10.480 around the task? It needs the backstory. The 00:06:10.480 --> 00:06:12.579 context. You need to give it the details about 00:06:12.579 --> 00:06:14.480 the audience, the purpose, the background, the 00:06:14.480 --> 00:06:17.459 tone, its whole worldview for this task. See, 00:06:17.560 --> 00:06:19.759 for context, this is the whole setting the story 00:06:19.759 --> 00:06:21.959 so far. You said this is the step people skip 00:06:21.959 --> 00:06:24.459 most often. Yeah, I think so. And it's like asking 00:06:24.459 --> 00:06:27.319 an actor to perform without telling them anything 00:06:27.319 --> 00:06:29.199 about the play or their character's history. 00:06:29.339 --> 00:06:31.300 The result is going to be flat, give it motivation, 00:06:31.500 --> 00:06:34.379 the full scene. So you mentioned four pillars 00:06:34.379 --> 00:06:38.410 for great context. First. Audience. Who's this 00:06:38.410 --> 00:06:41.870 for? Beginners. Experts. Loyal customers. Second, 00:06:42.029 --> 00:06:44.230 background. Where does this live? Is it for a 00:06:44.230 --> 00:06:47.629 business blog? Social media? A specific scenario. 00:06:47.949 --> 00:06:51.269 Okay. Audience. Background. Third, purpose. Why 00:06:51.269 --> 00:06:52.769 are we even doing this? To teach something. To 00:06:52.769 --> 00:06:55.110 persuade. To sell. Just to entertain. Yeah. And 00:06:55.110 --> 00:06:58.129 fourth, tone. How should it sound? Witty. Super 00:06:58.129 --> 00:07:01.120 formal. Empathetic. Casual. Let's take an example. 00:07:01.240 --> 00:07:04.079 Write a 500 -word blog post about AI video tools 00:07:04.079 --> 00:07:05.959 without context. You get a generic overview. 00:07:06.160 --> 00:07:08.379 Probably not very useful. Yeah. Just a list of 00:07:08.379 --> 00:07:10.759 tools, maybe. But if we add context, audience, 00:07:10.980 --> 00:07:13.500 non -tech, small business owners, background. 00:07:13.740 --> 00:07:16.240 It's for a marketing agency's blog. Purpose dot. 00:07:16.379 --> 00:07:18.360 Get them excited about turning podcasts into 00:07:18.360 --> 00:07:21.220 YouTube shorts easily. Tone. Simple, practical, 00:07:21.399 --> 00:07:24.480 inspiring. See, now it's not just a generic list. 00:07:24.959 --> 00:07:27.360 It becomes a highly targeted, persuasive piece 00:07:27.360 --> 00:07:30.060 that actually speaks to those business owners 00:07:30.060 --> 00:07:32.459 about their needs. Totally different outcome. 00:07:32.720 --> 00:07:34.160 And you shouldn't be afraid to give it a lot 00:07:34.160 --> 00:07:37.300 of context for complex tasks, like a brain dump. 00:07:37.560 --> 00:07:40.060 Absolutely not. AI can process pages of text 00:07:40.060 --> 00:07:42.420 almost instantly. Better to give it too much 00:07:42.420 --> 00:07:44.680 background than too little. Over -deliver on 00:07:44.680 --> 00:07:46.800 the context. What about for ongoing projects? 00:07:46.860 --> 00:07:49.240 Like if you're always writing for the same brand 00:07:49.240 --> 00:07:51.360 voice? That's where a persona document comes 00:07:51.360 --> 00:07:53.759 in super handy. It's like a master file. With 00:07:53.759 --> 00:07:55.939 all your standard context, your audience profile, 00:07:56.240 --> 00:07:58.120 your brand voice guidelines, your company mission, 00:07:58.220 --> 00:08:00.040 stuff like that. So you create this document 00:08:00.040 --> 00:08:03.399 once. And then for future prompts, you just tell 00:08:03.399 --> 00:08:06.720 the AI, hey, first, read and fully internalize 00:08:06.720 --> 00:08:08.959 this persona document. Everything you generate 00:08:08.959 --> 00:08:11.259 needs to align with it. It's a huge time saver. 00:08:11.279 --> 00:08:13.980 It keeps everything consistent. Okay. Roll, instruction, 00:08:14.339 --> 00:08:17.079 context. We're building a really detailed prompt. 00:08:17.660 --> 00:08:19.980 Now, how do we make sure the quality or the style 00:08:19.980 --> 00:08:22.160 of the output matches what we want? Something 00:08:22.160 --> 00:08:25.360 really specific. Ah, that's where examples come 00:08:25.360 --> 00:08:28.500 in. You show, don't just tell. Provide concrete 00:08:28.500 --> 00:08:31.560 examples of excellent output. Mid -roll sponsor 00:08:31.560 --> 00:08:36.019 read insert point. Okay, we're back. We were 00:08:36.019 --> 00:08:38.379 just talking about ensuring the AI output matches 00:08:38.379 --> 00:08:41.720 a specific quality or style. You said E for examples 00:08:41.720 --> 00:08:44.779 is key. Yeah, absolutely. If you need a very 00:08:44.779 --> 00:08:47.279 specific structure or a particular tone or a 00:08:47.279 --> 00:08:50.179 certain format, showing the AI really high quality 00:08:50.179 --> 00:08:53.100 examples, that often has the biggest, most immediate 00:08:53.100 --> 00:08:55.440 impact on the output. It's like training an apprentice, 00:08:55.539 --> 00:08:57.779 right? You don't just describe the perfect report, 00:08:57.919 --> 00:09:00.080 you show them one. Exactly. It's way faster. 00:09:00.220 --> 00:09:02.580 This technique, by the way, is often called few 00:09:02.580 --> 00:09:05.440 shot prompting. You're giving the AI a few shots, 00:09:05.500 --> 00:09:08.250 a few examples to learn from directly. Let's 00:09:08.250 --> 00:09:10.110 think about a newsletter. If you just say, write 00:09:10.110 --> 00:09:12.549 a newsletter about the top three AI news stories 00:09:12.549 --> 00:09:14.690 this week. You'll likely get a blob of text, 00:09:14.809 --> 00:09:17.929 probably unstructured, maybe kind of dry, generic. 00:09:18.230 --> 00:09:19.870 But if you provide an example from a previous 00:09:19.870 --> 00:09:22.429 week, you say, act as my newsletter co -writer. 00:09:22.549 --> 00:09:25.289 You need to follow the exact same tone, style, 00:09:25.450 --> 00:09:27.929 and format as this example below. And then you 00:09:27.929 --> 00:09:31.070 paste in your best one. Now the AI gets it. It 00:09:31.070 --> 00:09:33.230 sees your custom sections, your specific way 00:09:33.230 --> 00:09:35.529 of phrasing things, your intro style, your outro. 00:09:35.850 --> 00:09:38.590 It learns not just what topics to cover, but 00:09:38.590 --> 00:09:40.889 how you actually write the newsletter. The quality 00:09:40.889 --> 00:09:43.789 jump must be significant. Oh, it's huge. And 00:09:43.789 --> 00:09:45.389 this isn't just for writing. Remember, you can 00:09:45.389 --> 00:09:47.850 use examples for anything structured. Getting 00:09:47.850 --> 00:09:51.169 code output in the right JSON format. Complex 00:09:51.169 --> 00:09:53.610 logic patterns. Show it an example. That makes 00:09:53.610 --> 00:09:56.360 sense. And you had a pro -level tip for sales 00:09:56.360 --> 00:09:57.980 and marketing here. Yeah, this one's clever. 00:09:58.240 --> 00:10:01.919 Try giving the AI an anonymized email or inquiry 00:10:01.919 --> 00:10:04.279 you receive from your absolute ideal customer, 00:10:04.460 --> 00:10:07.299 the perfect lead. Okay. The AI analyzes their 00:10:07.299 --> 00:10:09.559 language, the specific pain points they mentioned, 00:10:09.659 --> 00:10:12.679 their professional tone, everything. Then you 00:10:12.679 --> 00:10:15.279 ask it to craft new outreach messages or marketing 00:10:15.279 --> 00:10:17.980 copy based on that analysis. So it learns how 00:10:17.980 --> 00:10:20.059 your best customers actually think and talk. 00:10:20.559 --> 00:10:23.120 Precisely. It helps you create messages that 00:10:23.120 --> 00:10:25.559 really resonate deeply with that specific customer 00:10:25.559 --> 00:10:28.600 profile. Super powerful for targeting. We're 00:10:28.600 --> 00:10:31.220 building quite a prompt here with RACE, role, 00:10:31.539 --> 00:10:34.080 instruction, context, examples, what's needed 00:10:34.080 --> 00:10:37.000 to keep the AI on track. Stop it from just drifting 00:10:37.000 --> 00:10:39.500 off into generic AI speak. That's where constraints 00:10:39.500 --> 00:10:42.320 come in. You need to set clear rules, boundaries, 00:10:42.519 --> 00:10:45.039 the rules of the game for the output. The second 00:10:45.039 --> 00:10:48.259 C, constraints. Okay, so. If context is the worldview, 00:10:48.580 --> 00:10:51.299 constraints are the hard and fast rules. Exactly. 00:10:51.379 --> 00:10:54.559 Think about any game. It needs rules to be fun, 00:10:54.700 --> 00:10:57.480 challenging, interesting. Without rules, it's 00:10:57.480 --> 00:11:00.840 just chaos. Your AI needs those boundaries, too, 00:11:00.940 --> 00:11:03.960 to force it to produce focused, high -quality, 00:11:04.059 --> 00:11:06.320 professional output. Because by default, these 00:11:06.320 --> 00:11:09.419 models can be a bit verbose, a bit hedging. Yeah. 00:11:09.480 --> 00:11:11.480 They often default to being wordy, playing it 00:11:11.480 --> 00:11:13.620 safe. Constraints are how you tame that. You 00:11:13.620 --> 00:11:16.059 force it to be concise, maybe more opinionated, 00:11:16.139 --> 00:11:18.419 stay aligned with exactly what you need. So what 00:11:18.419 --> 00:11:21.200 are some examples of effective constraints? Well, 00:11:21.240 --> 00:11:23.429 they can be structural. Your response must be 00:11:23.429 --> 00:11:26.269 under 100 words or include exactly one direct 00:11:26.269 --> 00:11:28.629 quote and one statistic. They can be stylistic. 00:11:29.049 --> 00:11:32.350 Do not use corporate buzzwords like synergy or 00:11:32.350 --> 00:11:36.110 leverage or the tone must be warm and conversational, 00:11:36.169 --> 00:11:39.330 not formal. Or even persona -based, like the 00:11:39.330 --> 00:11:41.490 output needs to sound like it came from a competent 00:11:41.490 --> 00:11:44.149 founder, not an academic. Precisely. Or even 00:11:44.149 --> 00:11:46.970 do not use any hashtags. They are lame. You're 00:11:46.970 --> 00:11:49.070 giving it direct commands to shape the final 00:11:49.070 --> 00:11:51.860 output. And for managing these over time, especially 00:11:51.860 --> 00:11:54.740 on big projects. You can add your core constraints 00:11:54.740 --> 00:11:57.580 right into the AI's custom instructions or settings 00:11:57.580 --> 00:11:59.980 if the platform supports it. It's like establishing 00:11:59.980 --> 00:12:02.679 the official rulebook for that AI on that project 00:12:02.679 --> 00:12:05.399 ensures consistency. Now, you mentioned a negative 00:12:05.399 --> 00:12:08.120 constraint technique. What's that? This is really 00:12:08.120 --> 00:12:10.500 useful when you notice the AI keeps making the 00:12:10.500 --> 00:12:13.019 same mistakes or falling into bad habits. You 00:12:13.019 --> 00:12:16.059 explicitly tell it what not to do. Oh, okay. 00:12:16.139 --> 00:12:18.620 So you'd literally say something like, do not 00:12:18.620 --> 00:12:21.100 use exclamation points. Do not use marketing 00:12:21.100 --> 00:12:23.820 buzzwords. Do not make the response longer than 00:12:23.820 --> 00:12:27.179 150 words. Exactly. It's very direct. It helps 00:12:27.179 --> 00:12:29.679 quickly steer the AI away from those default, 00:12:29.879 --> 00:12:32.879 often annoying AI sounding patterns and towards 00:12:32.879 --> 00:12:35.029 something more unique. I have to admit, I still 00:12:35.029 --> 00:12:37.610 wrestle with prompt drift myself sometimes. The 00:12:37.610 --> 00:12:40.190 AI just seems to forget the earlier instructions 00:12:40.190 --> 00:12:42.610 or constraints over a longer conversation. Oh, 00:12:42.629 --> 00:12:44.970 totally. It happens. Negative constraints are 00:12:44.970 --> 00:12:46.889 a great tool to help tighten that up when you 00:12:46.889 --> 00:12:49.169 see it starting to drift. So we've guided the 00:12:49.169 --> 00:12:52.470 AI to create some amazing content using RICEC, 00:12:52.730 --> 00:12:55.870 role, instruction, context, examples, constraints. 00:12:56.230 --> 00:12:58.250 How do we make sure the final presentation is 00:12:58.250 --> 00:13:01.429 useful, immediately usable? Right. That final 00:13:01.429 --> 00:13:04.580 step is crucial. You need to specify the exact 00:13:04.580 --> 00:13:08.360 structure, the presentation, how you want the 00:13:08.360 --> 00:13:10.759 information packaged. Oh, for output format. 00:13:10.799 --> 00:13:13.179 This is about how the AI structures the response. 00:13:13.500 --> 00:13:15.539 Yeah. Think of it like having a really fine wine. 00:13:16.139 --> 00:13:18.700 Just getting the wine, the content isn't enough. 00:13:18.860 --> 00:13:21.279 A messy wall of text. That's like dumping the 00:13:21.279 --> 00:13:23.519 wine straight onto the table. Oh, okay. The output 00:13:23.519 --> 00:13:26.000 comment is the crystal glass. The perfect vessel. 00:13:26.120 --> 00:13:28.700 It presents the content cleanly, makes it organized, 00:13:28.919 --> 00:13:31.480 easy to understand, and importantly, ready to 00:13:31.480 --> 00:13:33.899 use. And it's not just about making it look nice, 00:13:33.919 --> 00:13:36.720 is it? Not at all. It's about making it actionable, 00:13:36.879 --> 00:13:40.500 sometimes even machine readable. A well -structured 00:13:40.500 --> 00:13:43.840 output like a neat table or a JSON object can 00:13:43.840 --> 00:13:46.360 often be plugged directly into other software, 00:13:46.480 --> 00:13:49.779 other automations. Saves a ton of manual reformatting. 00:13:50.029 --> 00:13:51.970 What are some common formats we can ask for? 00:13:52.090 --> 00:13:55.549 Oh, lots. Simple tables are great for comparisons. 00:13:56.049 --> 00:13:58.370 JSON is fantastic if you need structured data 00:13:58.370 --> 00:14:00.730 for apps or APIs. It's kind of the universal 00:14:00.730 --> 00:14:03.169 language for that. Markdown is good for creating 00:14:03.169 --> 00:14:06.850 web -ready text or documents with basic formatting 00:14:06.850 --> 00:14:09.389 like headings and lists. You can even ask for 00:14:09.389 --> 00:14:12.559 a Twitter thread structure. Absolutely. Format 00:14:12.559 --> 00:14:14.360 this as a five -tweet thread. It'll break it 00:14:14.360 --> 00:14:16.320 down accordingly. Let's take that software comparison 00:14:16.320 --> 00:14:19.379 idea again without specifying format. You get 00:14:19.379 --> 00:14:22.100 dense paragraphs. Hard to compare tool A versus 00:14:22.100 --> 00:14:24.840 tool B versus tool C. But if you add the constraint, 00:14:25.139 --> 00:14:27.419 present your answer as a three -column markdown 00:14:27.419 --> 00:14:30.700 table. Columns, tool name, key features, ideal 00:14:30.700 --> 00:14:33.320 use case. Instantly usable. Perfectly formatted. 00:14:33.399 --> 00:14:35.919 Easy to scan. You get the information you need 00:14:35.919 --> 00:14:38.379 at a glance. That single line about format makes 00:14:38.379 --> 00:14:41.549 all the difference. Okay, here's a thought. Could 00:14:41.549 --> 00:14:44.029 you ask for the same information in multiple 00:14:44.029 --> 00:14:48.009 formats in one go, like a chain of formats? Whoa. 00:14:48.269 --> 00:14:50.830 Yes, you absolutely can. That's a really efficient 00:14:50.830 --> 00:14:53.649 technique. So I could ask for, say, a summary 00:14:53.649 --> 00:14:56.649 of an article and request it as a LinkedIn post 00:14:56.649 --> 00:14:59.389 and as a five point bullet list for Twitter and 00:14:59.389 --> 00:15:02.690 maybe that three column comparison table. all 00:15:02.690 --> 00:15:05.750 from the same core content in one single response 00:15:05.750 --> 00:15:08.629 you got it that is a passive productivity hack 00:15:08.629 --> 00:15:11.429 especially for content repurposing generate a 00:15:11.429 --> 00:15:13.750 whole week's worth of social media posts formatted 00:15:13.750 --> 00:15:16.649 for different platforms with just one smart prompt 00:15:16.649 --> 00:15:19.289 that's seriously impressive okay so after going 00:15:19.289 --> 00:15:22.470 through all six rice eco steps role instruction 00:15:22.470 --> 00:15:26.230 context examples constraints output format What 00:15:26.230 --> 00:15:27.929 does a really top -tier prompt actually look 00:15:27.929 --> 00:15:29.850 like when it all comes together? It looks like 00:15:29.850 --> 00:15:32.769 a precise, highly structured blueprint, something 00:15:32.769 --> 00:15:35.110 that consistently yields customized, actionable, 00:15:35.309 --> 00:15:38.049 and genuinely valuable results, not just generic 00:15:38.049 --> 00:15:40.049 text. Let's try that real estate example. The 00:15:40.049 --> 00:15:42.909 bad prompt was just, how can I use AI and automation 00:15:42.909 --> 00:15:45.490 in my real estate business? Yeah, super vague. 00:15:45.629 --> 00:15:48.009 You'll get a useless, generic answer that doesn't 00:15:48.009 --> 00:15:50.299 help your specific business at all. But a Rice 00:15:50.299 --> 00:15:52.600 Eco -enhanced prompt would be much more detailed. 00:15:52.799 --> 00:15:55.460 Right. You define the role. You are a business 00:15:55.460 --> 00:15:57.639 growth strategist specializing in real estate 00:15:57.639 --> 00:16:01.519 tech. The instruction. Identify the top three 00:16:01.519 --> 00:16:04.240 AI opportunities for my specific business. Create 00:16:04.240 --> 00:16:07.159 a prioritized action plan focusing on time savings 00:16:07.159 --> 00:16:10.059 and revenue generation. Okay, then add context, 00:16:10.299 --> 00:16:13.399 details about the agency size, location, current 00:16:13.399 --> 00:16:16.000 challenges, specific goals, maybe the tech staff. 00:16:16.139 --> 00:16:18.419 Then maybe examples. Here's an example of a lead 00:16:18.419 --> 00:16:20.440 qualification automation that worked well for 00:16:20.440 --> 00:16:23.700 a similar agency. Add constraints. The solutions 00:16:23.700 --> 00:16:26.120 must be implementable by a non -technical person 00:16:26.120 --> 00:16:29.360 within a $500 monthly budget and require less 00:16:29.360 --> 00:16:32.100 than 10 hours setup time. And finally, the output 00:16:32.100 --> 00:16:34.730 format. Present this as an AI and automation 00:16:34.730 --> 00:16:37.309 playbook with clear sections for opportunity, 00:16:37.669 --> 00:16:40.429 action steps, estimated time saved, and required 00:16:40.429 --> 00:16:43.250 budget. You see the difference. The output from 00:16:43.250 --> 00:16:44.769 that prompt is going to be completely customized. 00:16:45.029 --> 00:16:47.330 It addresses the specific market, the budget, 00:16:47.389 --> 00:16:49.509 the time limits, the unique pain points. And 00:16:49.509 --> 00:16:51.389 you mentioned it could even project things like 00:16:51.389 --> 00:16:54.230 8, 12 hours of weekly time savings. Yeah, because 00:16:54.230 --> 00:16:56.429 you gave it the context and constraints needed 00:16:56.429 --> 00:16:59.149 to make those kinds of specific relevant calculations. 00:16:59.429 --> 00:17:01.990 It's not generic advice anymore. It's a tailored 00:17:01.990 --> 00:17:05.150 strategy. Now, that full Ristoko framework seems 00:17:05.150 --> 00:17:08.049 incredibly powerful for these big, complex tasks. 00:17:08.329 --> 00:17:11.250 But you also mentioned it might be overkill for 00:17:11.250 --> 00:17:14.450 simpler everyday prompts. That's right. For probably 00:17:14.450 --> 00:17:17.269 80 % of the things you ask an AI to do day to 00:17:17.269 --> 00:17:19.890 day, you don't necessarily need all six steps 00:17:19.890 --> 00:17:22.289 every single time. You can get fantastic results 00:17:22.289 --> 00:17:25.130 focusing on what I call the ICC method. ICC, 00:17:25.309 --> 00:17:28.390 instruction, context, and constraints. Exactly. 00:17:28.450 --> 00:17:31.289 Think of it as the 80 -20 of prompting. These 00:17:31.289 --> 00:17:33.019 three are often the most... crucial ingredients 00:17:33.019 --> 00:17:35.240 for getting good results quickly so the must 00:17:35.240 --> 00:17:38.140 -haves a clear instruction what exactly should 00:17:38.140 --> 00:17:41.279 it do essential context who is this for why does 00:17:41.279 --> 00:17:44.119 it matter and clear constraints the key boundaries 00:17:44.119 --> 00:17:46.539 or rules nail those three and you'll already 00:17:46.539 --> 00:17:49.200 be way ahead of most generic prompts let's try 00:17:49.200 --> 00:17:51.960 that twitter example again with just icc the 00:17:51.960 --> 00:17:54.339 weak prompt was write a twitter post about chat 00:17:54.339 --> 00:17:57.980 gpt tips generic forgettable output guaranteed 00:17:57.980 --> 00:18:01.869 but a strong icc prompt might be Instruction. 00:18:02.450 --> 00:18:04.750 Write a Twitter thread. Give me five practical 00:18:04.750 --> 00:18:07.509 lessons learned from using chat GPT daily. Okay. 00:18:07.769 --> 00:18:10.569 Context. My audience is indie creators and solo 00:18:10.569 --> 00:18:13.230 entrepreneurs. They want practical AI tips, not 00:18:13.230 --> 00:18:16.210 hype. Got it. And constraints. Keep the tone 00:18:16.210 --> 00:18:18.670 clear, non -technical, and casually insightful. 00:18:19.029 --> 00:18:21.930 No marketing buzzwords or hype. No hashtags, 00:18:22.049 --> 00:18:26.029 please. Perfect. That ICC prompt is clean, clear, 00:18:26.230 --> 00:18:29.029 and gives the AI enough direction to produce 00:18:29.029 --> 00:18:31.869 something much, much better than the vague version. 00:18:32.069 --> 00:18:35.170 It hits the core requirements. So Rye Seco for 00:18:35.170 --> 00:18:38.190 complex stuff, ICC for the daily grind. Makes 00:18:38.190 --> 00:18:40.490 sense. But even with a great prompt, the first 00:18:40.490 --> 00:18:42.470 output isn't always perfect, right? You said 00:18:42.470 --> 00:18:44.490 the real magic is often in the follow -up. Absolutely. 00:18:44.869 --> 00:18:47.029 Crafting the perfect prompt often doesn't happen 00:18:47.029 --> 00:18:48.890 on the first try. It's an iterative process. 00:18:49.210 --> 00:18:51.730 That's where the EIO process comes in. Evaluate, 00:18:51.849 --> 00:18:54.400 iterate. Write and optimize. E -I -O. Okay, first 00:18:54.400 --> 00:18:57.140 E is for evaluate. So don't just accept the first 00:18:57.140 --> 00:18:59.599 thing the AI spits out. Right. Put on your detective 00:18:59.599 --> 00:19:02.059 hat. Really interrogate the results. Don't just 00:19:02.059 --> 00:19:04.940 passively accept it. Ask clarifying questions. 00:19:05.559 --> 00:19:08.519 Like what? Ask things like, what assumptions 00:19:08.519 --> 00:19:10.920 did you make when generating this? Or what context 00:19:10.920 --> 00:19:13.319 might be missing here? And my personal favorite 00:19:13.319 --> 00:19:16.759 power move. Now critique your own output. Be 00:19:16.759 --> 00:19:20.130 brutally honest about its flaws. Whoa! Asking 00:19:20.130 --> 00:19:22.829 the AI to critique itself. Yeah, it's surprisingly 00:19:22.829 --> 00:19:25.849 effective. It forces the AI to perform self -analysis, 00:19:25.970 --> 00:19:29.049 often revealing weaknesses or areas for improvement 00:19:29.049 --> 00:19:31.410 without you having to figure it all out yourself. 00:19:31.789 --> 00:19:34.430 Okay, so after evaluating, I is for iterate. 00:19:34.950 --> 00:19:36.950 This is the back and forth. Exactly. This is 00:19:36.950 --> 00:19:38.769 the creative partnership part. You collaborate 00:19:38.769 --> 00:19:41.690 with the AI, refine the output. Don't be afraid 00:19:41.690 --> 00:19:43.500 to be a demanding director. So you'd say things 00:19:43.500 --> 00:19:46.279 like, OK, rewrite this paragraph to be more concise. 00:19:46.380 --> 00:19:48.819 Or can you add a bit more humor here? Or give 00:19:48.819 --> 00:19:51.200 me three different versions of that opening sentence. 00:19:51.559 --> 00:19:53.740 Yep. It's this rapid conversational iteration 00:19:53.740 --> 00:19:55.859 that helps you discover new angles, nail the 00:19:55.859 --> 00:19:58.880 tone, and really perfect the message. And finally, 00:19:59.019 --> 00:20:02.380 O is for optimize. What does that involve? Once 00:20:02.380 --> 00:20:04.700 you've iterated and arrived at an output that's 00:20:04.700 --> 00:20:08.069 perfect, don't just stop there. optimize the 00:20:08.069 --> 00:20:10.430 original prompt itself, incorporate everything 00:20:10.430 --> 00:20:12.809 you learned during the evaluation and iteration 00:20:12.809 --> 00:20:15.509 phases. Ah, so you're improving the recipe for 00:20:15.509 --> 00:20:18.309 next time, making a reusable tool out of that 00:20:18.309 --> 00:20:20.809 perfect result. Exactly. You're creating the 00:20:20.809 --> 00:20:23.230 reusable mold from your perfect sculpture. And 00:20:23.230 --> 00:20:25.089 you can even ask the AI for help with that step 00:20:25.089 --> 00:20:28.069 too. For sure. You can literally say, okay, analyze 00:20:28.069 --> 00:20:31.380 our entire conversation, especially my... Corrections 00:20:31.380 --> 00:20:34.160 and requests. Now rewrite my very first prompt 00:20:34.160 --> 00:20:36.799 so that it would produce this final perfect output 00:20:36.799 --> 00:20:39.400 directly on the first try. That's brilliant. 00:20:39.460 --> 00:20:41.700 Turning a successful one -off interaction into 00:20:41.700 --> 00:20:45.299 a reliable, reusable asset. So when we pull all 00:20:45.299 --> 00:20:49.359 this together, RISCO, ICC, EIO, what's the really 00:20:49.359 --> 00:20:51.640 big takeaway here about how we should be thinking 00:20:51.640 --> 00:20:54.220 about AI, our role in it all? I think the biggest 00:20:54.220 --> 00:20:56.599 shift is moving away from just being a passive 00:20:56.599 --> 00:20:59.460 prompter. Someone just asking basic questions. 00:20:59.640 --> 00:21:01.819 And becoming. Becoming a thoughtful architect, 00:21:02.099 --> 00:21:04.859 an architect of the AI's interaction, its reasoning, 00:21:04.920 --> 00:21:07.640 its output. The Riceco framework, then, it's 00:21:07.640 --> 00:21:10.019 not just about technical tricks for writing prompts. 00:21:10.099 --> 00:21:12.759 It's more fundamental, a mindset shift. Exactly. 00:21:12.759 --> 00:21:16.079 It's about moving from making vague wishes to 00:21:16.079 --> 00:21:19.000 providing a clear, detailed blueprint for the 00:21:19.000 --> 00:21:21.619 AI to follow. And the benefits are faster workflows, 00:21:22.019 --> 00:21:25.509 fewer revisions. and consistently more professional, 00:21:25.750 --> 00:21:28.930 more reliable, more useful results. It all boils 00:21:28.930 --> 00:21:31.430 down to communicating with the AI using clarity, 00:21:31.609 --> 00:21:34.630 structure, and precision. So for everyone listening, 00:21:34.829 --> 00:21:37.069 the next time you sit down to work with an AI, 00:21:37.349 --> 00:21:40.329 think ICC for those quick daily tasks, instruction 00:21:40.329 --> 00:21:43.160 context constraints. Get those right first. And 00:21:43.160 --> 00:21:44.980 for the bigger, more complex projects, bring 00:21:44.980 --> 00:21:48.059 out the full Rice Eco framework. Role, instruction, 00:21:48.440 --> 00:21:51.220 context, examples, constraints, output format. 00:21:51.420 --> 00:21:53.960 Master this way of thinking, this structured 00:21:53.960 --> 00:21:56.259 approach, and you really will unlock the true 00:21:56.259 --> 00:21:58.720 potential of AI. It becomes less of a magic box 00:21:58.720 --> 00:22:01.119 and more of an indispensable partner. Because 00:22:01.119 --> 00:22:03.299 ultimately, as we've kind of unpacked today, 00:22:03.440 --> 00:22:07.119 every truly great AI output. It really does start 00:22:07.119 --> 00:22:09.099 with a great prompt, a well -architected prompt. 00:22:09.339 --> 00:22:11.359 Couldn't agree more. So maybe something to think 00:22:11.359 --> 00:22:13.960 about. What possibilities does this kind of structured 00:22:13.960 --> 00:22:17.400 prompting open up for your work, your creativity? 00:22:17.839 --> 00:22:19.920 Worth considering as you go out and architect 00:22:19.920 --> 00:22:22.960 your AI future. Thanks for diving deep with us 00:22:22.960 --> 00:22:24.099 today. Until next time.