Master The Art of Context Engineering

All right, Creative Con, let's get real for a second. Everybody is talking about AI, but almost no one is talking about the thing that actually makes AI powerful. Context. Because here's the truth.

AI doesn't create magic out of thin air. It mirrors the quality of the context you feed it. Your inputs, your framing, your perspective, and mastering that. That's what separates the average user from the ones who are truly dangerous with this new technology.

Our next speaker knows this better than anyone. He's been deep in the eye trenches testing, building, and you know, breaking prompts to understand how to get human level nuance out of machine logic. Today, he's here to teach you the hidden skill every creative needs to master in the AI era, context engineering, and how to use it to generate smarter, sharper, more human creative. So, let's get ready to take some notes and get ready to level up our AI prowess.

Welcome to the Creative Con stage, Jimmy SGel. Hey everyone, my name is Jimmy Slaggel. Super excited for you guys to be tuning in and listening to uh this presentation on one of my favorite topics, which is context engineering. Um, quick background about me.

I've been in the AI and creative strategy space since about 2022. I I actually had co-founded a uh startup that was trying to use AI to generate UGC style ads back before chat GBT was even a thing. So I've been interested in the intersection of AI and creative strategy for man almost three uh yeah three years now. Uh which is crazy.

Um Alex Cooper and I also launched a uh program where we were teaching uh creative strategists how to use AI. We had over 300 different brands, I think 500 creative strategists go through that that program. Um, and this was by far and away everyone's favorite uh topic in that 8-week course. And so, super excited for you guys to learn a little more about context engineering.

If you've never heard of context engineering, uh don't worry. I'll be going over exactly what it is and describing it in um as much detail as I can. If you are super uh uh advanced, uh this will be valuable for you. If you are a basic Chad GBT user, this is going to be value val valuable for you and I will do my best to be able to present to both um of those audiences.

So, first just like looking at what context engineering is and I think Toby who is the CEO of uh Shopify does a great job of defining it which is really the art of providing all the context for the task to be plausibly solvable by the LLM. Now, an LLM is just like chatbt. uh that is essentially the the generic term for uh chatbt and claude and gemini um and all the other ones that that exist. Uh and and so really there's two kind of skills that are required to use AI well or AI well within creative strategy and that's context engineering and prompt engineering.

I think they're separate. Toby kind of uh indicates here that they're the same. Um but really we're going to be going through and showing you why context engineering is important and even what it is. So, don't worry.

Um, I just love this this tweet by him and I thought it was really valuable. Uh, so today's agenda, we're going to be looking and defining what context engineering is. Uh, we're going to be comparing an output with no context versus an output with context. Uh, and then I'm going to show you guys how you can actually start to build your context library.

Um, and so that is really what what we are going to be going over uh today. So, what exactly is context engineering? Um, before I answer that, uh, and this is going to be pretty creative strategy focused, but I want you to think through what makes a really good creative strategist. And I have some ideas here, um, to help you get started.

So, first, maybe you're someone that is just an art of the game, if you will. Uh, so you love going on YouTube, you love watching Alex's videos, um, you love watching Dra's videos. You're just trying to learn as much as you can from these these industry experts who talk about creative strategy and and learn from them who are in ad accounts that spend millions and millions of dollars every single year. Uh maybe you're someone that's just actively on foreplay.

You're looking at all the different uh uh types of ads that your competitors are running. You've got great swipe files. Um, but you're kind of just storing this rolodex of ads in your your head so that you have this understanding of knowing which ads uh different brands are testing and all the different formats that exist. Uh maybe you're someone that spends a ton of time in motion uh and and you're looking at really that breakdown of creative versus performance trying to figure out what actually uh influences and impacts the the performance of your different creatives.

um maybe you attend all their webinars and and you're just trying to learn as much as possible, but you're really just kind of that that in between of like, hey, I need to know the data and I need to understand the creative strategy piece as well. Maybe you're someone that again just loves the fundamentals of creative strategy. You love studying uh oglev. You love psychology.

You love just looking at um really what it is about humans that gets them to take action uh and to persuade them. So maybe you just love psychology. Uh and and maybe you're someone that just loves this idea of okay system systematizing how can we make the best creative ops system. So maybe you're someone uh uh that is trying to generate hundreds of ads every every single month um for for Facebook's uh Armageddon or whatever the heck that uh you know endless supply of creatives uh for the media buying is going to look like.

Um, but I think all of those things are are really like what context is. Everything that you take from um an ad that has performed well to something that you read in or learned from Oglev, like all of that is context. That chat GBT, it's not that it doesn't have that understanding. It just doesn't actually know when to use it and it's not going to use it uh uh out of random.

Um, so like I said, you know, I asked Chat GBT like, "Can you actually help me with creative strategy?" Um and and you you know Chachbt tells it or this might have been quad but they tell you uh themselves like the most valuable thing I can probably ask you or do is ask you the right questions to think about your process systematically. Without context I'm telling you chatbt knows nothing and and uh that is really this art of context engineering is is trying to take chatbt by itself uh in its current form to turn it into something valuable. And if we go back to Toby's um uh tweet, it's really this art of like what do we need to provide chat GBT with in order to get valuable outputs. And when you have the right context in any of your prompts, I'm telling you, uh you are going to get so much more value out of chat GBT and the the responses and ideas are going to be significantly better.

Uh and so that is what context engineering is. It is simply giving AI the right information in order to get a valuable response. Uh so we're going to dive into this a little deeper. Um but there are two types of context.

Uh the first one is going to be brand context and the second one is going to be domain context. So when it comes to brand context, that's the first thing. That's going to be everything about uh your brand. So uh these are just some examples.

This is not a a definitive list. There there can be many more that you add to this. There could probably be less. Um, but this is just hopefully to get your idea spinning.

So, first of all, products that you sell. Maybe you give it the breakdown of uh what your best performing or or top selling products are that you really want to focus on, why people might buy those. Um, and uh and just the overall like what you guys are maybe focused on when it comes to the products that you're trying to sell. The next one is is who buys and why.

Uh, the most important thing is the why. So, customer persona. So trying to get chat to BT or or just you know when we think of context um uh trying to document who is actually buying the product, why, what painoint is it solving, what barriers are they overcoming, what identities do they give themselves, all of these different things are really important. Uh the next one is is kind of your unique value proposition.

So what does the market look like? Where are you positioned? And what are your unique value propositions? And sometimes that's easy if you're like a problem solution brand.

If you're a lifestyle brand, uh this could maybe just be like what is that feeling or or that uh future state that you want the person um to to feel when they are buying. All of those are are are great for for unique value props. Um this one's really important, the the brand tone and just kind of the words used and not used. So existing uh and it ties into five and six, but like existing scripts that you guys have used for ads that have performed well, existing headlines that have performed well.

Um, really you want to document all of that. Uh, the next one is just like the ad formats that you've tested. Obviously, there are a ton. This could be UGC style, this could be founder, this could be VSSLs.

You just want to to document all of the different ad formats that you've tested and what has worked and why. Uh, and then finally, customer reviews are great. That is a gold mine of of information for the brand. Um, and so it being a uh AI being able to see kind of the the overall picture of your customer reviews is super valuable.

Um, the best way that I describe writing this and and we'll be able to show you some examples too, so don't worry too much. But assume that you hired an intern and you had to describe to this intern exactly uh uh the different parts. So like let's just say that you are going to create a contact stock on brand tone and words use and not use. Imagine you have an intern sitting next to you and you're just going to talk and you're going to explain the words that you use, the tone, examples and and show them everything like that is kind of what a context doc ends up um being.

Now domain context is a little bit different. Uh so domain context instead of it being you know for example uh uh what ads have worked in general maybe this is just a document outlining uh the best creative strategy for 2025. Or maybe instead of it being the the static ad and video ad formats that you guys have used and the performance of it, maybe this is just a list of all the different static ad formats and all the different video ad formats. I think Alex calls it an ad cookbook.

Um, this could be a document on how to actually create good headlines. So, how do you come up with good headlines? Like that alone is a contact stock. How do you write good scripts?

Like what is your process? Um, uh, maybe this is something that, uh, you saw a podcast with Barry and it was talking all about the the formula to create ugly ads or something along those lines. Um, you could take that podcast script uh, or transcript and use that as a contact stock. Um, you could also look at what your competitors are doing and take their creative strategy and that could be a contact stock for the model to know.

All of this chatbt has absolutely zero idea. I'm I'm I'm not joking. I'll show you an example. Um, Chatbt is really not great about knowing all of this.

Uh, and Chatbt is not good at knowing all of this. So, this is proof. Okay. So, I went into Claude.

I said, "Write a 30 secondond uh, script for direct response Facebook ad for the perfect gene.nyc. I even linked the URL so it won't get confused because the perfect gene can mean many different things." Um, do not give me the scenes or visuals. I only want the verbal script. Okay.

This is Claude. This is Opus. So this was their, you know, at the time most advanced model. Um, and and the perfect gene is a pretty big company, right?

Like it's not like they're they're like a six-figure brand. Like this is a company that is is doing quite well. Um, and this is this is how uh little Chat GBT actually knows about you and your company. And I do not want you to miss this point.

Okay. Uh, so here is the verbal script for the perfect gene that I came up with. Um, what is funny about this is The Perfect Gene, if you're not aware, is a jean brand for guys. I don't think they have ever targeted women.

Uh, and they don't have a women's line, but as you can see, you know, the fourth sentence in Chatbt says over 50,000 women have already made the switch. And right now, you can get 20% off your first pair. That is how little ChatBT knows about uh uh your um your company. So, I hope that is just eye opening for you.

Uh unless you're maybe Nike or Lulu or or a massive uh uh billion-dollar company, the odds are chatbt claw do not know that much about your brand. Um so I'm going to show you the difference that prompt and context engineering can make uh and and show you the the results. So here is a prompt that I came up with directly, you know, in a new chat, but right after this. So, same claw opus for um and I said write a 30 second uh uh script for a Facebook ad for the perfect gene.

Don't give me the visuals. Um first, you know, and this is where the art of prompt and context engineering is. First, I said, I want you to write one script for one customer segment. So, within this uh document, you can kind of see here there's three documents at the bottom.

That's the extra context that I gave chatbt, which I'll be able to show you. But one of them is just like the ICP analysis. So, who actually is uh the perfect gene going after? Uh, and who are they at their core?

So, what I'm telling Claude in this case is, hey, I want you to write a script for one customer segment per script. You need to look through that doc and essentially um choose one of them per script. Then I want you to uh read through the doc of how to create killer ads. This was a document about performance marketing um uh from Harry and it was just great.

like it's a 12page something like that 14page document on like this is how to create killer ads. Uh so I I told it I want you to use this as your frame of reference on how to create really good ads. Um you know when thinking of pain points and and the emotion that they feel just pick one emotion that you want to focus heavily on. Um and I gave it a quick example.

Finally I pasted a a video ads template. So all the different video ads that exist. So again like founder UGC ugly ads like all of these different things. UGC mashups.

Um, and I told it to pick one of those video ad formats uh for each of the different customer segments. Uh, there was a little more. I know this kind of got distorted, but go back. Uh, so after you write the script, go back and review them.

Analyze it deeply. Do we stop them with the hook? Is it actually compelling? Uh, do they feel something based on this?

I'm telling it it's first response is probably a four out of 10. Go back and make it a 10 out of 10. And then uh, rewrite them. And then at the bottom, like this is kind of where prompting context docs comes in is I just tell it like for context, here are the different types of video templates.

For context, here's how to make performance ads. For context, here are five examples of scripts that we've used in the past. Examples are are so key. Um so in this case, I pasted in um five different scripts that the perfect gene has used uh that have performed well just so it has some understanding.

And this is the the uh difference in the quality. So final rewritten scripts. Uh, so one is choosing the ugly single take UGC core emotion frustration to liberation is is kind of that emotional transition. And as you can see, like right away, you know, what pisses me off?

Like that's a pretty good hook. Way better than what was originally uh created. Spending 5 years building these legs just have every pair of jeans telling me I'm wrong. I'm so sick of having to choose.

Either my thighs go numb or I'm swimming in fabric at the waist. It's humiliating. Then my buddy shows up in the Perfect Gene Athletic Fit. Turns out they make jeans for guys who actually lift.

Like overall it's not perfect. Um, and maybe they want to actually run this, but nonetheless, it's so much better than what AI had originally come up with. Um, the next one is is kind of shame to dignity. So, so problem, agitation, storytelling is is the format that it chose.

Core emotion, shame to dignity. Every time I walk into a regular store, they look at me like I don't belong. So, think bigger guys, big and tall, uh, where they have specific stores for them. Um, so if you can imagine a bigger guy, uh, uh, you know, that's being vulnerable and open here.

Again, not great, but um, this is this idea that perfect jean goes up to size 50. Uh, same style, same washes, same price as everyone else. You start to get that emotional tie in. Again, it's not perfect u, and there's probably a lot of tweaking that you would do, but it's such a better start than just going into chatbt and saying, "Hey, write a script for me." And so that uh is is really like simplified in in you know 15 minutes the art of prompt and context engineering.

Um so now I'm going to spend the rest of the time just showing you how to create these documents because that is what is most important. I know you're probably sitting there like yeah that sounds great but how did you actually go through and create these documents? Um so a couple different tools and and resources that I want you to be aware of that I can show you. So, first of all, um if you do not use deep research, I strongly strongly strongly strongly recommend using deep research to help you with these these contact stocks.

The reason why is it has by far and away the longest output. So, you'll be able to see as I go through this how long the output is. Um but two, it will go and scrape all the relevant uh sites. And so, it just will have way more real-time information than what you know the the original cutoff date of chatbt was.

Um, so in this case, I said you need to go and find all the ICPS for Ollie, but more importantly, we need their psychographics. This is what we're looking for. So, this is where it gets really fun. Uh, there was a video that Sarah Leinger did with the operators um and and uh and this is just something that what I did is I ended up actually taking the transcript of this um I took the transcript of this video uh threw it into ChatgBT and essentially said like, "Hey, I need the summary of what this is." is so as you can see here uh what customer personas are why they matter the role of identities and emotional drivers why identities matter emotional drivers uh two big examples so giving examples within that so I just pasted that overview from chatbt uh and then I essentially said you know I I entered it it asks a follow-up question and now this is the document that I got defining all these customer personas and why they matter um so you can see here it will actually start to go through all his team has listen uh learned to listen for key emotions beyond why customers seek uh wellness supplements frustration and aversion.

Um a common feeling that traditional vitamins are unpleasant or daunting. Uh guilt and shame, nurturing love. So two key customer personas for Olly with examples. So Mia, the the millennial wellness seeker, everyday uh light hero.

So her profile, her goals and needs, her behavior. So, Maya discovered Olly on a Target run, drawn in by the the the packaging, emotions, and identity. Um, she often feels slightly guilty that she's not doing more. I should really be taking vitamins.

So, all of this uh Olivia, the onthe-go mom, family's chief wellness officer. So, all of this is is really like who uh AI is coming up with for these um uh specific ICP. So, what I can do is I can just copy. If I just wanted, you know, uh if I wanted this whole doc, I could copy this whole doc.

Um I could just come here, hit copy, and then I have this whole doc at my disposal um uh to be able to then put into chatbt or claude and say, "Hey, for context, here are my uh ICPS." So, that's one example. Use deep research. Uh take transcripts from YouTube videos if you need like a really good starting point. Um but uh but nonetheless that is how you can start to um uh create some ICPS without having to sit down for two hours and write them all out.

Um so that's how you can get your ICPS. The next one is uh writing ad scripts and visuals. So getting uh or sorry getting winning ad scripts and visuals. So there's a tool called Google AI Studio.

If you're unaware of Google AI Studio, what makes it so unique to every other AI tool is that it can actually um look at video files. So, in this case, I prompted it to say, "All right, you're a video analyzer agent. Your job is to extract a summary of the video. What is happening in enough detail that someone who is blind would still be able to understand what is happening if the video uh overview was just read to them." Now, what you have to see is that uh and and I'm not going to necessarily play this video um but you can see kind of like what the overall uh vibe of it is.

And you can see here this video is so this is the output. This video is a 40 second fast-paced ad for the perfect gene. It starts with the male narrator asking do jeans give you saggy diaper but visually there's a split screen. So this model Google AI studio is able to actually watch the footage.

So what you can do is just paste in all of your winning ads and and you know use a prompt of like hey describe this video u visually to me and what you can see is it's going to give you a summary. Now I can go and I can copy this summary uh so analysis the hook text the angle uh the product the time stamp or the time stamp the product was introduced I can copy this I can throw this into a Google doc I can throw this into a not notion page I'll do this maybe three four five more times just to have examples of some of our winning ads um but that's how you can go from like not just getting this the the transcript but for AI to actually see what's happening so that is a really easy and good way for you to start to create context docks around winning ads that you have. Obviously with images and static ads, you can use that within chatbt or claude as as they're able to view images, but um Google AI Studio Gemini 2.5 Pro is really the only one that can watch video files both visually and getting the transcript. Um so again you can go through uh you can copy the output here and you can paste that into claude to to start to get um to start to get just better outputs uh from from AI.

Um the next one I would say is is just like creative strategy principles. So this is again is a doc uh that I was talking about from Harry Delm. Um and this is how to make killer performance ads. Uh so the golden rules our ads must target emotions.

There must be a promise. There must be proof. there must be zero confusion. So all of these things are so helpful for AI to know.

So what you're going to do is is essentially just go to like chatbt or claude um and and what you're going to do is essentially just say like hey um you know so I would go back here. I would uh um copy in this doc. I would come back to this um and I could just paste this doc in. So now Claude has contacts.

Um, so let's just say I do that. And let's just say that I copy in um this uh this two key personas for Olly. Um, and I'll just copy a lot of this just so we can get uh get everything. Okay.

So, we're going to go back to Claude here. I'm going to paste that in. So, now Claude has actual context for these two things. I don't have examples of all these ads, but now what I can say um is like I need you to come up with ad ideas for Ali.

Um the only uh material I want you to use is what is within the how to make killer ads. Also, you only can target the key uh customer personas that we have listed here. Make sure that for each ad idea, you're choosing one of the ICPS and one of the emotions uh that they are experiencing. So, um I'm going to then and this isn't a great prompt.

If I wanted to like spend a lot more time, I I probably would, but just based on these two two docs, all you have to do now is hit enter. And as you can see, it's going to be just using this these two documents to come up with ideas. And I'm telling you, the outputs are going to be so much better than if you just used uh generic um chatbt. So, we'll see what it comes up with.

But that's actually how you use the contact docs. Uh just so you can see, you can paste it into a chat. Uh what gets really cool, and I'm probably not going to have enough time to do it, but uh there are things called cloud projects. And when you go through and create a cloud project, uh so here like this is a script writer.

Uh actually, let me go through and show you the one that we actually used. Uh so this is like context libraries if you will. So here add inspo types of static and video ad formats. Um cloud projects you're essentially just able to add in different files.

So either uploading from a device you can connect it to Google Drive. Uh you can add text context. Uh but I added here like Eugene Schwarz's five stages of awareness the types of video ads. How to make killer ads.

So that document winning ads for the perfect gene. Um how to create really good static. So, just a document of like here's how to create really good static ads, um, customer reviews, and, uh, yeah, I mean, that's just kind of the the overview of, um, of what it looks like. And I don't think that that, uh, chat ended up working for whatever reason.

Maybe Claude is is down. Of course, it happens on the demo. But, nonetheless, hopefully that was helpful. I'm telling you, context engineering is going to completely change the the quality of outputs that you get.

Um, and I hope this was just a good video uh to introduce you to this world and and get your brain spinning on how you can start to create um uh really good contact stocks for AI. Uh if you have any questions, make sure to hit me up. I'm always down to riff on on this topic. And if you have any other ideas for us, uh let me know.

But thanks for tuning in. Um and uh would love to connect with you all