Big tech just announced plans to spend trillions ($600B from Meta, $250B from Google, $115B from OpenAI) on AI infrastructure over the next 5 years. Not billions. Trillions. They're literally in a cage match over who gets to power the future of… everything.

With that amount of investment, AI is going to continue to evolve at a rapid pace. Add to that the fact that current capabilities are already quite good, and you’re fast in a situation where every brand and agency on the planet will be using AI to plan content, analyze what works, and create the actual assets. Even Zuckerberg is indicating this. The brands pretending this isn't happening are the same ones who said "we don't need a website" in 1995.

If you're not already embracing this shift, you're not being cautious—you're falling behind. No, AI didn’t put that emdash there, I did. I wanted your attention, because I needed to make sure you get how important this is!

The good news is, if you haven’t been doing much with AI content, you picked the right newsletter to fix that.

We've all seen the garbage. Subject lines that sound like they came from a bot farm in 2003. Product descriptions that use "revolutionary" three times in two sentences. Images with too many hands, fingers, or other appendages.

Here's why this happens: everyone treats AI like a magic wand instead of what it actually is… your best employee ever, but one you forgot to onboard.

Think about it. You wouldn't hire a copywriter and immediately ask them to write your Black Friday campaign without telling them anything about your brand. But that's exactly what people do with AI. They throw "write me an email about our new product" at ChatGPT and wonder why it sounds like corporate vomit.

The secret isn't "prompt engineering" (90% of that advice is lead magnet bait anyway). It's context engineering. Give the AI what it needs to understand your brand, your customers, and the specific job you're hiring it to do.

Great AI results are 80% context, 20% prompt. Everything else is just noise from people selling courses.

After helping 200+ brands and agencies over the last 8 months (yes, I counted) directly within Raleon and personally, here’s what separates those that are getting killer AI content from those getting robot soup.

First, we’re going to I’ll quickly cover what context engineering is. Then, I’ll outline the 6 kinds of basic context assets that will provide the fastest value. Lastly, I’ll guide you through how to setup a Claude project with this context. We’re going to cover this both for text, and images.

Everyone has heard of prompt engineering at this point. Here’s the quick secret from those on the inside: 90% of what you’ve heard about prompt engineering is a lie.

Great results are 80% context engineering, 20% prompt engineering. In fact, more often than not when your AI results suck, it’s because you didn’t give the AI enough context. Not because you didn’t ask nicely.

Context engineering, put simply, is giving the AI the information (context) it needs to be able to complete the task. Kind of like how you would onboard an employee.

Now, here’s the 6 basic context’s you need for great text generation, whether that’s analysis, copy for ads, emails, landing pages, you name it.

Not your mission statement that sounds like it was written by a committee of consultants. Your actual personality. How you talk when no one's watching. What makes you different beyond "we care about quality" (everyone says that).

If I was True Classic, for instance, I would do something like:

Specific examples, not vague descriptions. Don't say "we're fun and approachable." Show it.

Beyond the features and benefits everyone else lists. The real problem you solve. The transformation you deliver. The job your product gets hired to do emotionally, not just functionally.

You would repeat the above for each product. You could go lighter on the details above if you have a lot of products.

Real customer language from reviews, support tickets, and actual conversations. Not demographic data like "millennial moms aged 25-35." How they actually describe their problems and your solutions in their own words.

Your best-converting copy. The email subject lines that made your phone blow up with Stripe notifications. The ad copy that scaled to six figures. The product descriptions that turned browsers into buyers.

To generate this for Meta, I would:

What bombed and why. This teaches AI what to avoid and helps it understand your boundaries. "This email got a 0.3% open rate because it sounded like we were announcing a corporate merger."

Do the same thing we mentioned above, but then ask for the bottom 3 performers.

You may be wondering why the bottom 3? It’s because AI has a problem with focusing on the bad.

You frequently want to use “semantic negative prompts” where you describe what you want to see explicitly as opposed to too much of what you don’t want to see. Reason being, AI gets confused and can sometimes include what you said not to do purely because you mentioned it.

I’m not generally an enabler, but because I am nice, I will show you how to do this in ChatGPT. Truth be told, you should do this in Claude, not ChatGPT.

If you’re using ChatGPT, that’s part of why your results suck. Especially since the GPT-5 update (scroll further for why in the “What AI Models Should You Use?” section).

You could create a Custom GPT around this, but to be honest, I am not sure if it’s actually worth the extra overhead for this type of use case.

Step 1: Click “New Project” on the left when logged into ChatGPT

Step 2: Fill in the project name you want, in this case, I put in “Example DTC Project”. I’d add “for ads” or “for emails” based on the performance data you put in.

Step 3: Add your files by clicking “Add files”

Step 4: Profit! That’s it. Now just start a new chat in your project.

You’ll notice I underlined ChatGPT 5 Thinking. I wouldn’t use any other model from ChatGPT right now for any legitimate work.

You should do this instead of ChatGPT. There’s a reason why Microsoft is eyeing a partnership with Anthropic (the creator of Claude) afterall. 

Thankfully, the process is about the same.

Step 1: Goto Projects, and click “Create Project”

Step 2: Now, we’re going to add our 6 essential context files like before.

Step 4: That’s it! Now you can chat with your custom Claude Project.

Notice what I didn't say about context: I didn’t say dump your entire brand book and every document you ever wrote into the AI.

I get this question all the time. We've seen brands try this approach and it actually performs worse, not better. AI gets confused by contradictory information and irrelevant details just like humans do.

A key insight for you here that everyone is currently missing. AI tools are now at a stage where different tools excel at different jobs.

In other words, using ChatGPT for everything is like using a hammer for brain surgery. As a result, here’s your AI model cheat sheet for September 2025.

Now that you understand how to create a project, and what models to use, it’s time to dive into how to bring this all together.

Below is the process that we see top brands and agencies using right now to get great AI content results.

Step 1: Analysis in ChatGPT 

Upload your top-performing content. Ask it to identify what makes it work. Not just "it's engaging" but the specific patterns, word choices, and structural elements.

Step 2: Chat with Your New Claude Project

This is where the magic happens. Not because your prompt is awesome, but because you built great context into this project. Now you can give it the results of the analysis from step 1, and run a prompt like:

Yes, I did just throw some pro prompt-fu in there at the end. Use it and thank me later.

Step 3: Iterate with Claude 

It’s rare that a chat conversation is 1-shot and done. Go back and forth a few times to get to the right result. If your context is good, it shouldn’t take more than a few back and forth.

Step 4: Create with Nano Banana 

I couldn’t say this model name without laughing during my presentation. The problem is it’s “official name” (Gemini 2.5 Flash Image) was worse.

In either case, take the results from Step 3, and turn your content suggestions into real static ads, or email infographics in Nano Banana.  

Multi-turn editing beats hoping for perfection on attempt number one. If you’re not sure what I am talking about here, wait for our deep dive into image context next week.

Yes, this is manual. Yes, it takes actual effort. I know this myth busts the "magical unicorn" of AI being sold on X that automates their entire marketing department. But you come here for the reality of what’s possible. You still need people in the loop.

What I will say is if this feels like a lot to you, it is. That’s part of why we created Raleon. We literally automate everything I just walked you through from context engineering to the workflow process. Done. As close to a unicorn as you can get right now.

But whether you use Raleon or do it manually, the context and process works.

The brands and agencies that are really crushing it with AI aren’t the ones looking for a simple shortcut. They're building systematic approaches to consistently create content that doesn't suck.

Remember what I said at the beginning? Every brand on the planet will be using AI for content creation. We’re already well on our way to this being true.

You now have everything you need to either join that wave or get crushed by it.

You've got the 6 essentials for building context that actually works. You know which tools excel at different jobs. You have the exact process to create content that doesn't suck.

The brands starting now are like those early into Meta ads. The compounded learnings and experience they will gain will make it nearly impossible to catch up to later. While everyone else is still trying to find the perfect prompt, you'll have a content engine that understands your brand better than most employees.

Now that you have this information, here’s 3 things you should do right now.

Start with the above, you’re going to see awesome results.

If you liked this for text based content, you’ll love next week. I’m going to run through the same process but for VISUAL content (images). I’ll also include a sweet mini-app I built that you’re going to love (and will save you a ton of time).

And that's it for this week's edition. Stop treating AI like a slot machine and start treating it like your most important hire. Give it context, show it examples, and watch your content go from robot vomit to revenue driver.

Just don't expect it to happen with a single "ultimate prompt" you found on Twitter.