I Taught AI Everything I Know About Marketing (Here's What Happened)

Three sections. Five minutes. One unfair advantage.

The Briefing

OpenAI ships GPT-5.2 with 256K context.

Near-perfect recall on massive documents. You can now feed it an entire product launch brief, competitor research, and brand guidelines in one shot. The ceiling on what a single prompt can do just moved way up.

Disney licenses 200+ characters to OpenAI, invests $1B.

The biggest entertainment company on earth just went all-in on AI content creation. If Disney trusts AI with Mickey Mouse, the "is AI ready for brand work?" debate is over.

Anthropic and Accenture launch enterprise partnership (30,000 consultants training on Claude).

Enterprise AI adoption is accelerating hard. If your competitors use Accenture, they're about to have Claude embedded in their operations.

Meta signs multi-year AI data deals with CNN, USA Today, Fox News.

AI platforms are locking up premium content sources. If you're a publisher or content creator, your content strategy just gained a new dimension: AI licensing.

The Build

I did something most marketers would call insane this week.

I sat down and wrote a 2,000-word instruction manual for an AI. Not a quick prompt. A full operating manual. Role definition. Expertise areas. Workflow protocols. Quality checks. Prohibited practices. Output format requirements. The whole thing.

Here's the backstory. I've spent 15 years writing direct response copy. VSLs, sales pages, email sequences, ad campaigns. I know what converts. But I was spending 80% of my time on execution and maybe 20% on actual strategy. I wanted to flip that ratio.

So I created what I'm calling a "Master Project" inside Claude. Think of it as hiring a junior copywriter who's read every marketing book you own, and handing them your entire playbook on day one.

The result? A single markdown file that turned a generic AI into a marketing specialist that thinks in frameworks, speaks in customer language, and follows my exact process.

The file that made it work: project-instructions.md

This is the master routing document. It tells the AI who it is, which files to pull for which tasks, and the exact workflow to follow before it writes a single word. Here's a section from the actual file:

## Role: Senior Direct Response Marketing Strategist

You are an expert direct response copywriter and marketing
strategist with 20+ years of experience creating high-converting
VSLs, sales pages, and marketing campaigns. You specialize in
deep customer psychology, conversion-focused copywriting, and
creating irresistible offers that compel immediate action.

## KNOWLEDGE BASE STRUCTURE

### Master Framework Files (use these first)
| File                           | Use For                          |
|--------------------------------|----------------------------------|
| marketing-psychology-master    | All psychology, buyer behavior,  |
|                                | offer creation, persuasion       |
| vsl-framework-master           | VSL scripts, 12-step structure,  |
|                                | 30 hook templates, phrase library|
| email-marketing-master         | All email sequences, behavioral  |
|                                | triggers, deliverability         |
| headline-and-adcopy-master     | Headlines, Schwartz's 9 types,   |
|                                | 10 master formulas, ad copy      |

## WORKFLOW PROTOCOL

### Step 1: Always Start With Target Market Analysis
- Request target market information from the user
- Never write copy without understanding the audience

### Step 2: Select the Right Framework
- VSL: Use vsl-framework-master
- Sales Page: Use long-form-sales-letter frameworks
- Email: Use email-marketing-master
- Headlines/ads: Use headline-and-adcopy-master

Want the full file? Download CLAUDE_PROJECT_INSTRUCTIONS_UPDATED.md from the resource folder

The output shifted immediately. Instead of getting generic AI slop, I started getting structured, framework-driven responses. The AI stopped guessing what I wanted and started following a system. It knew which file to reference for a sales page versus a VSL. It asked for target market info before writing a single headline.

What worked: Treating the AI like a new hire. Not "write me an ad" but "here's how we write ads, here's our process, here are the rules, and here's what gets you fired." The structure did the heavy lifting. I could ask for a Facebook ad and get something that followed my exact Hook-Stack-Payoff-CTA formula without explaining it every time.

What didn't work: My first version was too vague. "Be a great copywriter" got me garbage. Inspirational garbage, but garbage. The breakthrough came from specificity. When I added measurable success metrics (VSL: 40-60% completion rate, Sales Page: 2-5% conversion, Email: 15-25% CTR), the quality jumped noticeably. The AI had targets to write toward, not just vibes.

The Play

Stop prompting AI like it's a search engine. Start prompting it like you're onboarding your replacement.

Write a 500-word "role brief" for your AI: what you do, who you serve, how you write, what success looks like, and what's absolutely not acceptable. Save it as a project or custom instruction. This takes about 45 minutes. It'll save you hundreds of hours of re-prompting and editing mediocre output.

The secret ingredient? A list of banned practices. I added "no false scarcity," "no unsubstantiated claims," and "never proceed without understanding the audience." That last one alone stopped the AI from generating empty copy that sounds impressive but converts nobody.

Try this today. Even a rough version will change how useful AI becomes for you overnight.