The first few times I used Claude for client work, I honestly wasn’t that impressed.
I’d spend five minutes writing a prompt, hit enter and watch it generate something that looked... mediocre.
Not terrible.
Not amazing.
Just painfully average.
It sounded polished enough that you couldn’t call it bad, but generic enough that you would never send it to a client without rewriting half of it.
So I’d start editing, rewriting, changing around the hook, then rewriting the ending.
Twenty minutes later, I’d realize I probably could have written the whole thing myself.
The problem wasn’t Claude.
It was the way I was talking to it.
That realization sent me down a rabbit hole. Instead of reading another “50 ChatGPT prompts” blog post, I went straight to Anthropic’s own technical documentation.. the guides their engineers publish for developers building on top of Claude.
Buried inside those docs was one sentence that explained almost every frustrating AI experience I’d had.
“Think of Claude as a brilliant but new employee who lacks context on your norms and workflows.”
Everything clicked.
Claude wasn’t producing generic work because it wasn’t capable of anything better. It was producing generic work because I was giving it almost nothing to work with.
No context, standards or examples.
No explanation of what success actually looked like.
I was handing an employee a sticky note and expecting agency-level work.
Once I understood that, the quality of my output changed almost overnight.
What Vague Gets You
Before I show you what actually works, I want to show you the gap.. because seeing it once is worth more than any explanation.
Here is how most people prompt Claude for client work:
Write a Facebook ad for a weight loss supplement.
Here is what comes back. You have seen this before:
“Struggling to lose weight? Our premium formula is scientifically formulated to support your weight loss journey. With natural ingredients and zero compromises, you can finally achieve the results you’ve been working toward. Try it today, your transformation starts now. Shop now →”
It is not wrong. It is just nothing. No one reads that and stops scrolling. No client sees that and says “H*LL YES, this is what I needed.”
Now here is a prompt that applies what Anthropic’s docs actually recommend:
You are a Meta ads specialist with experience running campaigns for health and wellness brands. You understand what the Facebook algorithm rewards and what gets people to stop scrolling.
Client: a weight loss supplement targeting women 40 to 55 who have tried other programs and are skeptical of anything that sounds like a promise.
Write three ad variations. Each needs a hook that leads with a pain point this specific woman would recognize.. not a product benefit. Body: 2 to 3 sentences that connect her experience to what this product actually does. CTA that is low-pressure.
Before finishing: verify none of these hooks could appear in any other weight loss ad. If they could, rewrite them.
Here is the kind of output that comes back:
“I did everything right for three years. Counted every calorie. Walked five miles a day. Lost twelve pounds. Gained fifteen back. If that sounds familiar, this is for you.”
Same model..different result. The only thing that changed was the prompt.
The Three Things That Actually Change the Output
Anthropic’s docs outline a set of prompting principles that apply directly to the kind of work digital service providers do. I am going to walk through the three that matter most, with real before and after examples for each one.
Comments
Nothing yet. Say the first thing.
Sign in to join the conversation.