The owner of a 12-person agency built her Context Stack on a Tuesday afternoon. It took her 25 minutes. By Wednesday morning she'd produced a client content calendar, a launch brief, and a five-email nurture sequence — all in her agency's voice, all client-ready, all from the same one-line prompt.
Her note to me after: "The brief is better than what I usually write manually."
This is what 25 minutes of context actually looks like.
Tuesday at 2:14 PM.
She opened a blank document. I gave her six prompts to answer:
Who is the client you're writing for? Two sentences. Real persona, not a marketing fiction.
What does the agency sound like? Three sentences someone on her team would say in a meeting. The phrasings the agency uses. The ones it doesn't.
What does "good" look like for this client? Paste one example of writing she'd happily send. Paste one example she'd reject.
What are the constraints? Channel, length, deadline, anything fixed.
What does this client not want? The anti-pattern list. The claims the client avoids. The topics off-limits. The phrasings that get her flagged in review.
What's the QA gate before something ships? Two checks. The two she actually runs, not the ones a process doc says she should run.
She wrote six paragraphs. Total length: 740 words.
By 2:39 PM, she had her Context Stack. Twenty-five minutes start to finish.
Wednesday at 8:50 AM.
She opened ChatGPT and pasted the document. Then she ran the world's shortest prompt:
"Read the context document. Build a 4-week content calendar for the [client] webinar series, then write the launch brief."That was the prompt. Eight words plus the placeholder.
The output came back in her voice. Specific to the client. The launch brief opened with the client's actual buyer pain, not "in today's competitive landscape." The calendar referenced the channels the client actually uses. The CTAs followed the format her client had approved for the last three launches.
She edited two paragraphs.
She told me later it was the first time in eleven months of using ChatGPT that her first output felt closer to ship-ready than her manually written drafts usually do.
Same model. Same task. Different door.
Here's the part that matters: she didn't change models. Didn't upgrade to Plus. Didn't switch to Claude. Didn't enable any plugins.
Same ChatGPT. Same task. Different door in. The 25-minute document was the door.
Two months earlier, she'd tried building this exact campaign with what she called "the best prompt I've ever written" — a 350-word brief generator she'd refined four times. The output came back generic. She rewrote 70% of it. The whole exercise took three hours.
This time, the prompt was eight words. The output needed two edits. The exercise took 22 minutes including the prompt.
The difference wasn't smarter. The difference was what the model knew before the prompt arrived.
What the document actually contained.
Six sections, in this order:
Persona (the client's customer, written so the model can answer "who am I writing to?")
Voice (the agency's voice, with three real-text examples — not adjectives like "professional yet approachable")
"Good" examples (one shippable, one rejected, both annotated)
Constraints (length, channel, deadline, anything fixed)
Anti-patterns (the language to avoid, the claims off-limits)
QA gate (the two checks she runs before anything ships)
That's the architecture. Six layers, one read, every prompt after that one-line.
The fix isn't a longer prompt. It's a shorter one with the right document behind it.
Most of the "I can't get good output from AI" frustration this year is a structural problem dressed up as a prompting problem. The model isn't refusing to do good work. It just doesn't know what good means for you. Twenty-five minutes of writing down what good means is the upstream fix.
Every prompt after that one runs against the same context. The voice stays consistent. The structure holds. The output stops drifting. The 9 PM rewrite goes away.
This is what people mean when they say AI "compounds" — not that the model gets smarter, but that the same context document keeps paying back across every interaction. One build, every prompt after.
The complete six-layer framework, including the Campaign Strategy Brief template I use with enterprise clients.
Not a course. Not a prompt pack. The actual system.
See you next week.
— Chris
