Every marketer has tried ChatGPT. Most get generic output back. The difference between generic AI copy and copy that ships is almost never the model — it's the prompt.
Here are the six patterns we use daily.
1. Anchor with role + audience
Before you ask, tell the model who it is and who it's writing for. This one-liner triples the specificity of everything downstream.
2. Show, don't tell
Models imitate. If you paste two examples of the style you want before the ask, the output will match — far better than adjectives ever will.
3. Ask for the frame, then the fill
Don't ask for a finished blog post. Ask for an outline. Approve it. Then fill each section separately. Quality goes up 3x, revisions drop.
Step 1: "Give me 5 possible angles for a blog about X."
Step 2: "For angle #2, give me a 6-section outline."
Step 3: "Write section 2 of the outline, ~200 words, in the style of these examples: […]"4. Constraints are creative
"Write about X" → mediocre. "Write about X in exactly 3 paragraphs, each starting with a rhetorical question" → memorable.
5. Explicit disqualification
Tell the model what NOT to do. Every LLM has default habits (bulleted lists, hedging, "as an AI…"). Ban them explicitly.
6. Iterate in place
Don't restart the conversation. Say "tighter", "less hedgy", "more concrete" — the model has your whole context and can improve much faster than starting over.
The meta-tip
Save your best prompts as templates. A team's competitive advantage in AI-driven marketing is 90% institutional prompt knowledge — not model access.
FAQ
Frequently asked questions
Which model should I use in 2025?
The frontier keeps shifting. Try the same 3 prompts on the top-3 available models each quarter and pick the winner. Don't over-commit.
Are AI-detection tools a problem?
For SEO, Google cares about quality, not authorship. A well-edited AI draft that's genuinely useful ranks fine. A raw AI dump does not.