8 AI Prompting Tips for Digital Writers — Ship30for30
Nicolas Cole’s framework for getting usable writing outputs from AI tools (ChatGPT, Claude, Gemini). The core argument: bad outputs are a prompting problem, not an AI problem.
The 8 Tips
-
Use Claude Projects as a context library. Create per-format projects (social, newsletter, blog). Add finished pieces so the context library converges on your voice over time. More context = better output.
-
Break tasks into singular prompts. Don’t cram brainstorm + outline + draft into one prompt. Distinguish “singular prompts” (one task, 500-1500 words) from “modular prompts” (a sequence of singular prompts chained to produce a larger work like a book).
-
Name every format you train AI on. E.g., “Format 1: Paragraph Style” with attributes and examples. Naming forces clarity and gives AI a handle to reference later.
-
Use objective, not subjective, language. “Write really great content” is subjective. “Generate thought leadership short-form social content, 280 characters or less, no hashtags, with 3 bulleted items” is objective. AI is pattern recognition — specificity drives quality.
-
Reinforce instructions with positive attributes. Don’t just say “write social content.” Add: concise language, economical phrasing, strong opinions, alternating sentence length for rhythm. Bake in frameworks from books you’ve read.
-
Use examples that mirror instructions exactly. If your rules say “declarative opener, 3-5 bullets, one-sentence insight,” every example must follow that structure. Mismatched examples confuse the model.
-
Add delivery instructions. Tell AI how to present output: format, length, labels, number of variants. Most people skip this step and get poorly formatted results.
-
Iterate with AI on your prompts. Ask “Why did you do that? What could I add to improve this prompt?” Treat yourself as a Writing Prompt Engineer: write, feed, review, iterate.
RDCO Mapping
Tips 1-3 directly apply to Sanity Check production — building a Claude Project with past issues as context, using singular prompts per section, and naming the newsletter format explicitly. The objective-language principle (Tip 4) aligns with the CopyThat framework’s emphasis on concrete over abstract.