At the latest Claude Code Anonymous meetup in Brighton, someone gave a lightning talk on prompts for "humanising" AI writing. They showed long skill files and detailed style guides, then they showed that you can get most of the same effect with a single line: "Use Orwell's 6 rules for writing."
The model absorbed Politics and the English Language during pretraining, along with decades of writers referencing and applying those rules. It already knows Orwell's rules, you just need to point at them.
The pattern
When I'm doing product work with Claude, I include the phrase "use Marty Cagan's four product risks." Four words that activate a whole framework for evaluating product decisions: value risk, usability risk, feasibility risk, viability risk. The model knows what each one means, how they interact, and how practitioners apply them. I don't need to define any of it.
I've heard people use "TDD red-green-refactor" the same way. (I have reservations about that particular one, but that's a digression for another post.)
These named references act as decompression keys: compact instructions that unpack into rich, structured knowledge because the model already did the hard work of learning it.
Why it works
No one has a complete picture of how LLMs store and retrieve knowledge. But the broad mechanics are understood enough to explain why this pattern is so effective.
During pretraining, the model compressed millions of documents as patterns, relationships, and structures encoded in its weights. Well-known frameworks got reinforced thousands of times across textbooks, blog posts, conference talks, and practical guides.
A named framework is that key. When you say "Orwell's 6 rules," you're calling on something that already exists in the model's weights. The name acts as a decompression point, activating a cluster of related knowledge that would take hundreds of lines to replicate explicitly.
What makes a good decompression key?
Not all references are equal. A reliable decompression key tends to be:
- Broadly documented. Written about in books, articles, courses, and blog posts. The more sources, the stronger the signal during pretraining.
- Consistently applied. Practitioners use it the same way. There's a shared understanding of what the framework actually means in practice.
- Freely available. If the material is mostly behind paywalls, fewer copies made it into the training corpus.
- Unambiguous. The name refers to one thing. No competing definitions, no overloaded terms.
An unreliable reference tends to be the opposite: niche frameworks with a small following, contested ideas where different authors use the same name to mean different things, paywalled content with limited circulation, or names that collide with something else entirely.
"Orwell's 6 rules" scores high on every dimension. It's one of the most referenced pieces of writing advice in the English language. "Cagan's four product risks" is solid too. Widely taught, consistently defined, and distinct enough that the model won't confuse it with something else.
If a thousand practitioners could independently describe the framework and mostly agree, the model can too.
Beyond prompt tricks
Before spelling out a framework in detail, ask whether a named reference would do the same job. Often it will, and you get something more reliable. The model's internal representation of a well-known framework is richer and more consistent than most handwritten descriptions of the same thing.
It's not a replacement for explicit instructions when you need something specific or novel. But for established frameworks with broad consensus, you're better off pointing at the knowledge than trying to rewrite it.