One Anthropic Researcher's Prompt Changed How I Use AI Forever. Here's the Exact Template.
Most prompts ask AI to explain things. The best ones ask it to show you something instead.
That distinction sounds cosmetic. It isn't. It changes what the model generates, how you process it, and — more importantly — whether it actually sticks.
I came across this idea while watching an interview with Amanda Askell — a philosopher and researcher at Anthropic whose work sits at the intersection of AI alignment and what you might loosely call Claude's inner life. She's a primary author of the document that defines Claude's values and character — the framework that governs how the model reasons when the rules run out. Almost as an aside near the end of the interview, she mentioned a prompting technique she uses to understand complex concepts.
It stopped me cold. Not because it was elaborate. Because it was disarmingly simple, and it worked in a way I hadn't thought to ask for.
The Exact Prompt Template
Here it is, cleaned up and ready to use:
I want to understand [concept].
Please explain it by writing a fable — an indirect,
narrative version of the concept.
The story should embody the concept completely without naming it directly.
Ideally, the reader should only start to realize
what the concept actually is near the end of the story.
After the fable, add a short explanation that names the concept clearly
and connects it back to the key moments in the story.
That's it. No elaborate scaffolding. No chain-of-thought trigger. No persona assignment. Just a deliberate decision about the order in which understanding should arrive.
Why This Works (and Why Direct Explanation Often Doesn't)
When you ask AI to explain a concept directly, you get a definition. Definitions are accurate and forgettable. The model produces the statistical center of everything written about that concept — clear, complete, and utterly without friction.
Friction, it turns out, is how things get encoded.
When a concept arrives wrapped in a story, your brain does something different. It tracks characters, infers motivation, builds a model of cause and effect. You're working the whole time, even if it doesn't feel like effort. By the time the fable ends and the concept is named explicitly, you've already reconstructed its logic from the inside out. The explanation that follows isn't introducing something new — it's confirming something you've just experienced.
This is not a new idea in pedagogy. It's the structure of every Socratic dialogue, every good case study, every effective parable. What's new is that you can now invoke it in a single prompt, on demand, for any concept you encounter.
Author's Note: I tested this on half a dozen concepts I'd read about but never felt I truly understood — things like "information asymmetry," "reflexive equilibria," and "Simpson's paradox." In every case, the fable version landed differently than the definition version. The model found angles I wouldn't have thought to ask for. The concept felt inhabited rather than described.
What Askell's Work Tells Us About Prompting
The fable prompt is a small window into something larger about how Askell thinks about AI interaction. Her work on Claude's character — documented in Anthropic's Claude's Character overview — is built around the idea that a model shouldn't just follow rules. It should internalize values deeply enough to exercise judgment when the rules run out.
And it echoes the philosophy behind the fable prompt in a specific way: both approaches treat the path to understanding as the variable worth designing, not just the destination.
Most prompt engineers are obsessed with what to ask. Askell seems to spend considerable energy on when to reveal what, and how the understanding should assemble itself in the reader's (or model's) mind. That's a different craft entirely.
The Prompt Is a Path, Not a Query
Here's the reframe that changed how I approach prompting more broadly: a prompt is not a question. It's a designed sequence of cognitive steps.
When you ask "explain X," you're outsourcing the entire sequence to the model, which will default to whatever sequence is most statistically common. That sequence is almost always: definition → examples → caveats. It's thorough. It's generic.
When you specify a narrative first, then a reveal, then an explanation — you're not just asking for different content. You're specifying a cognitive choreography that the model then executes. You're deciding what the reader encounters first, what they have to infer, and when the payoff arrives.
This is precisely why a prompt like Askell's is interesting to people who think carefully about prompting. It's not a trick. It's evidence that the instruction layer can control the reader's cognitive experience, not just the model's output.
This operates at a level far deeper than surface-level prompting. If you analyze this structure through the lens of prompt design—as we do in The Anatomy of a Perfect Prompt—you'll recognize that it leverages the interplay between Task and Format. Except here, the format isn't merely structural; it is experiential. You're architecting a reader journey rather than just configuring an output shape.
Practical Pitfall Avoidance Guide
Don't make the concept too abstract. Fables work when concepts have some causal structure — they have agents, decisions, consequences. Pure mathematical abstractions (e.g., the Riemann hypothesis) may produce beautiful prose that doesn't actually illuminate the concept. Test first.
Don't rush to name the concept. The prompt explicitly delays the reveal. If you add "and the concept is X" to the fable request, you collapse the structure. Let the model work without the label. The constraint is what produces the interesting compression.
Do run the explanation separately if it's unsatisfying. Sometimes the fable is excellent but the closing explanation is thin. In that case, follow up with: "Now explain the concept directly using the specific events and characters from the story you just wrote." This forces the model to anchor the abstract explanation to the concrete narrative rather than retreating to generic definition.
Do save fables that work. A well-constructed fable holds long-term value as an intellectual asset. Because a good narrative has continuous reuse potential, archiving it in a system like Prompt Vault ensures you can reliably retrieve and deploy the exact sequence whenever you need to teach or explain the concept again—rather than relying on a half-remembered prompt typed in haste.
Variations Worth Trying
The fable structure is a starting point. Once you have it, you can modify specific elements:
Change the fable length. Add "Keep the fable under 300 words" for a compressed version that tests the concept more severely. Longer fables allow more nuance; shorter ones force the model to identify the concept's actual core.
Change the genre. Replace "fable" with "detective story," "myth," or "corporate memo from a future civilization." Each genre imports different narrative conventions that illuminate different aspects of a concept. Information asymmetry told as a detective story is not the same as information asymmetry told as a fable — and both differ from a direct definition in ways that are instructive.
Add a reader persona. Prepend "Write this for someone who has no background in economics" to constrain the vocabulary and assumed knowledge level. The fable's characters and events will shift accordingly.
Chain it with questions. After you receive the fable and explanation, ask: "What aspect of [concept] did the story fail to capture?" This surfaces the model's awareness of its own simplifications — and often reveals the most interesting edge cases of the concept.
The Broader Implication
Most people who use AI extensively eventually land on the same observation: the output is only as interesting as the setup. Throw a flat question at a model and you get a flat answer. Design the cognitive path the model walks, and you get something structurally different.
Askell's fable prompt is a clean example of what that looks like in practice. It's not prompt engineering in the maximalist sense — no XML wrappers, no role assignments, no five-paragraph scaffolds. It's just a decision to control sequence: story first, reveal second, explanation third.
This highlights the core limitation of standard conversational prompting. Rather than outsourcing the sequence of understanding to a model's statistical defaults—a pitfall we explore when analyzing why to Stop Using One-Liner Prompts—this approach consciously overrides those defaults. The fable prompt doesn't succeed by adding instruction weight; it succeeds by replacing a generic sequence with a designed one.
The research on narrative learning backs this up. According to Anthropic's own writing on Claude's character development, the company deliberately chose to train Claude with something closer to virtue ethics than rule-following — shaping dispositions rather than issuing instructions. The fable prompt embodies a similar philosophy: instead of telling AI what to output, you design the conditions under which the right output naturally emerges.
That's a harder thing to get right. But when it works, it works differently — and the difference is felt immediately.
The template, one more time:
I want to understand [concept].
Please explain it by writing a fable — an indirect,
narrative version of the concept.
The story should embody the concept completely without naming it directly.
Ideally, the reader should only start to realize
what the concept actually is near the end of the story.
After the fable, add a short explanation that names the concept clearly
and connects it back to the key moments in the story.
Copy it. Use it on the next concept that resists direct explanation. You'll notice the difference.
