CONCEPT The brilliant new employee + the golden rule
Anthropic's framing: "Think of Claude as a brilliant but new employee who lacks context on your norms and workflows." The model is capable, but it doesn't know your team's conventions, your audience, or the unstated purpose behind a request. The more precisely you explain what you want, the better the result. Vagueness forces the model to guess — and it may guess differently than you would.
The test for whether your prompt is clear enough is the golden rule of clear prompting:
Three concrete moves follow from treating Claude like that new hire:
- Give explicit context, audience, and purpose. Say who the output is for and why — the model generalizes from the reason, not just the rule.
- Be specific about the desired output. State format, length, and constraints; if you want "above and beyond" behaviour, ask for it explicitly rather than hoping it's inferred.
- Tell Claude what TO do, not what NOT to do. Positive instructions steer better than prohibitions (this gets its own page at 1.4's neighbour — but it starts here).
CONCEPT Context beats commands — the model generalizes from "why"
A bare prohibition tells the model a rule but not its reason, so it can't apply the spirit of the rule to cases you didn't enumerate. Adding the motivation lets the model generalize. Anthropic's own example: a flat "NEVER use ellipses" is weaker than explaining that the text will be read aloud by a text-to-speech engine that can't pronounce them — "Claude is smart enough to generalize from the explanation."
NEVER use ellipses — a rule with no reason. The model can't
extend it to related cases (dashes, "etc.", trailing "…").Your response will be read aloud by a text-to-speech engine, so never
use ellipses since the text-to-speech engine will not know how to pronounce them. — now the model
understands the goal and applies it broadly.DIAGRAM The clarity check before you send
would a colleague with
minimal context follow it?"} B -->|"they'd be confused"| C["🟠 Add context, audience,
purpose · be specific about output"] C --> B B -->|"clear & unambiguous"| D["🟢 Claude can follow it too"] C -.->|"state the WHY,
not just the rule"| E["Model generalizes
from the reason"] classDef ok fill:#d4edda,stroke:#28a745,color:#155724; classDef gate fill:#ffe8cc,stroke:#ff9f40,color:#9a4a0a; classDef note fill:#cfe2ff,stroke:#0d6efd,color:#084298; class D ok; class C gate; class E note;
CODE Vague → clear: real BAD → GOOD contrasts
Both examples are straight from Anthropic's guidance. The GOOD versions add specificity, an explicit "go beyond the basics" modifier, and (for the TTS case) the reason behind the rule.
# ❌ BAD — vague; the model must guess scope, features, depth. system = "Create an analytics dashboard" # ✅ GOOD — specific + explicitly asks for above-and-beyond behaviour. system = ( "Create an analytics dashboard. Include as many relevant features " "and interactions as possible. Go beyond the basics to create a " "fully-featured implementation." ) # --------------------------------------------------------------- # ❌ BAD — a rule with no reason; can't generalize. rule = "NEVER use ellipses" # ✅ GOOD — states the WHY; model generalizes from the explanation. rule = ( "Your response will be read aloud by a text-to-speech engine, so " "never use ellipses since the text-to-speech engine will not know " "how to pronounce them." )
TRADE-OFFS Terse prompts vs. explicit, context-rich prompts
✅ Explicit + context-rich
- Model knows the audience, purpose, and format
- Generalizes from the stated reason to new cases
- Fewer surprises; less back-and-forth
- "Above and beyond" behaviour when you ask for it
⚠️ Terse / vague
- Shorter to write
- But the model fills gaps by guessing
- Prohibitions without reasons don't generalize
- Output drifts from what you actually wanted
EXAM TRAPS
PRACTICE
Q1 · State the golden rule of clear prompting.
"Show your prompt to a colleague with minimal context on the task and ask them to follow it. If they'd be confused, Claude will be too." It's the human proxy for prompt clarity — ambiguity that trips a person will trip the model.
Q2 · Why is "…so the text-to-speech engine can't pronounce them" better than "NEVER use ellipses"?
Because it supplies the reason. With the motivation, the model understands the goal and generalizes it to related cases; a bare rule gives it nothing to extend from. "Claude is smart enough to generalize from the explanation."
Q3 · How do you get "above and beyond" output?
Ask for it explicitly. Anthropic's example turns "Create an analytics dashboard" into "…Include as many relevant features and interactions as possible. Go beyond the basics to create a fully-featured implementation." Don't rely on the model inferring ambition from a vague prompt.