E02¶


Research	R0053 — Prompt Claims
Run	2026-03-31-02
Claim	C002
Source	SRC01
Evidence	SRC01-E02
Type	Reported

Anthropic recommends positive framing over negative constraints

URL: https://eval.16x.engineer/blog/the-pink-elephant-negative-instructions-llms-effectiveness-analysis

Extract¶

Anthropic explicitly advises: "Tell Claude what to do instead of what not to do." Example reframing: instead of "Don't use mock data" use "Only use real-world data." A user resolved file-duplication issues by changing from "do not make new versions" to "Make all possible updates in current files whenever possible." However, negative constraints retain value for preventing harmful behaviors in system prompts and establishing firm boundaries.

Relevance to Hypotheses¶

Hypothesis	Relationship	Strength
H1	Contradicts	Official vendor guidance contradicts the claim's mechanism
H2	Supports	Shows enforcement works better with positive framing
H3	N/A	Does not address whether AI follows all requirements

Context¶

The nuance about negative constraints being useful for firm boundaries is important — the claim is not entirely wrong, just oversimplified.