E01¶


Research	R0023 — Counterproductive advice and prompt lifecycle
Run	2026-03-25
Query	Q002
Source	SRC04
Evidence	SRC04-E01
Type	Factual

Anthropic guide partially attributed; recommends persona prompting despite empirical evidence against it.

URL: https://docs.anthropic.com/en/docs/build-with-claude/prompt-engineering/overview

Extract¶

Anthropic's prompt engineering documentation is partially attributed to Rick Dakan and Joseph Feller. The documentation includes templates like "You are an expert AI tax analyst" for role prompting. The Wharton GAIL Report 4 specifically names Anthropic alongside Google and OpenAI as vendors whose persona prompting recommendations are not supported by empirical evidence for factual accuracy tasks.

The interactive tutorial covers 9 chapters of prompt engineering techniques. It is maintained by Anthropic's documentation team and informed by their Applied AI team.

Relevance to Hypotheses¶

Hypothesis	Relationship	Strength
H1	Partially supports	At least some attribution exists, but authors' research credentials are unclear
H2	Supports	Documentation writers, not necessarily AI researchers, appear to author vendor guides
H3	Supports	Vendor guides represent a distinct category mixing research insights with commercial documentation

Context¶

Anthropic is notable as the only major vendor that partially attributes authorship. This is a marginal improvement over OpenAI and Google but still does not provide the transparency expected of research-backed recommendations.