Q002 — Query Definition¶

Query as Received¶

Do mainstream prompt engineering guides and best-practice documents discuss techniques for reducing sycophantic behavior in AI responses?

Subject: Mainstream prompt engineering guidance (vendor documentation, industry guides, widely-referenced best practices)
Scope: Whether these documents explicitly address sycophancy — the tendency of AI models to agree with users rather than maintain accuracy — and provide actionable techniques for reducing it
Evidence basis: Official vendor documentation (OpenAI, Anthropic, Google), widely-cited prompt engineering guides, academic research on sycophancy mitigation

"Mainstream" is subjective — this research interprets it as vendor-published documentation, widely-cited guides (e.g., Lakera, Prompt Engineering Guide), and high-profile industry publications.
The query asks about "discuss techniques" which could range from a brief mention to in-depth methodology. Both depth and presence are examined.
Sycophancy exists on a spectrum from excessive praise to subtle agreement bias. The query encompasses all forms.

Do major AI vendor prompt engineering guides explicitly mention sycophancy?
What prompt-level techniques for reducing sycophancy are documented in the literature?
How effective are these techniques based on available evidence?
Is there a gap between academic research on sycophancy and mainstream prompt engineering guidance?

ID	Hypothesis	Description
H1	Yes, mainstream guides address sycophancy	Mainstream prompt engineering guides explicitly discuss sycophancy and provide actionable prompt-level techniques
H2	No, sycophancy is not addressed in mainstream guides	Mainstream guides do not discuss sycophancy; it remains an academic and research topic only
H3	Partially — emerging but inconsistent coverage	Some guides address sycophancy (particularly recent ones), but coverage is inconsistent, often indirect, and typically lacks the depth found in academic research