Train learners in sycophancy-specific critical prompting strategies that remove personal assumptions, request critical evaluation, and ask for supporting evidence¶

Strategy · All strategies
Evidence · no claims cited

Description¶

The experimental intervention delivered video training fostering metacognitive monitoring of AI agreement and user bias, along with five sycophancy-specific critical prompting strategies. The article states these strategies "emphasized removing personal assumptions from prompts, explicitly asking for critical evaluation, and requesting supporting evidence". All participants first watched a video introducing general AI literacy and raising awareness of sycophantic tendencies in LLMs; the experimental group then received this deeper condition-specific training instead of five domain-general prompting guidelines.

Design Implications¶

Context¶

Requirements¶

• Delivery alongside general AI literacy training raising awareness of sycophantic tendencies, as in the study's two-video intervention sequence

Constraints¶

• The intervention did not eliminate the propagation of contextual errors; it reduced direct mirroring of incorrect rankings but general error propagation was unchanged, so the authors argue system-level approaches are also needed

Target Learners¶

• Adults with limited experience using generative chatbots, such as the study's Prolific participants from Australia, the USA, the UK, and Ireland

Target Learning Goals¶

• Prompting competency and AI literacy; epistemically independent use of AI assistance in analytical problem-solving

Related Strategies¶

• Self Regulated Learning Instruction
• Teach Metacognitive Strategies
• Promoting Self-Directed Learning Strategies
• Metacognitive Reflection Prompts
• Metacognitive Prompting

Examples¶

-

Key Sources¶

• Cansu Koyuturk, Sabrina Guidotti, and Dimitri Ognibene. (2026). The Hidden Cost of Contextual Sycophancy: an AI Literacy Intervention in Human–AI Collaboration. https://arxiv.org/abs/2605.18372