S02¶


Research	R0054 — Prompt Claims v2
Run	2026-03-31
Claim	C002
Search	S02

WebSearch — LLM negation and instruction-following research

Summary¶

Field	Value
Source/Database	WebSearch
Query terms	KAIST negation prompts LLM models perform worse negative instructions research paper
Filters	None
Results returned	10
Results selected	2
Results rejected	8

Selected Results¶

Result	Title	URL	Rationale
S02-R01	Why Positive Prompts Outperform Negative Ones with LLMs	https://gadlet.com/posts/negative-prompting/	Directly addresses the positive vs negative prompt effectiveness question with research citations
S02-R02	LLMs Don't Understand Negation	https://hackernoon.com/llms-dont-understand-negation	Research on LLM negation processing weaknesses

Rejected Results¶

Result	Title	URL	Rationale
S02-R03	Hiding Prompt Injections in Academic Papers	https://www.schneier.com/blog/archives/2025/07/hiding-prompt-injections-in-academic-papers.html	About prompt injection security, not instruction types
S02-R04	LLMs and the "not" problem	https://seantrott.substack.com/p/llms-and-the-not-problem	Linguistic analysis of negation, tangentially relevant
S02-R05	Understanding the Relationship Between LLMs and Negation	https://swimm.io/blog/understanding-llms-and-negation	General overview, insufficient depth
S02-R06	Improving Negation Reasoning in LLMs	https://aclanthology.org/2025.findings-emnlp.761.pdf	Academic paper on improving negation, not on prompt design strategy
S02-R07	LLM Negative Prompts: avoid unintended consequences	https://www.linkedin.com/pulse/llm-negative-prompts-avoid-unintended-consequences-chris-clark-fdvwe	LinkedIn article, insufficient depth
S02-R08	IGNORE ALL PREVIOUS INSTRUCTIONS	https://statmodeling.stat.columbia.edu/2025/07/07/chatbot-prompts/	About prompt injection cheating, not instruction design
S02-R09	NegativePrompt: Leveraging Psychology for LLMs	https://www.ijcai.org/proceedings/2024/0719.pdf	Academic paper on negative prompting from psychology perspective
S02-R10	Negation: A Pink Elephant in the LLMs' Room?	https://arxiv.org/html/2503.22395v2	About negation processing in LLMs, tangentially relevant

Notes¶

The selected results provide both the research foundation (LLMs struggle with negation) and the practical implication (positive prompts outperform negative ones when used alone, but both are needed together).