# Flaw or Artifact? Rethinking Prompt Sensitivity in Evaluating LLMs

arXiv: https://arxiv.org/abs/2509.01790 · HTML: https://mlsanity.com/paper/2509.01790/

## Key facts

- Cited by 30 papers in the map (data through 14 Oct 2026).
- Placed in Prompting and chain-of-thought reasoning, one of 17 directions in LLM reasoning.
- Builds on 20 of its 24 references that are in the map.
- First posted on arXiv 1 Sep 2025.

## Placed in

- LLM reasoning / [Prompting and chain-of-thought reasoning](https://mlsanity.com/lr/prompting-and-chain-of-thought-reasoning/)

## Builds on (most cited first, 20 of 20)

- [Language Models are Few-Shot Learners](https://mlsanity.com/paper/2005.14165/) (2020)
- [Measuring Massive Multitask Language Understanding](https://mlsanity.com/paper/2009.03300/) (2020)
- [Measuring Mathematical Problem Solving With the MATH Dataset](https://mlsanity.com/paper/2103.03874/) (2021)
- [Think you have Solved Question Answering? Try ARC, the AI2 Reasoning Challenge](https://mlsanity.com/paper/1803.05457/) (2018)
- [HellaSwag: Can a Machine Really Finish Your Sentence?](https://mlsanity.com/paper/1905.07830/) (2019)
- [GPQA: A Graduate-Level Google-Proof Q&A Benchmark](https://mlsanity.com/paper/2311.12022/) (2023)
- [G-Eval: NLG Evaluation using GPT-4 with Better Human Alignment](https://mlsanity.com/paper/2303.16634/) (2023)
- [Can a Suit of Armor Conduct Electricity? A New Dataset for Open Book Question Answering](https://mlsanity.com/paper/1809.02789/) (2018)
- [LongBench: A Bilingual, Multitask Benchmark for Long Context Understanding](https://mlsanity.com/paper/2308.14508/) (2023)
- [Super-NaturalInstructions: Generalization via Declarative Instructions on 1600+ NLP Tasks](https://mlsanity.com/paper/2204.07705/) (2022)
- [Quantifying Language Models' Sensitivity to Spurious Features in Prompt Design or: How I learned to start worrying about prompt formatting](https://mlsanity.com/paper/2310.11324/) (2023)
- [CoQA: A Conversational Question Answering Challenge](https://mlsanity.com/paper/1808.07042/) (2018)
- [The NarrativeQA Reading Comprehension Challenge](https://mlsanity.com/paper/1712.07040/) (2017)
- [Is ChatGPT a Good NLG Evaluator? A Preliminary Study](https://mlsanity.com/paper/2303.04048/) (2023)
- [Large Language Models Are State-of-the-Art Evaluators of Translation Quality](https://mlsanity.com/paper/2302.14520/) (2023)
- [State of What Art? A Call for Multi-Prompt LLM Evaluation](https://mlsanity.com/paper/2401.00595/) (2023)
- [Can large language models reason about medical questions?](https://mlsanity.com/paper/2207.08143/) (2022)
- [Mind Your Format: Towards Consistent Evaluation of In-Context Learning Improvements](https://mlsanity.com/paper/2401.06766/) (2024)
- [Efficient multi-prompt evaluation of LLMs](https://mlsanity.com/paper/2405.17202/) (2024)
- [NutriBench: A Dataset for Evaluating Large Language Models on Nutrition Estimation from Meal Descriptions](https://mlsanity.com/paper/2407.12843/) (2024)

## Built on by (most cited first, 30 of 30)

- [WebDevJudge: Evaluating (M)LLMs as Critiques for Web Development Quality](https://mlsanity.com/paper/2510.18560/) (2025)
- [JudgeSense: A Benchmark for Prompt Sensitivity in LLM-as-a-Judge Systems](https://mlsanity.com/paper/2604.23478/) (2026)
- [Beyond Accuracy: Policy Invariance as a Reliability Test for LLM Safety Judges](https://mlsanity.com/paper/2605.06161/) (2026)
- [Measuring Pragmatic Influence in Large Language Model Instructions](https://mlsanity.com/paper/2602.21223/) (2026)
- [Where the Verifier Fails: A Category-Level Audit of Reward Signals in RLVR](https://mlsanity.com/paper/2609.01354/) (2026)
- [AI Native Asset Intelligence](https://mlsanity.com/paper/2605.09115/) (2026)
- [PromptAudit: Auditing Prompt Sensitivity in LLM-Based Vulnerability Detection](https://mlsanity.com/paper/2605.24171/) (2026)
- [Mitigating LLM-based p-Hacking by Preregistering for the Next LLM](https://mlsanity.com/paper/2606.27687/) (2026)
- [Banana100: Breaking NR-IQA Metrics by 100 Iterative Image Replications with Nano Banana Pro](https://mlsanity.com/paper/2604.03400/) (2026)
- [PRISM: A Benchmark for Programmatic Spatial-Temporal Reasoning](https://mlsanity.com/paper/2605.19382/) (2026)
- [Discourse-Role Labels as Presentation-Time Variables for Context Use in Language Models](https://mlsanity.com/paper/2606.04109/) (2026)
- [Document-Authored Control-Signal Impersonation: A Low-Cost Indirect Prompt Attack on RAG Safety Boundaries](https://mlsanity.com/paper/2606.09005/) (2026)
- [ExpressEdit: Fast Editing of Stylized Facial Expressions with Diffusion Models in Photoshop](https://mlsanity.com/paper/2604.03448/) (2026)
- [LLMs as annotators of credibility assessment in Danish asylum decisions: evaluating classification performance and errors beyond aggregated metrics](https://mlsanity.com/paper/2605.13412/) (2026)
- [Spurious Prompts: Can Irrelevant Prompts Steer Large Language Models?](https://mlsanity.com/paper/2605.29678/) (2026)
- [The Unsampled Truth: Quantifying Prompt Artifacts in LM Psychometrics](https://mlsanity.com/paper/2606.03357/) (2026)
- [Understanding Evaluation Illusion in Diffusion Large Language Models](https://mlsanity.com/paper/2606.29228/) (2026)
- [Agent Step Value: Auditing Evaluation-Channel Sensitivity in Black-Box Agent Traces](https://mlsanity.com/paper/2607.04419/) (2026)
- [Measuring GEO Visibility: Prompt Corpora Define the Answer Market](https://mlsanity.com/paper/2609.06811/) (2026)
- [AfriSyCo: Measuring Assertive Framing, Verification, and Wording Sensitivity Around African-Language Content](https://mlsanity.com/paper/2609.17853/) (2026)
- [What Changes When Fact-Verification Scores Improve? Evidence and Answer Accounting Across Trained Verifiers and LLMs](https://mlsanity.com/paper/2609.27064/) (2026)
- [The Decomposition Tax: LLM Pipelines Lose Up to 40 Accuracy Points at Their Own Interfaces](https://mlsanity.com/paper/2609.32825/) (2026)
- [Generalization Is Stability, Not Accuracy: Multi-Axis Evaluation of LLMs](https://mlsanity.com/paper/2610.01428/) (2026)
- [When Verifiable Counts Depend on Wording: Auditing Wording Robustness in Instruction Following](https://mlsanity.com/paper/2610.05278/) (2026)
- [Quantization Effects on Tool-Failure Recovery Vary Across Prompts and Evaluation Designs](https://mlsanity.com/paper/2610.07781/) (2026)
- [A Distributional Framework for Behavioral Characterization and Risk Management of Large Language Models](https://mlsanity.com/paper/2610.12718/) (2026)
- [Do correct answers have stable latent knowledge support?](https://mlsanity.com/paper/2610.15277/) (2026)
- [REA: From Correct LLM Judgments to Safe Decision Delivery](https://mlsanity.com/paper/2610.17473/) (2026)
- [Beyond Aggregate Accuracy: Item-Level Transitions in Prompt Optimization](https://mlsanity.com/paper/2610.19084/) (2026)
- [RCCM: Diagnosing Gains and Harms in Reasoning-Prompt Transfer](https://mlsanity.com/paper/2610.19872/) (2026)

## Abstract (arXiv)

Prompt sensitivity, referring to the phenomenon where paraphrasing (i.e., repeating something written or spoken using different words) leads to significant changes in large language model (LLM) performance, has been widely accepted as a core limitation of LLMs. In this work, we revisit this issue and ask: Is the widely reported high prompt sensitivity truly an inherent weakness of LLMs, or is it largely an artifact of evaluation processes? To answer this question, we systematically evaluate 7 LLMs (e.g., GPT and Gemini family) across 6 benchmarks, including both multiple-choice and open-ended tasks on 12 diverse prompt templates. We find that much of the prompt sensitivity stems from heuristic evaluation methods, including log-likelihood scoring and rigid answer matching, which often overlook semantically correct responses expressed through alternative phrasings, such as synonyms or paraphrases. When we adopt LLM-as-a-Judge evaluations, we observe a substantial reduction in performance variance and a consistently higher correlation in model rankings across prompts. Our findings suggest that modern LLMs are more robust to prompt templates than previously believed, and that prompt sensitivity may be more an artifact of evaluation than a flaw in the models.