# Beyond Prompt Engineering: A Systematic Analysis of Prompt Lexical Sensitivity and Its Impacts on Quality

arXiv: https://arxiv.org/abs/2608.20349 · HTML: https://mlsanity.com/paper/2608.20349/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 22 of its 31 references that are in the map.
- First posted on arXiv 16 Jun 2026.

## Builds on (most cited first, 22 of 22)

- [Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena](https://mlsanity.com/paper/2306.05685/) (2023)
- [Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities](https://mlsanity.com/paper/2507.06261/) (2025)
- [G-Eval: NLG Evaluation using GPT-4 with Better Human Alignment](https://mlsanity.com/paper/2303.16634/) (2023)
- [A Survey on LLM-as-a-Judge](https://mlsanity.com/paper/2411.15594/) (2024)
- [Calibrate Before Use: Improving Few-Shot Performance of Language Models](https://mlsanity.com/paper/2102.09690/) (2021)
- [Fantastically Ordered Prompts and Where to Find Them: Overcoming Few-Shot Prompt Order Sensitivity](https://mlsanity.com/paper/2104.08786/) (2021)
- [Large Language Models are not Fair Evaluators](https://mlsanity.com/paper/2305.17926/) (2023)
- [Large Language Models Can Be Easily Distracted by Irrelevant Context](https://mlsanity.com/paper/2302.00093/) (2023)
- [A Survey on In-context Learning](https://mlsanity.com/paper/2301.00234/) (2022)
- [GPT Understands, Too](https://mlsanity.com/paper/2103.10385/) (2021)
- [WizardMath: Empowering Mathematical Reasoning for Large Language Models via Reinforced Evol-Instruct](https://mlsanity.com/paper/2308.09583/) (2023)
- [Making Retrieval-Augmented Language Models Robust to Irrelevant Context](https://mlsanity.com/paper/2310.01558/) (2023)
- [True Few-Shot Learning with Language Models](https://mlsanity.com/paper/2105.11447/) (2021)
- [State of What Art? A Call for Multi-Prompt LLM Evaluation](https://mlsanity.com/paper/2401.00595/) (2023)
- [FLASK: Fine-grained Language Model Evaluation based on Alignment Skill Sets](https://mlsanity.com/paper/2307.10928/) (2023)
- [Batch Calibration: Rethinking Calibration for In-Context Learning and Prompt Engineering](https://mlsanity.com/paper/2309.17249/) (2023)
- [Mind Your Format: Towards Consistent Evaluation of In-Context Learning Improvements](https://mlsanity.com/paper/2401.06766/) (2024)
- [Aligning Language Models to Explicitly Handle Ambiguity](https://mlsanity.com/paper/2404.11972/) (2024)
- [How are Prompts Different in Terms of Sensitivity?](https://mlsanity.com/paper/2311.07230/) (2023)
- [Prompt Perturbation Consistency Learning for Robust Language Models](https://mlsanity.com/paper/2402.15833/) (2024)
- [Towards LLMs Robustness to Changes in Prompt Format Styles](https://mlsanity.com/paper/2504.06969/) (2025)
- [Evaluating Robustness of Large Language Models Against Multilingual Typographical Errors](https://mlsanity.com/paper/2510.09536/) (2025)

## Abstract (arXiv)

Large Language Models (LLMs) exhibit extreme sensitivity to surface-level prompt variations, in which minor lexical changes can trigger disproportionate performance fluctuations. Moving beyond black-box optimization and coarse-grained templates, we present the first large-scale, n-gram token-level mechanistic analysis of prompt stability, leveraging a dataset of 132,000 prompt variants. Our investigation reveals a fundamental Scaling Law of Prompt Performance Stability: higher average task performance is strongly associated with lower variance and greater robustness across prompt perturbation. We identify two core linguistic drivers underlying this robustness: (1) Domain-Specific Terminology, which tightly anchors semantic boundaries, and (2) Explicit Action Directives, which formalize reasoning trajectories. Together, these elements constrain the model's interpretative space, effectively ``locking in'' more deterministic generation behavior. Building on these insights, we introduce an automated Prompt-Refining Agent that systematically restructures input queries by injecting domain anchoring and operational constraints. Empirical evaluation shows that our approach reduces performance variance by 40.7% in code generation task, while preserving or improving mean performance. These findings provide a statistically grounded and mechanistically interpretable framework for achieving robust prompt engineering.