Elamur Research⌕
☰
home / papers / 2026 / What Prompts Don't Say: Understanding and Managing Underspecification in LLM Prompts

What Prompts Don't Say: Understanding and Managing Underspecification in LLM Prompts

Paper · 2026 · Chenyang Yang, Yike Shi, Qianou Ma et al.

Summary

This paper examines prompt underspecification: developers leave out requirements that matter, so a model can produce many valid but inconsistent behaviours. The authors argue that ad-hoc, trial-and-error prompt writing and the fast parallel evolution of prompts and models make this problem worse than in classical software or ML. They built a benchmark of 60 plausible requirements across three tasks drawn from Anthropic's AI usage report. They wrote 240 synthetic prompts, each covering a different subset of those requirements, and scored outputs with human-validated LLM-as-a-judge validators. Models often satisfy unstated requirements on their own, but this behaviour is unreliable and tends to break when the prompt or model changes. Listing every requirement does not fix this, because instruction-following degrades as prompts grow, and generic prompt optimizers help little. The authors propose requirements-aware optimizers that tune how requirements are phrased or choose which ones to state, raising accuracy and shortening prompts. They also outline a process for discovering, validating and monitoring requirements. This matters for anyone maintaining LLM-based applications over time.

Key points

  • Models infer unspecified requirements in 41.1% of cases by default. However, unstated requirements are twice as likely to regress after prompt or model changes, sometimes losing more than 20% accuracy.
  • Writing out all requirements can lower performance by 19%, because instruction-following is limited and requirements can conflict.
  • Requirements-aware optimizers improve results by 4.8% on average over baselines. The variant that selects which requirements to state produces prompts with 43% fewer tokens.
  • The study covers 60 requirements, 240 prompts and about 8.4k aggregated data points. It required over 1.5 million LLM calls, and the three tasks come from the Office, Business and Computer categories.
  • Code and data are available at github.com/malusamayo/underspec-analysis. The authors list as limitations the small scale, possible same-family bias in LLM validators, and the absence of a user study.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🎓Carnegie Mellon UniversityPittsburgh · United States598 works · 191,225 citations

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksArts and Humanities · 17,450 worksLanguage and Linguistics · 17,296 works

🧭 Topics 1Research subjects and languages the work is about.Language models & pretraining · 3,894 works

Cite

Cite as Chenyang Yang, Yike Shi, Qianou Ma, Michael Xieyang Liu, Christian Kaestner and Tongshuang Wu. 2026. What Prompts Don't Say: Understanding and Managing Underspecification in LLM Prompts. Findings of the Association for Computational Linguistics: ACL 2026.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.findings-acl.441

@inproceedings{yang2026what,
  title = {{What Prompts Don't Say: Understanding and Managing Underspecification in LLM Prompts}},
  author = {Chenyang Yang, Yike Shi, Qianou Ma, Michael Xieyang Liu, Christian Kaestner and Tongshuang Wu},
  booktitle = {Findings of the Association for Computational Linguistics: ACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-acl.441},
  url = {https://doi.org/10.18653/v1/2026.findings-acl.441}
}

References

49 references, 11 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Sam Adeniyi. 2026. Lesson plan dataset. https://huggingface.co/datasets/samadeniyi/ lesson_plan. ↗
  2. All-Hands-AI. 2025. Openhands system_prompt.j2. https://github.com/All-Hands-AI/ OpenHands/blob/main/openhands/agenthub/ codeact_agent/prompts/system_prompt.j2. Accessed: 2025-05-14. ↗
  3. Anthropic. 2025. System prompts: Claude 3.7 sonnet. find
  4. Rahul K Arora, Jason Wei, Rebecca Soskin Hicks, Preston Bowman, Joaquin Quiñonero-Candela, Foivos Tsimpourlas, Michael Sharman, Meghan Shah, Andrea Vallone, Alex Beutel, and 1 others. 2025. Healthbench: Evaluating large language models towards improved human health. arXiv preprint arXiv:2505.08775.
  5. Hamed Barzamini, Mona Rahimi, Murteza Shahzad, and Hamed Alhoori. 2022. Improving generalizability of ml-enabled software through domain specification. In Proceedings of the 1st International Conference on AI Engineering: Software Engineering for AI, CAIN ’22, page 181–192, New York, NY, USA. Associ… find
  6. James Bergstra, Rémi Bardenet, Yoshua Bengio, and Balázs Kégl. 2011. Algorithms for hyper-parameter optimization. Advances in neural information processing systems, 24. find
  7. Natasha Bernal and Amanda Hoover. 2024. We asked ai to take us on a tour of our cities. it was chaos. WIRED. find
  8. Chris Bogart, Christian Kästner, James Herbsleb, and Ferdian Thung. 2021. When and how to make breaking changes: Policies and practices in 18 open source software ecosystems. ACM Transactions on Software Engineering and Methodology (TOSEM), 30(4):1– 56. find
  9. Philippe Laban, Hiroaki Hayashi, Yingbo Zhou, and Jennifer Neville. 2025. Llms get lost in multi-turn conversation. arXiv preprint arXiv:2505.06120.
  10. Bowen Cao, Deng Cai, Zhisong Zhang, Yuexian Zou, and Wai Lam. 2024. On the worst prompt performance of large language models. arXiv preprint arXiv:2406.10248. — [arXiv:2406.10248](https://arxiv.org/abs/2406.10248) find
  11. Mingqi Li, Karan Aggarwal, Yong Xie, Aitzaz Ahmad, and Stephen Lau. 2024. Learning from contrastive prompts: Automated optimization and adaptation. arXiv preprint arXiv:2409.15199. — [arXiv:2409.15199](https://arxiv.org/abs/2409.15199) find
  12. Lingjiao Chen, Matei Zaharia, and James Zou. 2024. How is chatgpt’s behavior changing over time? Harvard Data Science Review, 6(2). find
  13. Jenny T Liang, Melissa Lin, Nikitha Rao, and Brad A Myers. 2024. Prompts are programs too! understanding how developers build software containing prompts. arXiv preprint arXiv:2409.12447. — [arXiv:2409.12447](https://arxiv.org/abs/2409.12447) find
  14. Alexander D’Amour, Katherine Heller, Dan Moldovan, Ben Adlam, Babak Alipanahi, Alex Beutel, Christina Chen, Jonathan Deaton, Jacob Eisenstein, Matthew D Hoffman, and 1 others. 2022. Underspecification presents challenges for credibility in modern machine learning. Journal of Machine Learning Researc… find
  15. Ning Ding, Yulin Chen, Bokai Xu, Yujia Qin, Zhi Zheng, Shengding Hu, Zhiyuan Liu, Maosong Sun, and Bowen Zhou. 2023. Enhancing chat language models by scaling high-quality instructional conversations. arXiv preprint arXiv:2305.14233.
  16. Lisa Dunlap, Krishna Mandal, Trevor Darrell, Jacob Steinhardt, and Joseph E Gonzalez. 2024. Vibecheck: Discover and quantify qualitative differences in large language models. arXiv preprint arXiv:2410.12851. find
  17. Kunal Handa, Alex Tamkin, Miles McCain, Saffron Huang, Esin Durmus, Sarah Heck, Jared Mueller, Jerry Hong, Stuart Ritchie, Tim Belonax, and 1 others. 2025. Which economic tasks are performed with ai? evidence from millions of claude conversations. arXiv preprint arXiv:2503.04761. find
  18. Yining Hong, Christopher S Timperley, and Christian Kästner. 2025. From hazard identification to controller design: Proactive and llm-supported safety engineering for ml-powered systems. arXiv preprint arXiv:2502.07974. — [arXiv:2502.07974](https://arxiv.org/abs/2502.07974) find
  19. Chip Huyen. 2022. Designing Machine Learning Systems. O’Reilly Media, USA. find
  20. Christian Kästner, Eunsuk Kang, and Sven Apel. 2021. Feature interactions on steroids: On the composition of ml models. arXiv preprint arXiv:2105.06449. — [arXiv:2105.06449](https://arxiv.org/abs/2105.06449) find
  21. Omar Khattab, Arnav Singhvi, Paridhi Maheshwari, Zhiyuan Zhang, Keshav Santhanam, Sri Vardhamanan, Saiful Haq, Ashutosh Sharma, Thomas T Joshi, Hanna Moazam, and 1 others. 2023. Dspy: Compiling declarative language model calls into self-improving pipelines. arXiv preprint arXiv:2310.03714.
  22. Hyuhng Joon Kim, Youna Kim, Cheonbok Park, Junyeob Kim, Choonghyun Park, Kang Min Yoo, Sang-goo Lee, and Taeuk Kim. 2024. Aligning language models to explicitly handle ambiguity. arXiv preprint arXiv:2404.11972. find
  23. Qianou Ma, Weirui Peng, Chenyang Yang, Hua Shen, Kenneth Koedinger, and Tongshuang Wu. 2025. What should we engineer in prompts? training humans in requirement-driven llm use. find
  24. Wanqin Ma, Chenyang Yang, and Christian Kästner. 2024. (why) is my prompt getting worse? rethinking regression testing for evolving llm apis. In Proceedings of the IEEE/ACM 3rd International Conference on AI Engineering-Software Engineering for AI, pages 166–171. find
  25. Steve McConnell. 1998. Software project survival guide. Pearson Education. find
  26. Niklas Muennighoff, Qian Liu, Armel Zebaze, Qinkai Zheng, Binyuan Hui, Terry Yue Zhuo, Swayam Singh, Xiangru Tang, Leandro von Werra, and Shayne Longpre. 2023. Octopack: Instruction tuning code large language models. arXiv preprint arXiv:2308.07124. find
  27. OpenAI. 2025. Prompt generation. find
  28. Long Ouyang, Jeffrey Wu, Xu Jiang, Diogo Almeida, Carroll Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, and 1 others. 2022. Training language models to follow instructions with human feedback. Advances in neural information processing systems, 35:27730–27744.
  29. Yiwei Qin, Kaiqiang Song, Yebowen Hu, Wenlin Yao, Sangwoo Cho, Xiaoyang Wang, Xuansheng Wu, Fei Liu, Pengfei Liu, and Dong Yu. 2024. Infobench: Evaluating instruction following ability in large language models. arXiv preprint arXiv:2401.03601.
  30. Chandan K. Reddy, Lluís Màrquez, Fran Valero, Nikhil Rao, Hugo Zaragoza, Sambaran Bandyopadhyay, Arnab Biswas, Anlu Xing, and Karthik Subbian. 2022. Shopping queries dataset: A large-scale ESCI benchmark for improving product search. arXiv preprint arXiv:2206.06588. — [arXiv:2206.06588](https://arxiv.org/abs/2206.06588) find
  31. Victor Sanh, Albert Webson, Colin Raffel, Stephen H Bach, Lintang Sutawika, Zaid Alyafeai, Antoine Chaffin, Arnaud Stiegler, Teven Le Scao, Arun Raja, and 1 others. 2021. Multitask prompted training enables zero-shot task generalization. arXiv preprint arXiv:2110.08207.
  32. Melanie Sclar, Yejin Choi, Yulia Tsvetkov, and Alane Suhr. 2023. Quantifying language models’ sensitivity to spurious features in prompt design or: How i learned to start worrying about prompt formatting. arXiv preprint arXiv:2310.11324. find
  33. Shreya Shankar, JD Zamfirescu-Pereira, Björn Hartmann, Aditya Parameswaran, and Ian Arawjo. 2024. Who validates the validators? aligning llm-assisted evaluation of llm outputs with human preferences. In Proceedings of the 37th Annual ACM Symposium on User Interface Software and Technology, pages 1–1… find
  34. Ian Sommerville. 2015. Software Engineering, 10th edition. Pearson. find
  35. Mingjie Sun, Yida Yin, Zhiqiu Xu, J Zico Kolter, and Zhuang Liu. 2025. Idiosyncrasies in large language models. arXiv preprint arXiv:2502.12150. — [arXiv:2502.12150](https://arxiv.org/abs/2502.12150) find
  36. Annalisa Szymanski, Simret Araya Gebreegziabher, Oghenemaro Anuyah, Ronald A Metoyer, and Toby Jia-Jun Li. 2024. Comparing criteria development across domain experts, lay users, and models in large language model evaluation. arXiv preprint arXiv:2410.02054. — [arXiv:2410.02054](https://arxiv.org/abs/2410.02054) find
  37. Mahan Tafreshipour, Aaron Imani, Eric Huang, Eduardo Almeida, Thomas Zimmermann, and Iftekhar Ahmed. 2024. Prompting in the wild: An empirical study of prompt evolution in software repositories. arXiv preprint arXiv:2412.17298. — [arXiv:2412.17298](https://arxiv.org/abs/2412.17298) find
  38. Axel Van Lamsweerde. 2009. Requirements engineering: From system goals to UML models to software, volume 10. Chichester, UK: John Wiley & Sons. find
  39. Sanidhya Vijayvargiya, Xuhui Zhou, Akhila Yerukola, Maarten Sap, and Graham Neubig. 2025. Interactive agents to overcome ambiguity in software engineering. arXiv preprint arXiv:2502.13069. find
  40. Chenyang Yang, Rishabh Rustogi, Rachel Brower-Sinning, Grace Lewis, Christian Kaestner, and Tongshuang Wu. 2023. Beyond testers’ biases: Guiding model testing with knowledge bases using llms. In Findings of the Association for Computational Linguistics: EMNLP 2023, pages 13504–13519. find
  41. Mert Yuksekgonul, Federico Bianchi, Joseph Boen, Sheng Liu, Zhi Huang, Carlos Guestrin, and James Zou. 2024. Textgrad: Automatic" differentiation" via text. arXiv preprint arXiv:2406.07496.
  42. J Diego Zamfirescu-Pereira, Richmond Y Wong, Bjoern Hartmann, and Qian Yang. 2023. Why johnny can’t prompt: how non-ai experts try (and fail) to design llm prompts. In Proceedings of the 2023 CHI conference on human factors in computing systems, pages 1–21. find
  43. Zhiyuan Zeng, Yizhong Wang, Hannaneh Hajishirzi, and Pang Wei Koh. 2025. Evaltree: Profiling language model weaknesses via hierarchical capability trees. arXiv preprint arXiv:2503.08893. — [arXiv:2503.08893](https://arxiv.org/abs/2503.08893) find
  44. Michael JQ Zhang and Eunsol Choi. 2023. Clarify when necessary: Resolving ambiguity through interaction with lms. arXiv preprint arXiv:2311.09469. find
  45. Tong Zhang, Peixin Qin, Yang Deng, Chen Huang, Wenqiang Lei, Junhong Liu, Dingnan Jin, Hongru Liang, and Tat-Seng Chua. 2024. Clamber: A benchmark of identifying and clarifying ambiguous information needs in large language models. arXiv preprint arXiv:2405.12063.
  46. Chenyang Zhao, Xueying Jia, Vijay Viswanathan, Tongshuang Wu, and Graham Neubig. 2024. Self-guide: Better task-specific instruction following via self-synthetic finetuning. arXiv preprint arXiv:2407.12874. find
  47. Lianmin Zheng, Wei-Lin Chiang, Ying Sheng, Siyuan Zhuang, Zhanghao Wu, Yonghao Zhuang, Zi Lin, Zhuohan Li, Dacheng Li, Eric Xing, and 1 others. 2023. Judging llm-as-a-judge with mt-bench and chatbot arena. Advances in Neural Information Processing Systems, 36:46595–46623.
  48. Han Zhou, Xingchen Wan, Ruoxi Sun, Hamid Palangi, Shariq Iqbal, Ivan Vulić, Anna Korhonen, and Sercan Ö Arık. 2025. Multi-agent design: Optimizing agents with better prompts and topologies. arXiv preprint arXiv:2502.02533. — [arXiv:2502.02533](https://arxiv.org/abs/2502.02533) find
  49. Jeffrey Zhou, Tianjian Lu, Swaroop Mishra, Siddhartha Brahma, Sujoy Basu, Yi Luan, Denny Zhou, and Le Hou. 2023. Instruction-following evaluation for large language models. arXiv preprint arXiv:2311.07911.

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/yang-2026-what-prompts-don-t-say/