Nwokeji, Joshua C., Ikwunne, Tochukwu A., Nkwo, Makuochi S. ORCID: https://orcid.org/0000-0002-9774-9602 and Yeerbo, Meiyeer (2026) Evaluating prompt engineering techniques for image-based classification of dark patterns with GPT-5. In: Artificial Intelligence in HCI. 7th International Conference AI-HCI 2026 Held as Part of the 28th HCI International Conference HCII 2026. Montreal, QC ,Canada, July 26-31, 2026. Proceedings Part III. Lecture Notes in Computer Science ((LNAI) -International Conference on Human-Computer Interaction, 16745 . Springer Nature, Cham, Switzerland, pp. 507-526. ISBN 978-3032308481 (doi:10.1007/978-3-032-30849-8_30)

	PDF (AAM Book Chapter) 54417 NKWO_Evaluating_Prompt_Engineering_Techniques_For_Image-based_Classification_(AAM BOOK CHAPTER)_2026.pdf - Accepted Version Restricted to Repository staff only until 9 July 2027. Download (587kB) | Request a copy
	PDF (Conference Proceedings) 54417 NKWO_Evaluating_Prompt_Engineering_Techniques_For_Image-based_Classification_(CONFERENCE PROCEEDINGS)_2026.pdf - Published Version Restricted to Repository staff only Download (70MB) | Request a copy

Abstract

Even though there is a broad awareness of foundational MLLM prompting techniques within the HCI community, there is a dearth of empirical evidence about which prompt technique gives the best performance in image-based dark pattern classification. Therefore, this research aims to evaluate prompting techniques and investigate their differential performance for binary classification of dark patterns from UI images. We conducted our evaluation using a labeled dataset of UI images sourced from the AIDUI study, comprising both dark pattern and non–dark pattern instances. All experiments were performed using GPT-5, accessed via API without fine-tuning. Model performance was measured using standard classification metrics, including accuracy, precision, recall, and F1-score. The results indicate that prompting techniques that combine in-context learning (via example-based conditioning) and structured stepwise reasoning tend to achieve superior performance in image-based dark pattern classification. This study provides novel empirical evidence on classical prompting techniques that can widen the scope of image-based dark pattern detection and classification, and offers recommendations, informing responsible AI use in interface evaluation and design evaluation.

Item Type:	Conference Proceedings
Title of Proceedings:	Artificial Intelligence in HCI. 7th International Conference AI-HCI 2026 Held as Part of the 28th HCI International Conference HCII 2026. Montreal, QC ,Canada, July 26-31, 2026. Proceedings Part III
Uncontrolled Keywords:	Artificial Intelligence, prompt engineering, prompt design, user experience, Human-Computer Interaction, UX, AI, LLM, dark patterns · HCI · Large Language Model · LLM · Image Classification · Ethical UX
Subjects:	Q Science > Q Science (General) Q Science > QA Mathematics Q Science > QA Mathematics > QA75 Electronic computers. Computer science
Faculty / School / Research Centre / Research Group:	Faculty of Engineering & Science Faculty of Engineering & Science > School of Computing & Mathematical Sciences (CMS)
Last Modified:	18 Sep 2026 11:25
URI:	https://gala.gre.ac.uk/id/eprint/54417

Actions (login required)

View Item

Downloads