Proper hazard recognition is fundamental to effective safety management in construction workplaces. Nevertheless, poor hazard recognition levels are a widespread and persistent problem in the construction industry. For example, recent investigations have demonstrated that a significant number of workplace hazards often remain unrecognized in construction workplaces. These unrecognized workplace hazards often remain unmanaged and can potentially translate into devastating and unexpected safety incidents. Therefore, interventions targeted at improving hazard recognition levels are foundational to enhancing safety management in construction workplaces. The main objective of the current investigation was to examine if ChatGPT, a language model recently launched by OpenAI, can aid hazard recognition when integrated into the curriculum of students pursuing a career in the construction industry. The investigation was carried out as an experimental effort with 42 students enrolled in the construction program at a major state university in the United States. First, prior to the introduction of ChatGPT as an intervention, the pre-intervention hazard recognition ability of the students was measured. Next, ChatGPT and its capabilities were introduced to the students in a classroom setting. Guidance was also offered on how the students could leverage ChatGPT to aid hazard recognition efforts. Finally, the post-intervention hazard recognition ability of the students was measured and compared against their earlier performance. The result suggests that ChatGPT can be leveraged to improve hazard recognition levels. Accordingly, integrating ChatGPT as part of safety education and training can yield benefits and prepare the next generation of construction professionals for industry success.

The recent emergence of multimodal large language models (LLMs) has introduced new opportunities for improving visual hazard recognition on construction sites. However, despite growing interest in their applications, there has been limited investigation into how different LLMs perform in safety-critical visual tasks within the construction domain. To address this gap, this study conducts a comparative evaluation of five state-of-the-art LLMs: GPT-4o, GPT-5, GPT-4.1, Claude 4.1 Opus, and Gemini 2.5 Pro, to assess their ability to identify potential hazards from real-world construction images. Each model was tested under three prompting strategies: zero-shot, few-shot, and chain-of-thought (CoT). Quantitative analysis was performed using precision, recall, and F1-score metrics across all conditions. Results reveal that prompting strategy significantly influenced hazard recognition performance. CoT prompting consistently produced the highest accuracy across models, with GPT-5 and GPT-4.1 achieving superior scores in most settings. The results suggest that structured prompt design can elevate LLM performance to significant levels, offering a cost-effective pathway for small firms and contractors to improve hazard recognition. Furthermore, LMM outputs can be repurposed as explanatory content for safety training and toolbox talks, supporting long-term safety culture development.

The recent emergence of multimodal large language models (LLMs) has introduced new opportunities for improving visual hazard recognition on construction sites. Unlike traditional computer vision models that rely on domain-specific training and extensive datasets, modern LLMs can interpret and describe complex visual scenes using simple natural language prompts. However, despite growing interest in their applications, there has been limited investigation into how different LLMs perform in safety-critical visual tasks within the construction domain. To address this gap, this study conducts a comparative evaluation of five state-of-the-art LLMs: Claude-3 Opus, GPT-4.5, GPT-4o, GPT-o3, and Gemini 2.0 Pro, to assess their ability to identify potential hazards from real-world construction images. Each model was tested under three prompting strategies: zero-shot, few-shot, and chain-of-thought (CoT). Zero-shot prompting involved minimal instruction, few-shot incorporated basic safety context and a hazard source mnemonic, and CoT provided step-by-step reasoning examples to scaffold model thinking. Quantitative analysis was performed using precision, recall, and F1-score metrics across all conditions. Results reveal that prompting strategy significantly influenced performance, with CoT prompting consistently producing higher accuracy across models. Additionally, LLM performance varied under different conditions, with GPT-4.5 and GPT-o3 outperforming others in most settings. The findings also demonstrate the critical role of prompt design in enhancing the accuracy and consistency of multimodal LLMs for construction safety applications. This study offers actionable insights into the integration of prompt engineering and LLMs for practical hazard recognition, contributing to the development of more reliable AI-assisted safety systems.

Severe head injuries can leave construction workers with permanent disability, prolonged work absence, or high medical costs. Although case counts identify sectors with many reports, workforce comparisons may identify different prevention priorities. This study examined 1485 OSHA Severe Injury Reports with head-region codes across 15 U.S. construction sectors from 2015 to 2024 and linked them with Quarterly Census of Employment and Wages data. Ten pairwise tests examined relationships among construction sector, event type, injury source, head region, and injury nature, followed by intracranial proportions within common event-source combinations. Three findings emerged. (1) Building Equipment Contractors recorded the most cases but fell below workforce expectation. (2) Structural Steel and Precast Concrete Contractors had the largest workforce-relative observed-to-expected ratio, followed by Masonry and Framing Contractors. (3) Transportation incidents were strongly associated with vehicle sources, whereas lower-level falls were associated with intracranial injuries. Within struck-by reports, machinery showed a numerically higher intracranial proportion than tools or equipment; these proportions were descriptive rather than statistically ranked. These results can help agencies allocate prevention resources across sectors and help contractors focus reviews on traffic movement, elevated access, material handling, and electrical work.