Truace
Subscribe
64/100

Moderate evidence

Cite
More
Export CSVExport JSONEmailLinkedInX

In a 51-person test, AI helped spot more construction site hazards than checklists

Study title: Assessing the Effectiveness of Generative Artificial Intelligence in Hazard Identification on Construction Sites

Evidence type
Controlled or validated comparison
Sample
n = 51
Sources
1 sourcePeer-reviewed
Published
2 min read
Truvace record
TRV-2026-2567cite or check it

Study title: Assessing the Effectiveness of Generative Artificial Intelligence in Hazard Identification on Construction Sites

A within-subjects experiment with 51 participants compared hazard identification on construction sites using traditional approaches against generative AI support. Scored against expert-verified ground truth, the AI-assisted condition produced significantly higher F1 scores than the traditional approaches at both the pre-training and post-training stages (p < 0.05).

Read the study abstract

The construction industry remains one of the most perilous, where hazard identification is often inconsistent. Hazards are still missed when teams rely mainly on traditional approaches like checklists, Job Safety Analysis (JSAs)/Job Hazard Analysis (JHAs), and individual experience. This study evaluated whether a GEN-AI-assisted approach improved hazard identification performance compared with traditional approaches, using expert-verified ground truth for scoring. A quantitative within-subjects experiment was co…

The gain

In an experiment with 51 participants, generative AI support produced significantly higher hazard identification F1 scores than traditional checklist and job hazard analysis approaches, both before and after a short training session.

In brief

Construction remains one of the most perilous industries, and hazards are still missed when teams rely mainly on checklists, job safety analyses and individual experience. The study suggests AI support could make proactive hazard prevention more consistent, though the authors say final outputs need human validation.

The short hazard identification training delivered between stages added little. The overall intervention effect was small and the method-by-intervention interaction negligible, with no statistically significant change over time (p > 0.05).

Main points

  1. Fifty-one participants each completed hazard identification in four conditions: traditional before training, generative AI before training, traditional after training and generative AI after training.
  2. Effectiveness was measured with the F1 score, which combines precision and recall, scored against expert-verified ground truth.
  3. Paired-sample t-tests found generative AI support scored significantly higher than traditional approaches at both stages (p < 0.05).
  4. A 2 × 2 repeated measures ANOVA confirmed a strong overall effect of method.
  5. The overall effect of the training intervention was small, and the method-by-intervention interaction was negligible.
  6. The authors conclude that final AI outputs need human validation.

The full story

Participants worked through hazard identification under four conditions, with a short training session on hazard identification delivered between the pre- and post-stages. Traditional approaches included checklists and Job Safety Analysis or Job Hazard Analysis.

Scores were built on the F1 measure, balancing how many real hazards were found against how many flagged items were correct. Paired t-tests compared traditional and AI-assisted performance at each stage, and a repeated measures ANOVA tested the overall pattern.

AI support outperformed traditional methods at both stages, while the training session produced no statistically significant change over time. The authors frame the result as empirical evidence that generative AI strengthens proactive prevention, with human validation still required.

Sources

1 on file
  1. Buildingsdoi:10.3390/buildings16122401
    Peer-reviewed2026-06-17

Talk it over

Four questions to start onWho gets this benefit, and who was left out of the study or story?

Start with a question

In a 51-person test, AI helped spot more construction site hazards than checklistshttps://truvace.com/item/assessing-the-effectiveness-of-generative-artificial-intelligence-in-hazard-iden

  1. Who gets this benefit, and who was left out of the study or story?

    The gain In an experiment with 51 participants, generative AI support produced significantly higher hazard identification F1 scores than traditional checklist and job hazard analysis approaches, both before and after a short training session.

    Answer in the thread

  2. How do we know? What can this kind of evidence show, and what can’t it?

    The evidence This rests on a controlled comparison with a sample of 51, from Buildings.

    Answer in the thread

  3. What does it leave open? What else would you want to know before acting on it?

    The record’s own caveat The finding applies to this dataset and experimental setup, the training session showed no statistically significant change over time, and the authors say final AI outputs need human validation.

    Answer in the thread

  4. Have you seen AI used this way where you study, work or live? Did it go the same way?

    Answer in the thread

The questions are the same on every record. The details under them are taken from this record as written.

What readers say

More from Labor