Gains:
- Ability to build construction site risk assessment into AI with hierarchy of hazard, probability, severity and precautions
- Ability to verify the risk analysis, OHS instruction and toolbox speech produced by AI with legislation and field reality
- Ability to understand and apply the legal responsibility of OHS documents and that AI output requires expert approval
Construction is one of the sectors where work accidents occur most frequently and most severely; Falling from heights, falling materials, landslides, electrical and construction equipment accidents are the main causes of loss of life. Occupational Health and Safety (OHS) is the discipline of systematically identifying these hazards and taking precautions, and it is legally mandatory. OHS documentation (risk assessment, safety instructions, toolbox conversations, emergency plans) are documents that carry legal responsibility. AI is a powerful aid in quickly producing drafts of these documents, expanding hazard lists and simplifying instructions. But in OHS, a missing or incorrect precaution directly means life safety; AI output must be verified with field reality by an OHS expert/engineer. In this unit, we will see how to produce OHS documentation safely with AI.
Establishing Risk Assessment Correctly
The basic logic of risk assessment: identify hazards for each work step, evaluate the risk (probability × severity), take action and re-evaluate the risk after action. When setting up risk analysis to AI, provide the following information:
- Work step/activity: What is being done? (mold removal, working at height, excavation, lifting loads with a crane).
- Danger: What could hurt? (fall, crush, dent, electricity).
- Likelihood and severity: How frequent, how severe? (on a scale).
- Hierarchy of measures: Elimination > substitution > engineering measure (guardrail, barrier) > administrative measure (instruction, training) > personal protective equipment (PPE). This order is important; PPE is a last resort.
The hierarchy of precautions is a critical concept: completely eliminating a hazard is much more effective than simply providing a hard hat. AI sometimes jumps straight to PPE and skips high-level precautions; It is necessary to fix this.
Risk scoring also requires attention. Risk is often reduced to a number as the probability multiplied by the severity (e.g., on a scale of 1-5, probability 4, severity 5 → risk 20, high). But this number can be misleading: a rare but fatal hazard (low probability, very high severity) can never be neglected, even if it gives a low multiplier. AI can mechanically generate scores and downplay a high-severity hazard due to low probability. So don't blindly trust the multiplication result; Meet every danger with the potential for loss of life with serious precautions, regardless of the score. Be sure to also fill in the "remaining risk" column after the precaution: whether the risk has decreased to an acceptable level after the precaution has been taken is the main purpose of the risk assessment and this judgment belongs to the OHS expert.
Attention: OHS risk assessment is a legal document and must be made according to the actual site and actual manufacturing method. A “generic” risk analysis generated by AI is the starting point; Only an OHS specialist who examines site-specific hazards (ground of that construction site, neighboring structure, weather, equipment) on site can add them. Signing and hanging the AI printout is a grave liability.
Weak Prompt / Strong Prompt
WEAK:"Write a risk analysis for the construction site."(No activity, method, precaution hierarchy, site condition.)STRONG:"Produce a risk assessment draft for the 'formwork removal at height' activity on a reinforced concrete building construction site. Table:Danger | Possible outcome | Probability | Severity | Precaution | Residual risk.Rules:- Sort the measures by HIERARCHY (elimination first/engineering measure, PPE first at the end) - Don't just jump to PPE; prioritize high-level measures such as guardrails, safety nets, platforms - Put a reminder note for ME to add hazards that may be specific to the site. I will complete this draft according to the site condition as an OSH expert.
Validating AI Output
Hierarchy of measures. Does the AI recommend high-level (engineering) measures first, or does it immediately jump to PPE? Collective measures such as guardrails, platforms, safety nets etc. should take precedence over individual PPE.
Field reality. AI lists general hazards well, but it cannot know risks specific to that construction site (proximity to neighboring building, underground line, narrow work area). The expert adds these.
Regulatory compliance. Do the measures comply with applicable OSH legislation and regulations? AI may provide outdated or general information; Confirmation from current legislation is required.
Completeness. AI may skip some work steps or hazards. Check that the activity list is complete.
control
What to look for
frequent error
Hierarchy of measures
Is collective action before PPE?
Jumping straight to PPE
field reality
Are there any risks specific to the construction site?
general stay
legislation
Does it comply with current OSH legislation?
Outdated/missing information
completeness
Are all work steps covered?
skip step
Responsibility
Has expert approval been obtained?
use unsigned
Three Mini Cases
Case 1 – Omitted parent measure. AI directly says “safety belt to be worn” (PPE) for working at height, but skips collective measures such as guardrails and safety nets. The OHS expert first mandates edge protection (guardrails) and adds PPE as a complementary measure. The hierarchy fix produces a much safer plan.
Case 2 – Site specific risk. AI's excavation risk analysis includes the general collapse hazard, but it does not know the base load of an old building adjacent to this construction site. The expert adds shoring (excavation support) and monitoring measures in excavation close to the neighboring structure. The on-site inspection captures the critical risk that the general outline misses.
Case 3 – Toolbox conversation. For the daily toolbox (short field safety) conversation, the crew chief asks the AI for a short text specific to that day's job (crane lifting). AI produces a simple, understandable outline; The chief adds and uses site-specific warnings (wind, not standing under load). A low-risk, expediting use.
Case 4 – Misleading score. In a risk analysis, the AI scores the risk low, saying “probability is low” for a pedestrian crossing near a deep excavation. However, the OSH expert sees that the possible outcome (being buried under the rubble) is fatal; This danger, which has a very high severity, cannot be ignored despite the low probability. The expert adds physical barriers and area closure measures by looking at the weight of the result, not the multiplication result. Lesson: potentially lethal hazards require serious action, regardless of score.
Copiable prompt templates
PRECAUTION HIERARCHY CONTROL PROMPT:"Reorder the measures recommended in the following risk assessment according to the hierarchy: 1) elimination 2) substitution 3) engineering/collective measure (guardrail, net, platform) 4) administrative measure 5) PPE. Mark places that jump directly to PPE and recommend high-level action. Text: [paste]"
TOOLBOX TALK PROMP: "Write a 5-minute, concise draft of a toolbox (daily field safety) talk for today's construction site activity '[activity]'. The 3 main hazards, precautions to be taken and what is expected from the team. Leave a blank item for ME to add site-specific warnings."
SITE-SPECIFIC RISK REMINDER PROMPT: "You have given a general risk analysis. Now, list the risk categories that may be specific to this construction site and that you may not know (neighbouring structure, ground/water, underground lines, narrow space, air, traffic) as headings so that we can examine each of them on site and fill them in."
Common mistakes
- Bypassing the precautionary hierarchy and directly recommending/accepting PPE.
- Not complementing AI's general risk analysis with site-specific hazards.
- Not confirming measures in accordance with current OSH legislation.
- Leaving some work steps or hazards missing.
- Signing and using the AI output without the approval of an OHS expert.
- Assigning risk scores (likelihood/severity) from realistic assessment.
In summary
- OHS documents carry legal responsibility; AI output is invalid without expert approval.
- Establish the risk assessment with a hierarchy of hazard, probability, severity and precaution.
- In the hierarchy of measures, collective/engineering measures come before PPE.
- AI lists common dangers well; Expert adds site-specific risks.
- Confirm the measures from current OSH legislation.
- AI speeds up low-risk text like Toolbox and instructions; Expert in risk analysis confirms.
Application task
Select a construction site activity (e.g. working at height, excavation, craning). Ask the AI to draft a risk assessment with columns for hazard, probability, severity, precaution, and residual risk; Tell him to list the measures in a hierarchy and set reminders for site-specific points. Then: (1) check that the measures comply with the hierarchy (collective first), (2) add at least two site-specific risks, (3) confirm the measures with current legislation, (4) check for any missing work steps. Note that the document requires expert approval.
checklist
- [ ] I defined the activity with danger, probability, severity, and precaution.
- [ ] I checked that the measures comply with the hierarchy (collective/engineering first).
- [ ] I added site-specific hazards to the draft.
- [ ] I have confirmed the measures from the current OSH legislation.
- [ ] I verified that all job steps were covered.
- [ ] I evaluated the risk scores realistically.
- [ ] I have subjected the document to the approval of the OHS expert/engineer.