10 ChatGPT Prompts for Safety Professionals

Safety professionals do not need to become prompt engineers to use ChatGPT effectively.

But there is a major difference between asking:

“Make me a JSA.”

and giving ChatGPT enough information to produce something you can actually use. The following ten prompts are designed around that approach. They are not magic commands. They are starting templates. Copy them. Modify them. Add your organization's information. Then review and verify the output before using it.

Before You Use These Prompts

Remember one rule:

AI output is not automatically a fact.

ChatGPT can help you:

• organize,
• draft,
• brainstorm,
• challenge,
• compare,
• summarize, and
• ask better questions.

It can also be wrong.

For consequential safety work:

• verify regulatory claims against authoritative sources,
• confirm site conditions in the field,
• involve the people performing the work,
• protect confidential or restricted information,
• follow your organization's AI and information-security requirements, and
• keep a qualified person responsible for the final decision.

With that understood, here are ten practical prompts.

Prompt 1: Build a First-Draft Job Safety Analysis

A Job Safety Analysis—JSA or JHA—is a natural place to use AI because much of the early work involves organizing information. But ChatGPT has not seen your jobsite. The prompt therefore needs to prevent the AI from inventing conditions.

Copy/Paste Prompt

Context:
I am preparing a Job Safety Analysis for the following task:

[DESCRIBE THE TASK]

Known work conditions:

[LIST KNOWN CONDITIONS]

Equipment and tools:

[LIST EQUIPMENT]

People involved:

[LIST ROLES]

Known company requirements:

[PASTE OR SUMMARIZE VERIFIED REQUIREMENTS]

Role:
Act as a critical safety-planning assistant.

Task:
Break the work into logical task steps. For each step, identify potential hazards, existing controls I have provided, questions requiring field verification, and possible additional controls to consider.

Constraints:

• Do not invent site conditions.
• Do not invent company requirements.
• Do not claim a control exists unless I provided it.
• Do not make a final determination that the task is safe.
• Distinguish hazards from controls.
• Use the hierarchy of controls when suggesting possible improvements.
• Do not treat PPE as the default solution when a stronger control may be feasible.

Output:
Create a table with these columns:

1. Task step
2. Potential hazard
3. Existing verified control
4. Missing information
5. Field-verification question
6. Possible control improvement

Uncertainty:
Mark anything that cannot be determined from the information provided as:

[FIELD VERIFICATION REQUIRED] At the end, list the five most important questions I should ask the workers performing this task before finalizing the JSA.

Why This Prompt Works

It does not ask ChatGPT to create a finished JSA from thin air. It uses AI to prepare you for field verification. That distinction matters. The output becomes a working document for discussion—not evidence that the hazards have actually been assessed.

Prompt 2: Critique an Existing JSA

Sometimes the better use of AI is not creating something. It is challenging something that already exists.

Copy/Paste Prompt

I am going to provide an existing Job Safety Analysis. Act as a skeptical safety reviewer. Do not rewrite the JSA yet.

Review it for:

• missing task steps,
• vague hazard descriptions,
• controls that merely repeat the hazard,
• excessive reliance on PPE,
• controls dependent primarily on memory or attention,
• unclear responsibilities,
• weak verification methods,
• hazards created by simultaneous operations,
• changing conditions,
• assumptions about worker behavior,
• potential high-consequence exposures, and
• places where the written JSA may not reflect actual field work.

Separate your response into:

1. Strengths

What is already useful?

2. Questions

What needs clarification?

3. Weak Controls

Which controls may need strengthening and why?

4. Missing Information

What cannot be determined from the JSA?

5. Field Verification

What should I observe or ask workers before changing the document?

Do not invent conditions that are not in the JSA.

If a potential issue depends on information you do not have, label it:

[VERIFY IN FIELD]

Here is the JSA:

[PASTE JSA]

Why This Prompt Works

It asks ChatGPT to challenge the document instead of simply making the writing prettier. That is often a more valuable use of AI.

Prompt 3: Prepare for an Incident Investigation

AI can help organize investigation preparation. It should not determine the cause from incomplete information.

Copy/Paste Prompt

I am preparing for a workplace incident investigation. Below is the information currently available. [PASTE KNOWN INFORMATION] Act as an investigation-planning assistant.

Do not determine the root cause.

First separate the information into:

Confirmed Facts

Information currently supported by evidence.

Witness Statements

Information attributed to a person but not yet independently verified.

Interpretations

Conclusions or explanations contained in the material.

Assumptions

Things that appear to be assumed but are not established.

Unknowns

Important information that is currently missing.

Then help me develop an investigation plan covering:

1. Evidence that should be preserved or collected.
2. People who may need to be interviewed.
3. Neutral, open-ended interview questions.
4. Timeline gaps.
5. Equipment or physical conditions requiring examination.
6. Procedures, training or records requiring review.
7. Controls that should be evaluated.
8. Alternative explanations that should remain open.
9. Organizational or job-site conditions that may require examination.
10. Questions that must be answered before conclusions can be supported.

Do not invent evidence. Do not assume that a human error, procedure violation or equipment failure is the cause merely because one is mentioned.

Label unsupported possibilities:

[HYPOTHESIS — REQUIRES EVIDENCE]

Why This Prompt Works

The prompt forces ChatGPT to distinguish between:

what we know

and:

what we think.

That is fundamental to good investigation.

Prompt 4: Challenge an Investigation Conclusion

This may be one of the most useful prompts in the entire article. Ask AI to disagree with you.

Copy/Paste Prompt

Act as an independent critical reviewer of this workplace incident-investigation conclusion.

My proposed conclusion is:

[PASTE CONCLUSION]

The evidence I believe supports it is:

[PASTE EVIDENCE] Do not automatically agree with my conclusion.

Instead:

1. Identify which parts are directly supported by evidence.
2. Identify assumptions or logical jumps.
3. Identify evidence that would weaken the conclusion.
4. Provide reasonable alternative explanations that should be tested.
5. Identify organizational or job-site conditions that may have been overlooked.
6. Identify controls that should be evaluated.
7. Identify any use of hindsight bias.
8. Identify any conclusion that merely restates human error or a procedure violation without explaining it.
9. Tell me what additional evidence would strengthen or disprove the conclusion.
10. Rewrite the conclusion only if the available evidence supports doing so.

Separate:

• facts,
• interpretations,
• hypotheses, and
• unknowns.

Do not invent evidence.

Why This Prompt Works

People naturally become attached to their own explanations. AI can be useful as a structured devil's advocate. The objective is not to let ChatGPT determine the cause. The objective is to pressure-test your reasoning.

Prompt 5: Create a Better Toolbox Talk

Toolbox talks are often technically correct and painfully forgettable. AI can help make them more conversational.

Copy/Paste Prompt

I need a [5/10/15]-minute toolbox talk for:

[AUDIENCE]

Topic:

[TOPIC]

Actual work being performed:

[DESCRIBE WORK]

Verified company requirements:

[PASTE REQUIREMENTS]

Verified regulatory or technical information:

[PASTE SOURCE INFORMATION] Act as a safety communication assistant.

Create a toolbox talk containing:

1. A short realistic opening scenario.
2. Why this topic matters to this specific work.
3. No more than five key points.
4. The critical controls employees need to understand.
5. Three open-ended discussion questions.
6. One question asking workers what makes this task difficult.
7. One question asking what conditions could change today.
8. A short closing action statement.

Constraints:

• Use plain language.
• Do not invent company rules.
• Do not invent regulatory requirements.
• Do not use statistics unless they are contained in the verified source information.
• Do not make the talk sound like a legal memo.
• Do not make it a lecture.
• Do not use fear-based language.
• Identify anything that requires site-specific confirmation.

Mark those items:

[SITE-SPECIFIC VERIFICATION]

Why This Prompt Works

The toolbox talk starts with actual work rather than a generic topic. It also creates conversation. The supervisor should not simply read the AI output. The talk should become a discussion about today's work.

Prompt 6: Turn a Regulation Into a Field Checklist

This is a powerful use of AI—but also one that requires careful verification. Provide the actual regulatory text. Do not rely on the model to remember it correctly.

Copy/Paste Prompt

I am providing the exact text of a safety regulation or official requirement.

Jurisdiction:

[JURISDICTION]

Source:

[SOURCE NAME]

Regulatory section:

[SECTION]

Text:

[PASTE VERIFIED TEXT] Act as a regulatory-analysis assistant.

Using only the source text I provided, create a practical field-verification checklist.

For each requirement, provide:

1. Plain-language explanation.
2. What I should look for in the field.
3. What records or documentation may be relevant.
4. Questions I may need to ask.
5. Any condition, exception or qualifier contained in the source.
6. Any wording that should remain subject to professional or legal interpretation.

Do not add requirements that are not contained in the provided text. Do not infer that a requirement applies to my workplace unless applicability is established.

If the text does not answer a question, state:

[NOT ESTABLISHED BY PROVIDED SOURCE]

At the end, create a section titled:

Items Requiring Independent Verification

List anything I should verify against the official current regulation, agency guidance or other authoritative source before using the checklist.

Why This Prompt Works

You give ChatGPT the source. Its job is to organize and translate the source—not invent the regulation. You still verify the finished checklist against the original requirement.

Prompt 7: Prepare Learning Team Questions

Learning Teams require curiosity. ChatGPT can help facilitators prepare questions without scripting the entire conversation.

Copy/Paste Prompt

I am preparing to facilitate a Learning Team about:

[DESCRIBE WORK OR EVENT]

Reason for the Learning Team:

[PROACTIVE / POST-EVENT / CHANGE / RECURRING DIFFICULTY / OTHER]

What we currently know:

[PASTE INFORMATION] Act as a Learning Team preparation assistant. Do not determine causes. Do not propose corrective actions yet.

Help me prepare open-ended questions exploring:

1. How the work normally happens.
2. Normal sequence.
3. Variability.
4. Difficult conditions.
5. Worker adaptations.
6. Differences between written procedures and actual work.
7. What experienced workers know.
8. What normally makes the work successful.
9. Controls workers rely on.
10. Controls that are difficult to use.
11. Surprises or changing conditions.
12. How workers detect and recover from problems.
13. What worries the people performing the work.
14. What the organization may not understand about the task.

Avoid accusatory questions. Avoid repeatedly asking “why?”

Prefer prompts such as:

• “Walk me through…”
• “Tell me about…”
• “What happens when…”
• “What makes this difficult?”
• “What usually helps?”
• “What changes?”

Organize the questions into:

Opening

Normal Work

Variation and Adaptation

Controls

Successful Work

Difficult Conditions

Closing Questions

Do not enter problem-solving mode. The purpose of this prompt is to help me prepare to learn.

Why This Prompt Works

It keeps the first phase focused on understanding. Solutions can come later.

That helps prevent the facilitator from hearing the first problem and immediately deciding:

“We need retraining.”

Prompt 8: Analyze Safety Data Without Inventing Causes

AI can identify patterns. Patterns are not causes. This prompt makes that distinction explicit.

Copy/Paste Prompt

I am going to provide safety data. Act as a safety-data analysis assistant.

First inspect the data for:

• missing values,
• inconsistent categories,
• duplicates,
• date problems,
• unusual values,
• changes in reporting practices, and
• limitations that could affect interpretation.

Then analyze the data for potentially useful patterns involving:

• event type,
• severity,
• location,
• task,
• equipment,
• time period,
• recurring conditions, and
• other variables actually present in the dataset.

Important constraints:

• Do not infer causation from correlation.
• Do not invent missing exposure data.
• Do not assume that a higher count means higher risk if exposure is unknown.
• Do not describe a pattern as statistically meaningful unless appropriate analysis supports that statement.
• Clearly distinguish observed data from interpretation.

Organize your output into:

Data Quality Issues

Observed Patterns

Questions Raised by the Data

Possible Hypotheses to Investigate

Additional Data Needed

Recommended Field or Operational Follow-Up

Label hypotheses:

[HYPOTHESIS — NOT ESTABLISHED]

If exposure information is required to interpret a result, state:

[EXPOSURE DATA NEEDED]

Here is the data:

[UPLOAD OR PASTE DATA]

Why This Prompt Works

It prevents a common analytical mistake:

Pattern = cause.

AI can help you discover where to look. Investigation determines what the pattern means.

Prompt 9: Review a Safety Procedure

Procedures often look good on paper. The real question is whether people can use them.

Copy/Paste Prompt

I am providing a safety procedure. Act as a critical procedure-review assistant. Do not rewrite it yet.

Review the procedure for:

1. Ambiguous instructions.
2. Undefined terms.
3. Conflicting requirements.
4. Missing responsibilities.
5. Steps that rely heavily on memory.
6. Steps requiring subjective judgment without criteria.
7. Weak verification points.
8. Sequence problems.
9. Conditions the procedure appears to assume.
10. Potential mismatch between the written process and field reality.
11. Controls that may be difficult to execute.
12. Changes in conditions that the procedure may not address.
13. Places where workers may reasonably need to adapt.
14. Information that should be verified with the people performing the work.

For each issue, provide:

• exact section,
• concern,
• why it matters,
• question to ask workers, and
• possible improvement to evaluate.

Do not invent regulatory requirements. Do not rewrite the procedure until the operational questions have been answered.

End with:

Questions to Take to the Field

List the ten most important questions I should ask employees who use this procedure.

Here is the procedure:

[PASTE OR UPLOAD PROCEDURE]

Why This Prompt Works

The prompt does not assume the procedure is correct simply because it exists. It also does not assume the procedure is wrong. It generates questions that can be tested against actual work.

Prompt 10: Prepare a Supervisor Safety Conversation

Safety professionals spend significant time coaching supervisors. AI can help prepare those conversations.

Copy/Paste Prompt

I need to prepare a supervisor to discuss the following safety issue with an employee or crew:

[DESCRIBE ISSUE]

Known facts:

[LIST FACTS]

Unknowns:

[LIST UNKNOWNS]

Applicable company expectation:

[PASTE VERIFIED EXPECTATION] Act as a supervisor-coaching assistant.

Create a short conversation guide that helps the supervisor:

1. State the concern without exaggeration.
2. Ask the employee to explain what happened.
3. Understand how the work is normally performed.
4. Explore conditions that influenced the situation.
5. Clarify the company expectation.
6. Ask what makes compliance difficult.
7. Identify what support or system change may be needed.
8. Confirm the employee understands the expectation.
9. Establish an appropriate next action.
10. Close respectfully.

Do not assume misconduct. Do not promise that discipline will or will not occur. Do not invent facts. Do not write a speech.

Create:

Opening Statement

Questions to Ask

Information the Supervisor Should Listen For

Expectations to Clarify

Possible Follow-Up

Keep the conversation direct, respectful and focused on both accountability and learning.

Why This Prompt Works

A supervisor does not need a robotic script. They need a structure. The prompt helps create that structure without predetermining what the employee will say.

A Simple Formula for Writing Your Own Safety Prompts

You can build many useful prompts using six components:

1. Context

Tell ChatGPT what is happening.

“I am preparing a JSA for a crew replacing a hydraulic hose on mobile equipment.”

2. Role

Tell it how you want it to approach the task.

“Act as a critical safety-review assistant.”

3. Task

Tell it exactly what you want.

“Identify hazards and questions requiring field verification.”

4. Constraints

Tell it what not to do.

“Do not invent site conditions or regulatory requirements.”

5. Output

Tell it how to organize the response.

“Use a table with task, hazard, control and verification question.”

6. Uncertainty

Tell it what to do when information is missing.

“Mark missing information [FIELD VERIFICATION REQUIRED] rather than guessing.”

Put them together:

Context + Role + Task + Constraints + Output + Uncertainty

That formula works for far more than safety.

One More Instruction Worth Adding

For important work, add:

“Before answering, tell me what information you are missing that could materially change your response.”

This encourages ChatGPT to identify gaps instead of silently filling them with assumptions.

You can also say:

“Separate facts, assumptions, hypotheses and unknowns.”

Those two instructions can dramatically improve the quality of safety-related AI work.

Do Not Ask AI Only to Agree With You

Try prompts such as:

“Challenge my conclusion.”

“What am I missing?”

“What evidence would prove me wrong?”

“What assumptions am I making?”

“What would a worker performing this task see differently?”

“What alternative explanations should I investigate?”

“Which control in this plan depends most heavily on perfect human performance?”

That is where AI can become particularly useful. Not because it knows everything. Because it can help you examine your own thinking from another angle.

Keep the Human in the Loop

Every prompt in this article is deliberately designed around human review.

Notice the recurring instructions:

• verify,
• ask workers,
• check the field,
• identify assumptions,
• separate facts from hypotheses,
• use authoritative sources,
• do not invent conditions, and
• retain professional judgment.

Those are not limitations we should try to prompt our way around. They are part of responsible AI use. Safety decisions affect real people. The professional remains accountable for the work.

Use AI to Get Back Into the Field

The purpose of these prompts is not to create more documents.

The purpose is to reduce the administrative effort required to:

• prepare,
• organize,
• analyze,
• question, and
• communicate.

If ChatGPT saves you thirty minutes building the first draft of a JSA, use some of that time to talk with the people doing the job. If it saves you an hour organizing incident notes, use the time to examine the controls. If it helps build a toolbox talk, use the time to make the talk a real conversation.

AI should not pull safety professionals farther away from work. Used correctly, it can give them more time to go see it.

Supporting resources

Related commercial publications are disclosed under Authorship & disclosures below.