How Safety Professionals Can Use ChatGPT

Safety professionals spend an enormous amount of time working with information.

We:

• write procedures,
• review regulations,
• prepare training,
• analyze incidents,
• develop Job Safety Analyses,
• build presentations,
• examine data,
• communicate with leaders,
• answer questions,
• prepare toolbox talks, and
• turn complicated technical information into something people can actually use.

ChatGPT can help with many of those tasks.

But there is an important distinction:

ChatGPT can assist safety professionals. It should not replace professional judgment.

That distinction should guide everything that follows.

Used poorly, it can produce a polished answer that is incomplete, outdated or simply wrong.

The objective is therefore not:

Let AI do safety.

The objective is:

Use AI to increase the capacity of the safety professional while keeping the professional responsible for the work.

What Is ChatGPT?

ChatGPT is a generative artificial-intelligence system developed by OpenAI.

At a basic level, it can interpret instructions and generate responses using patterns learned from large amounts of information.

Depending on the version, account, available tools and workspace configuration, ChatGPT may also be able to work with things such as:

• uploaded documents,
• spreadsheets,
• PDFs,
• images,
• current web information,
• connected information sources, and
• data-analysis tools.

That creates opportunities far beyond simply asking it questions.

For safety professionals, the better way to think about ChatGPT is not as an electronic encyclopedia.

Think of it as an AI work assistant.

It can help you:

• draft,
• organize,
• compare,
• summarize,
• question,
• restructure,
• analyze,
• critique, and
• communicate.

You remain responsible for determining whether the result is correct and appropriate.

The First Rule: AI Output Is Not Evidence

This may be the most important rule in this article.

ChatGPT can produce an answer that sounds authoritative and is wrong.

It can potentially:

• state an incorrect fact,
• misunderstand a regulation,
• omit an exception,
• apply the wrong jurisdiction,
• use outdated information,
• invent a reference,
• misinterpret a document, or
• confidently answer a question that did not contain enough information.

The technical term often used for fabricated or incorrect AI output is hallucination.

The professional response is straightforward:

Verify important claims.

If ChatGPT tells you what an OSHA standard requires, verify the requirement against the actual standard and applicable official guidance.

If it gives you a regulation number, verify it.

If it cites a study, open the study.

If it gives you a statistic, check the source.

If it analyzes your company procedure, compare the analysis against the actual procedure.

ChatGPT can help you reach the source.

It should not become the source merely because the answer sounds good.

Use AI to Accelerate Judgment, Not Replace It

A safety professional brings something AI does not have:

responsibility for the real-world decision.

You understand:

• the organization,
• the operation,
• the equipment,
• the workforce,
• company expectations,
• applicable regulations,
• site conditions,
• history,
• operational constraints, and
• consequences.

AI may help you think.

It does not remove your responsibility to decide.

That suggests a useful operating model:

Professional judgment + verified source information + AI assistance

rather than:

AI output = answer

Where Can Safety Professionals Use ChatGPT?

There are many useful applications.

Here are some of the strongest.

1. Drafting Safety Communications

Safety professionals write constantly.

ChatGPT can help draft:

• safety alerts,
• supervisor messages,
• toolbox talks,
• leadership communications,
• employee notices,
• training introductions,
• meeting summaries,
• FAQs, and
• campaign messaging.

Suppose you need a short message about distracted driving.

Instead of starting with a blank page, provide:

• the audience,
• the issue,
• your company requirement,
• the desired length,
• the tone, and
• the action employees should take.

Then ask ChatGPT for a draft.

You review it.

Correct it.

Add company-specific language.

Verify technical claims.

Now AI has reduced drafting time without taking ownership of the message.

2. Turning Technical Language Into Plain Language

Safety professionals frequently work between two worlds.

One contains:

• regulations,
• standards,
• technical manuals,
• engineering language, and
• legal requirements.

The other contains people who need to know:

“What does this mean for the job I am doing today?”

ChatGPT can help translate complex language into clearer explanations.

For example, you can provide a verified regulatory passage and ask:

Explain this in plain language for a frontline supervisor. Do not add requirements that are not contained in the source. Identify anything that still requires professional interpretation.

That is much safer than asking:

What does OSHA require?

The first prompt provides the authoritative source.

The second asks the AI to supply it from memory.

Whenever practical, give the model the source you want it to use.

3. Building Toolbox Talks

A toolbox talk should be more than a paragraph someone reads to employees.

ChatGPT can help build a discussion around:

• the actual task,
• the hazard,
• the critical controls,
• a realistic scenario,
• questions for workers, and
• actions employees should take.

For example:

Create a 10-minute toolbox talk for delivery drivers loading mobile equipment onto trailers. Base it only on the company requirements I provide. Include a realistic opening scenario, three discussion questions, the critical controls and a short closing commitment. Flag anything requiring site-specific verification.

That is far more useful than:

Write me a toolbox talk about loading.

Better input usually produces better output.

4. Preparing a Job Safety Analysis

ChatGPT can assist with Job Safety Analysis or Job Hazard Analysis preparation.

The word assist matters.

AI has not stood at your worksite.

It may not know:

• the exact machine,
• the ground conditions,
• the configuration,
• the crew,
• simultaneous operations,
• environmental conditions,
• maintenance status, or
• what changed today.

But it can help structure the analysis.

For example, provide the known task information and ask ChatGPT to:

• break the task into logical steps,
• brainstorm potential hazards,
• identify questions requiring field verification,
• challenge weak controls,
• organize controls using the hierarchy of controls, and
• identify assumptions.

Then take that work to the field.

Talk with the people doing the job.

Observe the task.

Verify the conditions.

Revise the JSA.

AI can help you prepare for the conversation.

It should not eliminate the conversation.

5. Preparing for Incident Investigations

ChatGPT can be useful before and during the analytical stages of an investigation.

You can ask it to help:

• organize known facts,
• separate facts from assumptions,
• develop a timeline,
• identify missing information,
• generate neutral interview questions,
• compare competing hypotheses,
• identify controls requiring examination, and
• challenge premature conclusions.

For example:

Here are the confirmed facts from an incident. Separate confirmed facts, witness statements, interpretations and unknowns. Do not determine a cause. Identify questions the investigation still needs to answer.

That can be extremely useful.

Notice what the prompt does not say:

Tell me what caused this accident.

The objective is to support investigation—not outsource it.

6. Challenging Your Conclusions

This is one of my favorite uses of AI.

Do not only ask ChatGPT to agree with you.

Ask it to challenge you.

Suppose your investigation concludes:

The employee failed to follow the procedure.

Ask:

Challenge this conclusion. What additional evidence would be needed before we could support it? What alternative explanations should be tested? What organizational or job-site conditions might also need examination?

Now AI becomes a thinking partner rather than a writing machine.

You can do the same thing with:

• corrective actions,
• policies,
• risk assessments,
• presentations,
• audit findings, and
• proposed controls.

Ask:

What am I missing?

That question can be more valuable than:

Does this look good?

7. Reviewing Procedures

Provide a procedure and ask ChatGPT to examine it from several perspectives.

For example:

Review this procedure for ambiguity, conflicting instructions, undefined responsibilities, steps that depend heavily on memory, unclear verification points and places where the written process may be difficult to execute in the field. Do not rewrite it yet. Give me questions to investigate with workers.

That creates a useful review framework.

Then go ask the workers.

AI identifies questions.

People provide operational reality.

8. Preparing Learning Team Questions

ChatGPT can help prepare for a Learning Team.

Describe:

• the work,
• what prompted the Learning Team,
• what is already known, and
• what you want to understand.

Then ask for open-ended questions covering:

• normal work,
• variability,
• adaptations,
• difficult conditions,
• successful work,
• controls,
• surprises, and
• improvement opportunities.

Again, the AI does not conduct the Learning Team.

It helps the facilitator prepare.

9. Analyzing Safety Data

When appropriate data tools are available, ChatGPT can assist with structured data such as spreadsheets.

Potential applications include:

• organizing incident data,
• identifying recurring categories,
• summarizing observations,
• comparing periods,
• calculating rates,
• creating tables,
• exploring patterns, and
• developing charts.

But there is an important warning.

A pattern is not automatically a cause.

Suppose injuries increase on Mondays.

That does not prove Monday causes injuries.

You still need to investigate:

• exposure,
• hours worked,
• staffing,
• work mix,
• reporting practices,
• seasonality,
• sample size, and
• other possible explanations.

AI can help reveal a pattern.

The safety professional determines what the pattern means.

10. Preparing Leadership Presentations

Safety professionals often need to turn complicated information into an executive-level message.

ChatGPT can help transform:

• investigation findings,
• audit results,
• risk information,
• project updates,
• safety metrics, and
• technical recommendations

into concise leadership communication.

You can specify:

The audience is senior operations leadership. Give me a five-slide structure: issue, evidence, operational significance, recommended actions and decisions needed. Separate confirmed evidence from interpretation.

That can dramatically reduce the time required to structure a presentation.

You still own the facts and recommendation.

11. Adapting Training for Different Audiences

The same safety topic may need different explanations for:

• new employees,
• experienced employees,
• supervisors,
• managers,
• executives,
• contractors, or
• technical specialists.

ChatGPT can help adapt the same verified source material for each audience.

That is particularly useful when the underlying technical requirement must remain consistent but the explanation needs to change.

For example:

Using only the source material below, create three explanations of hazardous-energy verification: one for a new employee, one for an experienced technician and one for an operations manager. Keep the technical requirement consistent across all three versions.

That is a strong use of AI.

12. Creating Better Questions

Sometimes the most valuable output is not an answer.

It is a question.

Before:

• visiting a site,
• conducting an audit,
• interviewing an employee,
• facilitating a Learning Team,
• meeting with leadership, or
• reviewing a new operation,

give ChatGPT the context and ask:

What questions should I be asking that I may not be asking?

This can broaden your thinking.

You do not have to accept every question.

The value is in exposing possibilities.

Use ChatGPT as a Hypothesis Generator

This is an important mindset.

ChatGPT is often excellent at generating possibilities.

It can suggest:

• potential contributing factors,
• possible hazards,
• questions,
• alternative explanations,
• possible controls, and
• areas requiring investigation.

Treat those as hypotheses.

Not facts.

If AI suggests:

“Fatigue may have contributed.”

that does not mean fatigue contributed.

It means:

Fatigue may be something worth examining if the evidence supports doing so.

This distinction protects the integrity of safety work.

Separate Facts, Assumptions and Unknowns

One of the most useful instructions you can give ChatGPT is:

Separate confirmed facts, assumptions, interpretations and unknowns.

That is useful for:

• investigations,
• risk assessments,
• project planning,
• regulatory research, and
• operational decisions.

AI-generated language can blur these categories if you do not ask it to distinguish them.

Make the distinction explicit.

Ask AI to Show Its Uncertainty

Another useful instruction is:

If you are uncertain, say so. Do not invent missing information.

You can go further:

Identify which parts of your response require verification and tell me what source would be appropriate for verification.

That encourages a more disciplined workflow.

It does not guarantee correctness.

You still verify.

Regulatory Research Requires Special Care

Safety professionals work in a regulated environment.

That makes regulatory accuracy particularly important.

ChatGPT can help you:

• identify potentially relevant standards,
• explain regulatory language,
• compare requirements,
• organize research,
• develop questions, and
• locate areas requiring deeper review.

But regulatory answers should be checked against authoritative sources.

For U.S. occupational safety questions, that may include:

• the actual OSHA regulation,
• official OSHA interpretations,
• applicable state-plan requirements,
• other relevant federal agencies,
• consensus standards when applicable, and
• your organization's legal or compliance resources when interpretation is consequential.

Do not rely on:

“ChatGPT said OSHA requires it.”

Verify.

Current Information Requires Current Sources

AI systems can have knowledge limitations.

Regulations change.

Guidance changes.

Standards change.

Technology changes.

Agency interpretations change.

When recency matters, use current-source capabilities where available and inspect the underlying sources.

Ask for:

• the source,
• publication date,
• jurisdiction,
• applicable regulatory section, and
• any limitations.

Then open the source.

The citation is useful.

The source is what matters.

Be Careful With Confidential and Sensitive Information

Before entering workplace information into any AI system, understand:

• your organization's AI policy,
• the account or workspace you are using,
• applicable privacy and confidentiality requirements,
• contractual obligations,
• data-retention requirements, and
• what information is permitted to be shared.

Do not casually paste:

• employee medical information,
• personally identifiable information,
• confidential investigation records,
• protected customer information,
• trade secrets,
• attorney-client communications, or
• other restricted information

into an AI system simply because it would make the prompt easier.

Your organization should establish rules governing what can and cannot be used.

Where possible, remove unnecessary identifiers and provide only the information needed for the task.

Consumer and Organizational Accounts May Have Different Data Controls

Do not assume every ChatGPT account handles organizational information the same way.

Data controls and administrative protections can differ depending on:

• account type,
• workspace,
• organizational configuration, and
• user settings.

Before using AI with company information, understand the environment in which you are working.

For organizational use, follow your employer's approved technology, security and data-governance requirements.

Human Review Is Especially Important for Consequential Decisions

AI can help organize evidence related to people.

That does not mean it should make employment or other high-impact decisions for you.

For example, ChatGPT may help summarize:

• training records,
• interview notes,
• investigation evidence, or
• documented performance information

when organizational policies allow.

But the responsible human decision-maker must review the underlying evidence, applicable policies, context and legal requirements.

AI assistance should not become an automated substitute for accountable human judgment.

Do Not Upload a Document and Assume AI Read It Perfectly

File analysis can be powerful.

It can also fail.

Complex formatting, scanned pages, images, unusual tables and large documents can create extraction problems.

If a specific number, sentence, requirement or table matters:

verify it against the original file.

A useful instruction is:

Quote the exact section you relied on and identify the page or location so I can verify it.

Then verify it.

Better Prompts Produce Better Work

You do not need to become a “prompt engineer.”

But you should provide useful context.

A strong safety prompt usually contains six things:

Context

What is happening?

Role

What perspective should ChatGPT take?

Task

What do you want it to do?

Constraints

What should it not do?

Output

What format do you want?

Uncertainty

What should happen when information is missing?

For example:

Context: I am preparing a pre-job discussion for a crew loading construction equipment onto a trailer.

Role: Act as a critical safety-review assistant.

Task: Review the task description and identify hazards, critical-control questions and areas requiring field verification.

Constraints: Do not invent site conditions or company requirements. Do not state that a control exists unless I provide it.

Output: Create a table with task step, potential hazard, existing control, questions to verify and possible improvements.

Uncertainty: Mark missing information as [FIELD VERIFICATION REQUIRED].

That is far stronger than:

Make me a JSA.

The Best Workflow Is Iterative

Do not expect one prompt to produce the final product.

Try:

Draft → Review → Challenge → Verify → Revise

For example:

Draft

Ask ChatGPT to create the first version.

Review

Read it using your professional knowledge.

Challenge

Ask ChatGPT to identify weaknesses, assumptions and missing information.

Verify

Check important claims against authoritative sources and actual field conditions.

Revise

Create the final human-approved product.

AI works particularly well inside that cycle.

Ten Practical Ways to Start

If you are new to ChatGPT, try using it this week to:

1. Draft a toolbox talk from a verified procedure.
2. Turn a technical requirement into plain language.
3. Critique an existing JSA.
4. Prepare neutral incident-interview questions.
5. Organize investigation facts into a timeline.
6. Challenge a proposed root-cause conclusion.
7. Review a procedure for ambiguity.
8. Generate Learning Team questions.
9. Summarize safety data and identify questions raised by the patterns.
10. Turn a technical safety issue into a concise leadership briefing.

Start with work you already understand.

That makes it easier to recognize when the AI gets something wrong.

What Should Safety Professionals Not Do With ChatGPT?

Do not use it to avoid going to the field.

Do not use it to replace worker involvement.

Do not assume citations are real.

Do not assume regulatory interpretations are correct.

Do not let polished writing hide weak evidence.

Do not provide restricted information without authorization.

Do not let AI make consequential personnel decisions for you.

Do not use it as a substitute for engineering expertise, legal advice, medical expertise or another qualified professional when those are required.

And do not allow AI to become an excuse for stopping your own thinking.

AI Should Give Safety Professionals More Time to Be Safety Professionals

This may be the biggest opportunity.

How much safety-professional time is spent:

• formatting documents,
• rewriting paragraphs,
• summarizing meetings,
• reorganizing information,
• creating first drafts,
• converting technical language,
• building repetitive communications, and
• preparing routine materials?

Those activities are necessary.

But they can consume time that might otherwise be spent:

• observing work,
• talking with employees,
• coaching supervisors,
• evaluating controls,
• learning from operations, and
• solving difficult problems.

If AI reduces administrative friction, the goal should not simply be:

Produce more paperwork.

The better goal is:

Create more capacity for meaningful safety work.

The Safety Professional Remains Accountable

ChatGPT can generate words.

It cannot stand beside a worker and experience the task.

It cannot assume your professional responsibilities.

It does not know every condition in your operation unless that information is provided and correctly understood.

The safety professional still needs to:

Go see the work.

Talk with people.

Verify conditions.

Check sources.

Exercise judgment.

Challenge assumptions.

Understand controls.

Make decisions.

AI can support that process.

It should not replace it.

The Question Is No Longer Whether AI Can Help

It can.

The more useful question is:

Where can AI improve the quality or efficiency of safety work without weakening human judgment, field learning or professional accountability?

That is where safety professionals should focus.

Use AI for what it does well.

Verify what matters.

Protect information.

Keep humans involved.

And use the time AI gives back to understand the work more deeply.

That is how ChatGPT becomes more than a novelty.

It becomes a tool.

Go Deeper: AI: ChatGPT a New Tool for Smarter Safety

AI: ChatGPT a New Tool for Smarter Safety explores practical ways safety professionals can use generative AI while maintaining professional judgment and responsibility.

The focus is not replacing the safety professional.

It is making the safety professional more capable.

Available in paperback and Kindle on Amazon.