Jump to
Why fast AI RAMS can mean less thinking about the job
Who actually checks an AI-generated RAMS?
Complacency and generic RAMS: why workers stop reading
Why workers should help write the risk assessment
The toolbox talk: good communication or failure to engage?
Is AI de-skilling the people who write RAMS?
Who is responsible if an AI risk assessment gets it wrong?
Cheap, fast RAMS: what are you actually paying for?
AI-generated RAMS and the normalisation of poor practice
Conclusion: We need to get back to the purpose of safety documentation
More tradespeople are using AI to write their RAMS. It’s quick, but a RAMS document isn’t suitable and sufficient until a competent person has checked it against the actual job.
The bigger risk isn’t the time it takes to review. It’s that AI-generated RAMS make it easy to skip the thinking altogether. The document looks right, everyone signs it, and nobody has stopped to ask whether the work is actually safe.
This article looks at how that happens, who’s responsible when it goes wrong, and how to use AI for RAMS without losing the judgement that keeps people safe.
Why fast AI RAMS can mean less thinking about the job
There is an uncomfortable truth about the production of RAMS in some parts of the construction industry.
For many people, it is already regarded as a necessary administrative inconvenience. Something that has to be produced because the principal contractor, client or site manager demands it. Something to be signed, submitted and filed before work can start.
The document itself is often seen as the objective, rather than the process of thinking that should sit behind it.
Now introduce AI.
A contractor who previously had to spend time considering the work, identifying hazards and putting together a method statement can now obtain something that looks credible in a matter of minutes.
The temptation is obvious. Why spend time thinking about the activity when software can produce the paperwork?
But this creates a potentially dangerous behavioural shift. The effort involved in producing the document is removed, and with it may go the opportunity to think critically about the work.
The problem isn't that tradespeople are inherently careless or incapable of understanding risk. Far from it. Many have extraordinary practical knowledge and an instinctive understanding of the dangers associated with their particular trade.
The problem is that practical experience can coexist with a dismissive attitude towards formal safety processes.
And when the administrative requirement becomes easier to satisfy without genuine engagement, the distinction between compliance and competence becomes even more blurred.
Who actually checks an AI-generated RAMS?
One of the most concerning consequences of automating RAMS documents is that there's a big worry that everyone in the supply chain might start getting too relaxed and not pay enough attention.
Consider the chain of events:
- A contractor provides a brief description of the proposed work.
- An AI service generates a risk assessment and method statement.
- The contractor submits the document to the principal contractor.
- Someone checks that the required documentation has been received.
- The document is approved or accepted.
- The workforce signs an attendance sheet at the briefing.
On paper, everything appears to have happened correctly.
- But who has actually thought about the work?
- Who has challenged the proposed methodology?
- Who has considered whether the identified controls are practical?
- Who has checked that the people undertaking the work understand the hazards?
There is a real danger that every participant in the process assumes that somebody else has undertaken the necessary thinking.
The contractor assumes the AI has produced a suitable assessment. The principal contractor assumes the contractor has checked it. The supervisor assumes the workforce has understood it. The workforce assumes the document has already been checked by competent people.
And so the document travels through the system, accumulating signatures and approvals without necessarily accumulating any meaningful understanding.
This is not risk management. It is the administration of reassurance.
Complacency and generic RAMS: why workers stop reading
We need to be honest here, while being fair to the workforce.
Some people on site have little interest in the details of a risk assessment. They want to get on with the job. They’ve done similar work hundreds of times, they know their trade, and they don’t like being told how to do something they already understand.
“I’ve been doing this for 30 years, I don’t need somebody telling me how to do my job” is a familiar attitude.
Sometimes that confidence is backed up by experience. Sometimes it’s exactly the kind of overconfidence that leads to incidents.
Experience is valuable, but it doesn’t make anyone immune to risk.
In fact, familiarity can be a hazard in its own right. Doing a task again and again without incident slowly wears down how risky it feels. Shortcuts become normal, precautions become optional and procedures become something you follow when somebody is watching.
This is where poor-quality AI-generated RAMS can make an existing problem worse. If the contractor hasn’t properly thought about the job, the document is unlikely to challenge anyone’s assumptions. It becomes a set of generic statements everyone can sign without reading.
The tradesperson sees another RAMS and another signature. The supervisor sees another completed task. The organisation sees more evidence that the paperwork is done.
Do this often enough and people stop reading at all. They expect the same hazards, the same precautions and the same lines about PPE, housekeeping and competence. So when a RAMS does flag something serious, it gets lost in the noise.
Poor safety documentation can condition people to ignore safety documentation.
Why workers should help write the risk assessment
One of the fundamental weaknesses of automated RAMS production is that it can encourage a passive approach to safety.
Good risk assessments should involve the people who understand and undertake the work.
The tradesperson often knows things about the activity that cannot be adequately captured in a brief description supplied to an AI platform. They understand the practical difficulties, the awkward access arrangements, the limitations of equipment and the shortcuts that people are likely to take when faced with a particular task.
That knowledge is invaluable.
Yet the automated production model can bypass that contribution entirely.
Rather than involving the workforce in identifying hazards and considering controls, the organisation produces a finished document and presents it to them for acceptance.
This reverses the process.
Instead of the risk assessment being informed by the people undertaking the work, the people undertaking the work are expected to conform to an assessment produced without their meaningful involvement.
The result is a workforce that becomes a recipient of safety instructions rather than an active participant in safety management.
And that matters because people are much more likely to recognise, understand and take ownership of precautions when they have been involved in considering why those precautions are necessary.
The toolbox talk: good communication or failure to engage?
This deserves its own section because it is arguably the point at which the entire process succeeds or fails.
A RAMS document can be beautifully written and technically comprehensive, but unless its contents are communicated effectively, it has achieved very little.
We have all seen the ritual.
A group of workers gathers around a supervisor. A document is produced. Somebody reads through it, often with little enthusiasm. Names are entered on an attendance sheet. Everybody returns to work.
The briefing is recorded as completed.
But what has actually been achieved?
Did anyone ask questions? Did anybody challenge the proposed method? Did the supervisor explain the controls that are particularly important? Did the workforce demonstrate an understanding of what to do if the circumstances change?
Or did everybody simply sign because they wanted to get on with the job?
AI has the potential to make this problem worse by producing longer, more detailed documents that are even less likely to be read or understood.
There is a temptation to equate volume with thoroughness. A 30-page RAMS document can look impressive, but if the critical information is buried in repetitive paragraphs and generic control measures, it may be less useful than a carefully prepared five-page assessment.
We need to recognise that communication is not the act of transferring information from a document to a person.
It is the process of ensuring that the person understands what is required of them.
That requires human interaction, questioning, explanation and, sometimes, challenge.
No amount of automated text generation can guarantee that understanding has taken place.
Is AI de-skilling the people who write RAMS?
There is a further risk that deserves attention: the gradual de-skilling of the people responsible for producing and reviewing safety documentation.
Writing a good risk assessment is an intellectual exercise. It requires the author to understand the proposed activity, break it down into its constituent stages, identify foreseeable hazards, evaluate the risks and establish suitable controls.
It also requires the author to question assumptions.
Why are we doing it this way? Is there a safer alternative? What happens if the equipment fails? What if another contractor is working nearby? What happens when the job doesn't go according to plan?
These questions are the essence of risk assessment.
If we increasingly outsource this process to AI, there is a danger that the people responsible for safety documentation become editors of generated text rather than assessors of risk.
They may become very proficient at prompting software, adjusting formatting and making documents look professional, whilst becoming progressively less engaged in the actual process of identifying and managing hazards.
This is a particularly worrying prospect for younger or less experienced personnel.
Professional competence develops through questioning, investigation, observation and experience. It cannot be acquired simply by becoming proficient at generating documentation.
We should be developing competent safety professionals, not competent operators of document-generation software.
Who is responsible if an AI risk assessment gets it wrong?
There is a fundamental question that the AI RAMS industry needs to answer.
When an AI-generated risk assessment fails to identify a significant hazard, who is accountable?
The software developer? The company selling the RAMS service? The contractor who supplied the information? The person who submitted the document? The person who approved it?
Of course, legal responsibility depends on the circumstances and the duties of the parties involved. But one principle should be beyond dispute: using AI does not transfer the employer's or contractor's responsibilities to a software provider.
The problem is that automation can create psychological distance between people and their decisions.
If someone personally develops a risk assessment, they are more likely to feel ownership of its contents. They have considered the activity, made decisions and selected the controls.
If someone downloads a completed assessment, makes a few amendments and submits it, that sense of ownership may be considerably weaker.
It becomes somebody else's work, somebody else's thinking and, potentially, somebody else's problem.
This is particularly dangerous in safety management because accountability needs to be accompanied by understanding.
You cannot meaningfully take responsibility for a safe system of work if you have not understood how that system is supposed to operate.
Cheap, fast RAMS: what are you actually paying for?
We should also examine the commercial incentives behind the growth of AI RAMS services.
There is an obvious market for cheap, fast safety documentation. Contractors want to reduce administrative burdens, consultants want to improve productivity and service providers want to sell scalable products.
There is nothing inherently wrong with commercial innovation.
However, we need to distinguish between selling a document and providing a competent safety service.
A service that generates a RAMS document in minutes may be commercially attractive, but what exactly is being sold?
Is it a properly considered assessment of a particular work activity, or is it a collection of industry-standard phrases assembled into a convincing format?
The distinction is important.
A genuinely competent risk assessment requires information gathering, questioning, consideration of the working environment and an understanding of the proposed methodology.
These things take time.
If a business model depends upon producing hundreds of individual assessments with minimal human involvement, we should reasonably question how much genuine assessment is taking place.
The danger is that safety becomes another commodity, measured by the number of documents produced and the speed at which they can be delivered.
And when price and speed become the primary measures of quality, the actual purpose of the exercise can become secondary.
AI-generated RAMS and the normalisation of poor practice
Perhaps the most concerning long-term consequence is the possibility that poor-quality documentation becomes accepted as the industry standard.
If contractors increasingly use automated services to produce generic RAMS, and principal contractors continue accepting those documents because they satisfy administrative requirements, the market will have little incentive to improve.
The feedback loop becomes self-perpetuating.
Poor documents are produced, accepted, signed and filed. Their acceptance is interpreted as evidence of adequacy. More documents are produced in the same way.
Eventually, the distinction between an adequate risk assessment and an acceptable-looking risk assessment becomes difficult to recognise.
We should be particularly concerned about the effect this has on younger workers entering the industry.
What example are we setting when we teach people that risk assessment is something that can be generated automatically, signed off and forgotten?
What happens to the culture of questioning, challenging and thinking critically about work?
We should not be surprised if the next generation views RAMS as little more than a bureaucratic requirement if that is precisely how the industry presents them.
How to use AI for RAMS safely
I am not suggesting that AI should be excluded from health and safety management.
That would be neither realistic nor particularly productive.
There are legitimate opportunities to use AI to improve the quality and efficiency of safety documentation.
For example, it could help competent professionals:
- Identify potentially overlooked hazards.
- Challenge assumptions within an existing risk assessment.
- Review the consistency of proposed control measures.
- Improve the structure and readability of documents.
- Produce accessible workforce briefings.
- Identify gaps between a method statement and its associated risk assessment.
But there is an important difference between using AI to challenge and support professional thinking and using it to replace professional thinking.
The former has considerable potential.
The latter risks creating a generation of safety documents that are technically convincing but practically disconnected from the work they are supposed to control.
AI should be an assistant to competent safety professionals, not an alternative to them.
Conclusion: We need to get back to the purpose of safety documentation
The construction industry does not need more paperwork.
It needs better risk management.
It needs people who understand the work they are undertaking, supervisors who are prepared to challenge unsafe practices, contractors who take responsibility for their activities and safety professionals who understand that their role extends considerably beyond producing documents.
Most importantly, it needs a workforce that understands why safety procedures exist and recognises that following them is not simply an administrative obligation.
A risk assessment should make somebody stop and think.
A method statement should explain how work is to be undertaken safely.
A briefing should establish understanding.
And the controls identified should actually be implemented.
If AI helps us achieve these things, then it has a valuable place in the industry.
If it simply allows us to produce more paperwork, more quickly, with less thought and less involvement from the people undertaking the work, then we need to question whether we are making progress at all.
There is something profoundly wrong with a safety culture in which the production of a document is considered more important than the understanding of its contents.
We should be wary of any technology that makes it easier to satisfy the appearance of safety management whilst potentially removing the very thought processes that make it effective.
A RAMS document is not a safety control in itself. It is a means of communicating the controls that people must understand and implement.
And perhaps that is the most important point of all.
We can automate the production of words. We can automate formatting, cross-referencing and even some elements of hazard identification.
But we cannot afford to automate away responsibility, professional judgement, workforce engagement and the willingness to stop and think.
Because when the paperwork is complete, the signatures are collected and the document is filed, somebody still has to do the job.
And it is that person, standing in the workplace, exposed to the hazards of the activity, who ultimately depends on the quality of the thinking that went into that document.
Why should you use MyRAMSapp?
MyRAMSapp is built to save you time without taking the thinking out of the process. Our templates are written by qualified health and safety professionals with a decade of experience in construction, and they’re designed for specific trades, from electricians and plumbers to roofers and groundworkers.
For more information about each subscription type visit single-trade RAMS or multi-trade RAMS. If you’re a principal contractor reviewing subcontractor RAMS, see our principal contractor tier.