What should you check before issuing AI-generated RAMS?
Introduction
AI can draft RAMS quickly, but it can also produce something that looks credible while embedding the wrong assumptions. The risk is not “AI wrote it”, the risk is issuing a document that has not been validated against the actual job, constraints, interfaces and control measures you can genuinely implement on site.
This article is the pre-issue gate. It is written for planners, supervisors, HSQE, and anyone who signs off RAMS before they are issued, briefed, and used to start work. If you want the wider context on AI-generated RAMS governance, start with the Guide to AI-Generated Risk Assessments and Method Statements (RAMS)
Quick Answer..
Before issuing AI-generated RAMS, check that the inputs are locked and job-specific, the method steps match the real sequence on site, hazards and controls are tied to each step, permits and hold points are included, and a competent person has reviewed and approved a controlled revision. Finally, ensure briefing evidence is tied to the approved revision and that change triggers, re-approval rules, and re-brief requirements are defined before work starts.
The minimum standard before you issue AI-generated RAMS
“Issue-ready” means you can show:
-
The input pack is correct and job-specific (not generic assumptions)
-
The method statement sequence is realistic and implementable
-
Controls are applied where the risk occurs
-
Permits, hold points, and interfaces are not missing
-
A competent review has happened and an approval is recorded
-
The approved revision is the source of truth
-
The briefing is recorded against that revision
-
Change triggers are defined and controlled through revision, approval, and re-brief
If any of the above is missing, you have a draft, not an issued RAMS.
Check 1: Inputs are locked and job-specific
AI outputs reflect the inputs you give it. If your inputs are vague, your RAMS will be vague, or worse, confidently wrong.
Pre-issue input checks:
-
Scope and boundaries: What is in scope, what is out of scope, and where the work starts and ends
-
Location and site constraints: Access, egress, space restrictions, existing services, adjacent live areas, public interface
-
Interfaces: Other contractors, client operations, deliveries, shared plant routes, temporary works designers, traffic management provider
-
Plant and equipment: Actual plant to be used, lifting accessories, attachments, inspection status, operator competence
-
Materials and substances: Concrete additives, resins, fuels, COSHH implications where relevant
-
Environmental conditions: Ground conditions, contamination assumptions, weather sensitivity, working hours constraints
-
Competence requirements: Tickets, training, supervision level, specialist roles
-
Permits and authorisations: Permit to dig, hot works, confined space, lifting plan requirements, isolations, work at height controls
Practical gate: If you cannot point to where each critical input came from (drawing, survey, permit requirement, site rule, client constraint), do not issue.
Check 2: The method steps match the real sequence on site
A common failure mode is a method statement that reads logically but does not match the actual planned sequence, or misses set-up steps that create the highest risk.
Pre-issue checks:
-
Does the sequence include set-up and enabling works (exclusion zones, barriers, signage, isolations, permits, service locating, temporary works checks)?
-
Are interfaces made explicit at the step they occur (plant and pedestrian segregation, deliveries, adjacent trades)?
-
Are critical hold points included (permit approval, inspection sign-off, test results, client witness points)?
-
Is the plan implementable with the space, access, and resources actually available?
Rule: If a control cannot be implemented, it is not a control. Replace it with something you can actually deliver, evidence, and supervise.
Using ChatGPT for RAMS?
Draft Faster. Review Site Risks Properly.
Get access to our free RAMS generator now.
Check 3: Hazards and controls are tied to each step
Do not accept RAMS where hazards and controls live in a generic list detached from the method steps. That format hides gaps and creates briefing failures.
Pre-issue checks:
-
Each step has its specific hazards (not just “manual handling” everywhere)
-
Controls are step-specific and describe how the team will apply them
-
Responsibilities are clear (who sets up, who checks, who authorises, who supervises)
-
Residual risk is not “assumed safe” by generic statements
Examples of what “tied to steps” looks like:
-
Step: service locating and excavation start
Controls: permit to dig confirmed, CAT and Genny used and recorded, trial holes, exclusion zone, supervised dig, stop-work trigger if unknown services found -
Step: lifting and placing components
Controls: lift plan present, competent appointed person, slinger and signaller roles, exclusion zone, lift accessories inspected, wind limits, communication method
If the RAMS cannot be briefed step-by-step, it is not issue-ready.
Check 4: Permits, hold points, and interfaces are not missing
AI drafts often under-specify authorisations because they are context dependent. Your job is to make these explicit.
Pre-issue checks:
-
Permit requirements are listed and referenced at the step they apply
-
Hold points are clear, including what “good” looks like and who signs off
-
Interfaces are controlled (traffic management, shared access routes, exclusion zones, client operations)
-
Inspections and test evidence are included (temporary works inspections, pre-use checks, torque checks, test certificates, as relevant)
Failure pattern: “All permits to be obtained” is not a control. State which permits, when, by whom, and what evidence is recorded.
Check 5: Version control and approval evidence are in place
You need to be able to answer two questions instantly:
-
What is the approved revision?
-
What revision did we brief?
Pre-issue checks:
-
A single source-of-truth location exists (not multiple PDF copies in WhatsApp threads)
-
The RAMS has an ID and revision number
-
Reviewer comments are captured (even if brief), and actions are closed out
-
Approver identity and approval date are recorded
-
“Issued” status is explicit, and superseded copies are clearly marked
If you cannot show a controlled chain from draft to approved to issued, you cannot defend the process.
For deeper process detail, see: How to review and approve RAMS.
Check 6: Briefing evidence and change triggers are ready
Issuing is not the end. If the RAMS is briefed incorrectly, or changes are not controlled, the audit trail collapses.
Pre-issue checks:
-
Briefing record template is ready and includes RAMS ID and revision
-
Attendees and roles are captured
-
Supervisor sign-off is captured
-
Understanding checks are planned (not just “Any questions?”)
-
Evidence you intend to keep is defined (photos of set-up, permits, inspections, exclusion zones, as relevant)
Change control checks:
-
Define change triggers explicitly (scope change, method change, plant change, access change, interface change, incident/near miss, client instruction, weather constraints)
-
Define what requires re-approval (material change vs minor wording)
-
Define re-brief rules (do not continue until the revised approved revision has been briefed)
For practical site change control, see: RAMS change control on site
FAQs
Who is accountable for AI-generated RAMS?
The accountable person is the competent reviewer and approver who signs off the issued RAMS. AI may draft, but it cannot accept responsibility for the method, controls, or safe system of work.
What is the biggest failure mode with AI-generated RAMS?
Issuing plausible text that is not job-specific. It often passes a quick read but fails on site because constraints, interfaces, permits, and step-specific controls were assumed rather than validated.
When do you need to re-brief after an AI RAMS update?
Whenever the change affects the method, sequence, controls, plant, interfaces, permits, or any safety critical step. If in doubt, treat it as a re-brief requirement and record it against the revised approved revision.
What evidence should you keep for audits and client review?
Keep evidence that proves control: input pack references, version history, reviewer comments and actions, approval record, issued revision ID, briefing record tied to that revision, change requests, revised approvals, re-brief records, and archive location.
Conclusion
AI can accelerate RAMS drafting, but it cannot replace your pre-issue gate. Lock job-specific inputs, verify the real sequence, tie controls to steps, include permits and hold points, then issue only an approved controlled revision that can be briefed and evidenced. If you want the broader governance model for AI-generated RAMS, return to the hub: AI-generated risk assessments and method statements (RAMS).
Using ChatGPT for RAMS?
Draft Faster. Review Properly.
Get access to our free RAMS generator and speed up your first draft with site-specific review built in.
