As of 13 August 2026, AI can only partly create a health and safety incident procedure.
This still needs a person who signs their name to it.
Can you do it?
15 minutesto a draft.
1 hourto something you’d act on.
Cost, all in£0
Skill neededchat-fluent
Who has to check ita colleague
What the alternative costsNo price for a human alternative is supplied in the available data.
If this goes wrong: an incident is missed, evidence is lost or an unsafe control is relied on, leaving your organisation exposed to further harm and enforcement.
What to actually do
Hand it to a person
The route this page recommends
A person who owns the outcome does this end to end, worth it when the failure is dear.
Use a tool built for this
Second choiceDo it yourself
The distant thirdA chat interface, chat-fluent skill, and roughly 1 hour until you can act on the result.
How to actually do it
- Open a blank document and gather your organisation name, work locations, activities, working hours, emergency contacts, first-aid arrangements, worker groups, contractors and known hazards.
- Gather your existing risk assessments, accident book or incident forms, emergency plan, insurance reporting instructions and any current health and safety policy.
- Paste those facts and documents into a chatbot with the copyable prompt, replacing each bracketed slot and asking it to mark anything absent as [TO CONFIRM].
- Export the draft and compare its names, phone numbers, locations, first-aid arrangements, reporting routes and escalation triggers against your current workplace documents.
- Ask your health and safety lead or other competent person to check the legal source list, external reporting checks, investigation method, records section and controls against the actual risks at each site.
- Run a short tabletop exercise for a serious injury, a near miss and suspected immediate danger, record every step that fails or is unclear, then amend the procedure.
- Send the tested version to the responsible manager for approval, add an owner and review date, brief affected workers and store the approved version where they can access it.
Prompt
Create a UK workplace health and safety incident procedure for [organisation name], covering [site or work locations], [main activities], [number and type of workers], [contractors or visitors], and [known hazards]. Use only the facts I provide and clearly label every missing detail as [TO CONFIRM]. Do not invent legal duties, reporting deadlines, contact details, emergency arrangements or risk controls. Where a legal or regulatory point needs checking, identify the point and name the current UK source to consult rather than presenting an unverified claim. Write the procedure in plain British English with these sections: purpose and scope; definitions of incident, accident, near miss and dangerous occurrence; immediate response; emergency escalation; first aid and medical attention; making the area safe; preserving evidence; internal reporting; external reporting checks; investigation roles and method; corrective actions; communication and confidentiality; records and retention; worker training; review and version control; and a simple incident report form. Include a responsibility table for workers, supervisors, managers and the health and safety lead. Include separate routes for serious injury, suspected immediate danger, minor injury, near miss and property damage. Do not diagnose injuries. State that urgent or worrying symptoms should be directed to NHS 111, or emergency services where appropriate. Finish with a list of questions that a competent UK health and safety person must answer before approval.
Open it prefilled in ChatGPT or Claude, or copy it into Gemini, which takes no prefill link.
What it gets wrong
- AI cannot inspect your premises or recognise hazards that were left out of your description.
- AI cannot decide whether your controls are suitable for the actual severity and likelihood of a workplace risk.
- AI cannot take responsibility for reporting, investigation decisions or the safety of people following the procedure.
- AI cannot replace a competent person checking UK duties, site arrangements and the results of a practical exercise.
- AI cannot make workers follow the procedure or ensure that corrective actions are completed.
What caps this at PARTLY: legal accountability, context depth and verification cost.
How we scored this
Five axes, each scored nought to two by hand: ten means AI carries the task cleanly, and the thresholds that turn a total into YES, PARTLY or NO are published in the methodology. Each axis name links to its definition.
| Axis | Score (0–2) |
|---|---|
| Output | 2 |
| Inputs | 1 |
| Verification | 1 |
| Liability | 1 |
| Effort delta | 1 |
| Total | 6 / 10 |
The methodology and its thresholds are published in full.
FAQ
- Can ChatGPT write a health and safety incident procedure?
- Yes, it can produce a useful first draft covering response, reporting, investigation, corrective actions and records. It cannot inspect your workplace or approve the procedure, so a competent person must adapt and check it before use.
- What should a workplace incident procedure include?
- It should explain immediate action, emergency escalation, first aid, preserving evidence, internal and external reporting checks, investigation roles, corrective actions, records and training. It should also give workers clear routes for accidents, near misses, dangerous situations and property damage.
- Is an AI health and safety procedure legally compliant?
- You cannot assume that it is. UK legal duties depend on your activities, premises, risks and arrangements, so a competent person must check the procedure against current official guidance and your organisation's responsibilities.
- What should I do if someone is injured at work?
- Follow your emergency arrangements, provide appropriate first aid and get emergency help when needed. Do not use an AI tool to diagnose an injury: direct urgent or worrying symptoms to NHS 111, and use emergency services for an immediate emergency.
Nearby answers
Assessed by gpt-5.6-luna (gpt-5.6-luna) on 2026-08-13, second-checked by an independent model. Wrong somewhere? Email [email protected] and it gets re-checked.
The newsletter
AI news, new answers and product picks, straight to your inbox.