As of 13 August 2026, AI cannot create a workplace risk assessment.
This still needs a person who signs their name to it.
Can you do it?
15 minutesto a draft.
n/ait cannot be self-verified.
Cost, all in£0
Skill neededchat-fluent
Who has to check ita professional
What the alternative costsThe supplied tool data gives no price for a human workplace risk-assessment service.
If this goes wrong, an omitted hazard or inadequate control can contribute to serious harm while the employer remains accountable.
What to actually do
Hand it to a person
The route this page recommends
Someone with a licence or accountable authority has to sign this before it counts.
Use a tool built for this
Second choiceDo it yourself
The distant thirdA chat interface gets you a draft, but you cannot verify it yourself. That is the catch.
How to actually do it
- Open a blank document or spreadsheet and create a notes section for the workplace, activity, people exposed, equipment, substances, vehicles, process steps, existing controls, incidents and emergency arrangements.
- Walk through the workplace with the person who does the job and record what happens at each process step, including access, lifting, work at height, machinery, electricity, substances, vehicles, fire, lone working and members of the public where relevant.
- Paste the completed notes and the copyable prompt into ChatGPT, Claude or Gemini, replacing each bracketed slot with facts and leaving unknown items marked as unknown.
- Copy the generated table into the document or spreadsheet, keeping every [INFORMATION NEEDED] item and the separate gaps and site-check sections.
- Ask the workplace manager and the people who perform the work to compare each listed control with the actual workplace, then record corrections, missing hazards and evidence such as equipment instructions, training records and inspection records.
- Send the draft, site notes and evidence to a competent health and safety person for a completeness and control review, then have the responsible employer approve, communicate and schedule the actions before relying on the assessment.
Prompt
Create a draft workplace risk assessment for the activity described below. Use UK workplace terminology and plain English. Do not invent facts, measurements, regulations, hazards, controls or risk ratings. If information is missing, write [INFORMATION NEEDED] and list the question that must be answered. Treat this as a working draft, not a completed legal document or a substitute for competent safety advice. Activity and location: [describe the work and where it takes place] People exposed: [employees, contractors, visitors, members of the public, young or vulnerable workers] Equipment, substances and vehicles: [list them] Work process: [describe the steps in order] Known hazards and existing controls: [list what is already known] Incidents, near misses or previous concerns: [provide details or write none known] Emergency arrangements: [provide details or write unknown] Return a table with these columns: activity or hazard; who might be harmed and how; existing controls; further controls needed, using the hierarchy of controls where relevant; person responsible; target date; likelihood; severity; risk rating; residual risk after the further controls; monitoring and review trigger. Explain the risk-rating scale before using it, and mark any rating that depends on missing information. Add a separate section called "Gaps requiring a competent person" and another called "Questions to answer during the site check". Do not state that the assessment is legally compliant.
Open it prefilled in ChatGPT or Claude, or copy it into Gemini, which takes no prefill link.
What it gets wrong
- AI cannot inspect the workplace, notice an obstructed escape route or observe how work is actually done.
- AI cannot know whether your list of hazards is complete when the supplied description leaves out an activity, person or condition.
- AI cannot take responsibility for selecting controls, approving the assessment or making the workplace safe.
- AI cannot replace consultation with workers, practical checks of equipment and confirmation that controls are followed.
- AI cannot give you a reliable risk rating when the likelihood, severity or exposure information is uncertain.
What makes this a NO: legal accountability, stakes of error 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 | 0 |
| Liability | 0 |
| Effort delta | 1 |
| Total | 4 / 10 |
The methodology and its thresholds are published in full.
FAQ
- Can ChatGPT write a risk assessment?
- It can write a structured draft from the information you provide, including hazards, controls, actions and review fields. It cannot inspect the workplace or establish that the assessment is complete, so a competent person must check and approve the result.
- Is it legal to use AI for a risk assessment?
- AI may help with drafting, but using its output does not transfer the employer's responsibility for a suitable and sufficient assessment. This is not professional advice, and a serious or complex case needs a competent health and safety professional.
- What information does AI need to write a workplace risk assessment?
- Give it the work steps, location, people exposed, equipment, substances, vehicles, existing controls, incidents, emergency arrangements and any relevant restrictions. It should mark unknown information rather than fill gaps with assumptions.
- Does an AI-generated risk assessment replace a competent person?
- No. The draft still needs a workplace check, consultation with the people doing the work and review by someone with suitable health and safety competence. The employer remains accountable for the decisions and the controls.
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.