← Workflow Recipes
WHS recipe

How do I consult before an AI tool changes the work?

Builds the consultation and risk assessment pack for an AI tool before go-live, weighted to psychosocial hazards, with the questions workers will actually ask and a record of what was decided.

Tested on 21 August 2026 · Platforms: ChatGPT Enterprise, Claude Cowork · Time required: 45 minutes · For: WHS advisers, Safety managers, HSRs, Operations leads

The trigger

A decision has been made to introduce or change an AI tool that allocates, paces, measures or monitors work at [SITE], and someone is scheduling go-live. Introducing it is a workplace change, so consultation and risk assessment come before the tool arrives, not after the first complaint.

The inputs

  • The [TOOL] and the specific [CHANGE] it makes to how work is allocated, paced, measured or monitored
  • The affected [WORKER_ROLE] groups, and who represents them
  • The vendor material, including what it claims the tool measures and how errors are handled
  • The existing risk assessment for that work, and any current controls
  • The go-live [DATE], the pilot scope and who signs off

The steps

  1. Write the change in one paragraph in plain terms. Name what the tool decides, what it recommends, what it records and what stays with a person. A tool nobody can describe cannot be consulted on.
  2. Map the change against psychosocial hazards first, using the AI-assisted psychosocial risk assessment workflow. Job control, job demands, role clarity, support and surveillance are where AI rollouts land.
  3. Draft the answers to the questions workers will ask. An HSR asking how the tool changes the work needs more than a vendor slide, so answer in terms of this work at this site.
  4. Run the prompt below to assemble the consultation pack from your own material.
  5. Consult, and record it. Consultation is a genuine opportunity to influence the decision before it is made, so hold it while go-live can still move.
  6. Close the loop in writing. What was raised, what changed as a result, what did not change and why, and who signs off. In New South Wales, note that the digital work systems duty reaches how work is allocated by the tool, not only the tool itself.

The prompts

Paste only de-identified material into your approved AI tool.

Prompt
You are preparing a work health and safety consultation pack for
a change at [SITE]. Use only the material pasted below. Where
something is not supplied, write "not supplied".

The change: [DESCRIBE_TOOL_AND_WHAT_IT_CHANGES]
Affected roles: [WORKER_ROLE_GROUPS]
Vendor material: [PASTE_VENDOR_CLAIMS_AND_LIMITS]
Existing risk assessment and controls: [PASTE_CURRENT_CONTROLS]

Produce four sections in order:
1. What changes for the worker, in plain terms, one paragraph
   per [WORKER_ROLE] group.
2. Hazards to consult on, psychosocial first, each with the
   supplied evidence and the gap where evidence is missing.
3. The questions workers and HSRs are likely to ask, with the
   answer drawn only from the supplied material, or the words
   "we do not know yet" where it is not there.
4. What is still open before go-live, with an owner for each.

Do not state that a risk is controlled, that consultation has
been met, or that the tool is safe to deploy.

The checks

  • Every hazard line traces to supplied material or is marked as a gap.
  • No answer to a worker question is invented; unknowns say so.
  • The psychosocial hazards are addressed, not only the physical ones.
  • The pack names what stays a human decision and who holds it.
  • Sweep for names, worker identifiers and incident details before the draft leaves your hands.

Governance controls

  • The person with the duty owns the pack and the decision. AI drafts and organises; a competent person verifies each hazard line against the source material, and the accountable officer signs off.
  • Consultation is recorded, not simulated. An AI-generated list of likely questions is preparation for a conversation, never a substitute for holding one, and it must run before go-live while the decision can still change.
  • De-identify before drafting. Use [SITE] and [WORKER_ROLE], never names, and keep worker complaints and incident detail out of the tool.
  • Officers carry a separate due diligence obligation to acquire and keep up to date knowledge of the risks, so file the pack, the questions asked and the recheck date as officer evidence.
  • Set a review date at go-live plus a fixed period. What the tool does to the work is only observable once it is running.

The output

One consultation pack covering what changes, the hazards, the answers to the questions workers will ask and the open items with owners, plus a dated record of the consultation held and what changed because of it. Both are filed with the risk assessment.

Time it takes

The first run takes the full 45 minutes while vendor material and the current risk assessment are located. Once those live in one place, a later change comes in closer to 25.

TheAICommand. Intelligence, At Your Command.

← Back to the recipe library

General information and education only. Not legal, compliance, financial, or professional advice. This recipe describes a way of working with AI tools; the governance controls, including human review, are part of the procedure and are never optional. Confirm obligations against the primary source and your organisation's own policies.