Skip to content
SimplTraining

Journal

When an AI Answer Is Not a Work Instruction: A Practical Onboarding Exercise

SimplSolutions Editorial Team · 5 min read · Published

ShareLinkedIn
When an AI Answer Is Not a Work Instruction: A Practical Onboarding Exercise - featured article image

A plausible AI answer can be useful—and still be unsafe to follow. The key question is not whether the wording sounds professional. It is whether the answer establishes the current company rule that applies to this employee, task, location, and situation.

That distinction is an important onboarding skill. Employees need to know when an AI response is a starting point for finding an approved instruction, rather than the instruction itself.

The verification habit to teach

Before acting on an AI answer, ask five questions:

  1. What is the governing source? Find the approved policy, procedure, standard, or official notice behind the answer.
  2. Is it current? Check its effective date, revision status, and whether a newer instruction supersedes it.
  3. Does it apply here? Confirm the employee’s role, business unit, location, process, and type of transaction.
  4. Are there exceptions or approvals? Look for conditional language, thresholds, required sign-off, emergency rules, or recordkeeping requirements.
  5. What remains uncertain? Write down the missing fact or unresolved conflict instead of silently filling the gap.

If the governing source cannot be identified, two sources conflict, or an unresolved exception could change the decision, the employee should pause and escalate to the designated policy owner, manager, compliance contact, or other approved authority.

This is a practical translation of responsible AI risk management—not a claim that NIST prescribes this exact employee checklist. NIST describes its AI Risk Management Framework as a voluntary framework intended to help organizations manage risks associated with AI and incorporate trustworthiness considerations into the design, development, use, and evaluation of AI systems. Its Generative AI Profile addresses risks specific to generative AI. Read the NIST AI Risk Management Framework.

A short onboarding exercise

Give the employee a deliberately incomplete assistant response and three pieces of context:

  • a current company policy;
  • an older procedure that still appears in a shared folder; and
  • an exception that is not stated in either document but must be confirmed with the policy owner.

Ask the employee to complete four fields:

  • Governing source: Which document controls, and why?
  • Evidence: What passage or metadata supports that choice?
  • Uncertainty: What fact, exception, or conflict is unresolved?
  • Decision: Act, seek confirmation, or escalate? Explain the threshold.

The exercise should assess the employee’s reasoning, not whether they can produce a confident-sounding answer.

Worked example: travel reimbursement

Assume the following fictional company materials are provided for training:

  • Current policy: The company’s Travel and Expense Policy, effective July 1, requires manager approval before booking international travel and requires receipts for expenses above the stated documentation threshold.
  • Outdated procedure: A procedure last revised two years earlier says employees may book international travel after notifying their manager. It does not mention pre-approval.
  • Missing exception: The current policy refers to an exception for urgent customer incidents, but the exception criteria and approving authority are documented elsewhere and are not included in the exercise packet.

The assistant replies:

“You can book the international trip after notifying your manager. Keep your receipts and submit the expense report afterward.”

The answer sounds operational. It is also incomplete.

A strong employee response

Governing source: The current Travel and Expense Policy, because it has the later effective date and requires approval rather than notification. The older procedure should not control where it conflicts with the current policy.

Evidence: The policy’s effective date and its requirement for manager approval before international booking. The employee should record the exact policy section in the training response.

Uncertainty: The trip may qualify for the urgent customer-incident exception, but the exception criteria and approving authority have not been established in the exercise materials.

Decision: Do not book yet. Confirm whether the exception applies and who can approve it. If the trip is not covered by the exception, obtain the required approval before booking.

The employee has not “rejected AI.” They have used it as a lead, checked the governing source, noticed a conflict, and escalated the unresolved condition.

When acting is reasonable—and when it is not

An employee may be able to proceed when the applicable source is clear, current, within scope, and complete enough for the decision. The employee should still follow any required approval or recording step.

Escalation is appropriate when:

  • the answer cites no authoritative source;
  • the only available instruction is outdated or conflicts with a current policy;
  • a missing exception could change the outcome;
  • the action affects safety, privacy, legal obligations, money, access, or a customer commitment;
  • the employee lacks authority to interpret the rule; or
  • the system presents several plausible answers without showing which one governs.

Training should make these decisions observable. A useful record captures the source selected, the uncertainty identified, the action chosen, and the reason for escalation or proceeding. Avoid grading employees solely on speed or answer matching; that can reward confident guessing.

What managers should reinforce

Managers can repeat one simple distinction: an AI answer may summarize; an approved work instruction authorizes action. The employee’s responsibility is to connect the answer to the organization’s current rule before acting.

That distinction also helps organizations evaluate AI-enabled training and knowledge experiences responsibly. NIST’s framework is voluntary guidance, not a company policy and not a substitute for the organization’s own approval, documentation, and escalation controls.

The goal is not to make employees distrust every AI response. It is to build the judgment to know when a response is useful, when it is incomplete, and when a human decision owner must be involved.

When an AI Answer Is Not a Work Instruction: A Practical Onboarding Exercise - sharing image
When an AI Answer Is Not a Work Instruction: A Practical Onboarding Exercise - inline comparison
When an AI Answer Is Not a Work Instruction: A Practical Onboarding Exercise - inline explainer

Start with one role and one hard call.

Tell us which conversation keeps going wrong. We'll build the first practice scenario with you.

A small team meeting at the start of a shift
How to Separate AI Answers From Approved Work Instructions | SimplTraining