AI adoption
Turn a useful AI prompt into a working standard.
Give a successful prompt clear inputs, checks, ownership and a change route before the wider team relies on it.
Consider a hypothetical operations manager who has written an AI prompt that turns rough job notes into a clear customer update. It works well for them. They share it with the team, but the results soon vary: one colleague omits the latest status, another pastes information that should not be used, and nobody knows which version is current. The prompt was useful. It was not yet a working standard.
Describe the job before the wording
Start with the task the prompt supports. Name the person doing it, the point in the workflow, the source information available and the decision or output expected. This stops the team treating a block of wording as a complete process.
Define where the prompt must not be used. A draft based on confirmed service notes is different from a response that interprets a contract, makes a refund decision or commits to a delivery date. Boundaries should reflect the consequence of getting the task wrong, not how fluent the output appears.
Make the inputs repeatable
List the minimum information needed for a useful result and where each item comes from. If the source is incomplete, the instruction should ask for the gap to be identified rather than quietly filled with an assumption.
Use a simple input structure when consistency matters. For example, a customer-update prompt might require the confirmed job status, outstanding action, named owner and next review point. Test it with synthetic examples that include a missing field, conflicting notes and an unusual but realistic exception.
- What information is required?
- Which source is authoritative?
- What should happen when evidence is missing?
- Which information must not be entered into the tool?
Design the output for the next step
Describe where the output goes and what the next person needs to do with it. A customer update might need a subject line, the confirmed status, the outstanding action and the next review point in a form that can be transferred into the approved system. A polished answer in the chat window is not enough if someone must reshape it before the work can continue.
Decide which parts must remain consistent and where judgement is expected. Ask a colleague who did not create the prompt to follow the written standard. Note every point where they need a verbal explanation, then improve the instructions or input structure until the handover works without relying on the original author.
Give the standard an owner
Store the approved prompt with a short description of its purpose, owner, last review date and current version. Provide one place for proposed changes. A prompt copied into personal notes or several chat threads becomes difficult to improve consistently.
Review it when the underlying process, source system, policy or AI tool changes. The useful asset is not the original prompt; it is the maintained way of working around it.
Rethink’s AI working sessions can help turn one promising use case into a safe, testable team workflow. Longer-term adoption support can then help maintain the standard, learn from corrections and extend what works without losing control.