← Back to all postsJune 5, 2026

Give your business shared rules for using AI

A saved prompt can help with a task. Shared rules explain when to use it, what to check and who is responsible for the result.

An incoming folder moves through a human check to a completed-work tray.

A useful prompt can help someone complete a task. It does not explain which information the business permits, who reviews the result or what happens when the output is wrong.

If several people use AI at work, give them a shared set of working rules. Keep those rules close to the tasks, examples and sources they need.

You do not need a new software platform to start. The aim is a process people can follow consistently.

Choose the work you are approving

List the tasks for which the business has agreed to use AI. Describe the expected result and the limits of each task.

For example, a hypothetical cleaning business might approve drafting a service explanation from a public scope sheet, with a person checking it before use. That approval would not cover pricing a new contract or sending unreviewed customer replies.

Make it easy to ask about a new use. People should not have to infer permission from the fact that a tool is available.

Name the information and the reviewer

Identify which sources are current and approved. Someone needs to maintain the service details, policies or instructions the task depends on.

Set information-handling rules for the actual tool and workflow. Use non-sensitive examples for learning, and give staff a person to ask when they are unsure whether material may be entered.

Describe a satisfactory output with an example. Explain which facts, conditions and uncertainties must remain intact. Name the person who checks it and what they do when it fails.

Keep the prompt with the instructions around it

A prompt is easier to reuse when the next person can see its purpose and sources. Store the current version alongside:

  • The task it supports.
  • The approved source material.
  • An example of a useful result.
  • A rejected example and the reason it failed.
  • The review and approval requirements.
  • The owner and last review date.

If someone improves the instructions, record what changed and why. Keep a way to recover the previous working version.

This can be a short shared document. Choose a place people actually use rather than creating another folder nobody can find.

Define the exception route

The instructions should say what happens when information is missing, sources conflict or a tool is unavailable. Provide a manual route so ordinary work can continue.

Keep customer-facing commitments and consequential system actions within the agreed approval process. A draft that sounds confident should not skip review because the office is busy.

Record failures clearly enough to improve the process without copying unnecessary customer information into a general troubleshooting log.

Review whether the arrangement helps

Choose a review point and examine representative work. Include review time, corrections, maintenance and tool costs.

Ask the people using the process what became easier and which instructions they still have to interpret. A rule that repeatedly causes questions may need a clearer example.

Keep what helps and update what does not. A growing prompt collection is useful only if people can find a relevant, maintained way to do the work.

For help turning scattered experiments into a practical process, tell us which tasks people are using AI for today.

Free rank audit