Can an Internal AI Assistant Tell the Difference Between a Company Rule, a Recommendation, and an Example?

Yes, but only when the source material clearly labels each type of information. An internal AI assistant can distinguish a mandatory rule from optional guidance or an illustrative example when the organization records that authority explicitly.

Why the Difference Matters

Business documents frequently mix mandatory requirements, recommended best practices, examples, temporary workarounds, warnings and personal notes on the same page.

Consider: “Employees should submit purchase requests before ordering equipment. For larger purchases, it is usually helpful to obtain two quotes. Last year, the Houston office requested three quotes for a replacement vehicle.”

Submitting the request may be mandatory. Two quotes may be recommended. The Houston example describes one past event. Without labels, the assistant may present all three as requirements.

Define the Knowledge Types

Policy or rule: A mandatory requirement approved by an authorized owner.

Procedure: Required or approved steps for carrying out a task.

Recommendation: Preferred guidance that is not compulsory.

Example: A scenario showing how a rule may apply.

Exception: An approved departure under defined conditions.

Historical reference: Past information that may no longer apply.

Required: Purchases above $10,000 require Finance approval.

Recommended: Obtain two competitive quotes when practical.

Example: Operations obtained three quotes before replacing a vehicle in 2025.

Use Direct Language for Mandatory Requirements

Mandatory guidance should use words such as must, required, prohibited, may only and do not proceed until. Recommendations should use should, recommended, consider and when practical.

“Employees should obtain approval” is weaker than “Employees must obtain approval before submitting the order.”

Structure Examples So They Cannot Become Policy

Example only: A customer-service manager may assign first-week shadowing when the role requires live-call observation. This example does not replace the approved onboarding checklist.

Store Authority as Metadata

Useful SharePoint fields include knowledge type, authority level, approval status, owner, audience, effective date, review date, related policy and AI eligibility.

Microsoft’s guidance on creating and customizing content types shows how organizations can define reusable structures for different document classes.

Do Not Rank Authority by Tone or Popularity

AI should not assume a statement is authoritative because it appears frequently, was written by a senior employee or sounds detailed. Frequency does not create authority.

The NIST AI Risk Management Framework emphasizes governance, documentation, accountability and ongoing evaluation.

Test Questions That Expose the Difference

Ask: Is this required or only recommended? Do I have to obtain three quotes? Was that example a one-time decision? Who may approve an exception? Which source establishes the actual rule?

Where Maisy Fits

During the Pixeldust implementation process, policies, procedures, recommendations, examples and exceptions are separated and assigned clear authority.

The Microsoft 365 Knowledge Hub implementation guide describes how governed SharePoint content, metadata, permissions and Copilot Studio retrieval operate together.

An internal AI assistant should not turn every helpful sentence into a commandment. Companies already have managers for that.