Copilot: Block Bad Documents Before Send

Key insights

• Copilot in SharePoint Skill checks each Statement of Advice (SOA) for required clauses and writes any omission to a tracking list called SOA Compliance Gaps.
  When a gap is found the document status flips to Review Before Sending, preventing it from being sent until fixed.
• The solution uses two cooperating Skills: one runs the initial verification and logs gaps; the other re-runs the check after fixes and marks the file SOA Ready to Send.
  This closing loop makes sure nothing leaves unchecked.
• Core architecture relies on two files: SHAREPOINT.md (the site brain) and SKILL.md (the Skill definition).
  Every user on the site inherits both files so the rules and terminology apply consistently.
• Microsoft’s governance feature lets admins apply a Copilot Off sensitivity label via Microsoft Purview to block Copilot from referencing selected sites, libraries, or files.
  Admins enable the behavior with a PowerShell setting called block content analysis services.
• To implement: create a sensitivity label in Purview, enable the block using PowerShell, and apply the label to the site/library/file you want excluded.
  End users can also use scope constraints (type “/” in Copilot chat and lock to a file) to force Copilot to use a single approved source.
• Key benefits: the approach improves governance and accuracy, cuts hallucinations by excluding draft or archived content, supports DLP use cases, and gives site owners fine-grained control without turning Copilot off globally.

Overview of the video

In a recent YouTube walkthrough, Daniel Anderson [MVP] demonstrates a practical solution that prevents incomplete financial statements from being sent out by accident. He shows how two Copilot in SharePoint Skills work together to check Statements of Advice, flag missing sections, and enforce a verification gate before documents leave the system. Consequently, this video targets organizations that need simple, no-code controls to reduce compliance risk while keeping processes fast and transparent.

Anderson narrates a step-by-step build that requires no Power Automate and no custom code, which makes the approach accessible to many SharePoint site owners. He also tests the solution across 13 real SOAs to show how it behaves in realistic conditions. As a result, viewers can see how the design responds to both gaps and fixes in the documents.

How the Copilot Skills enforce SOA checks

The core is a pair of cooperating Skills. First, a verification Skill scans new or updated documents for specific sections, such as the disclosure statement and authority to proceed, and then records gaps in a tracking list called SOA Compliance Gaps. When a gap exists, the Skill flips the document status to Review Before Sending, which prevents the document from being dispatched unchecked.

Second, a follow-up Skill reruns the verification once authors fix the missing content and then marks the document SOA Ready to Send when it passes. This closed-loop design ensures nothing leaves unchecked and reduces human error by automating both the detection and the clearance steps. In effect, the Skills form an automated gate that fits into existing SharePoint workstreams.

Site brain and Skill architecture

Anderson explains that the whole setup relies on two simple files that distribute behavior across the site. The first file, named SHAREPOINT.md, acts as the site brain and loads into every Copilot chat to define terminology and rules that the agent must not break. The second file, SKILL.md, lives in the Agent Assets library and contains the Skill logic; all users on the site inherit both files automatically.

This architecture offers a clear separation between policy and action, which helps administrators maintain consistent behavior across sites. Moreover, because configuration lives in plain files, teams can update rules without changing code, lowering maintenance barriers while still providing governance. However, administrators must keep those files accurate, since errors in the central files propagate quickly.

Governance with Copilot exclusions and Purview

The video also arrives as Microsoft expands governance options by letting organizations block Copilot from referencing certain documents. Specifically, site or library owners can apply a sensitivity label such as Copilot Off that prevents AI content analysis on labeled items. This uses Microsoft Purview in combination with a PowerShell switch to enable the block content analysis services capability.

Consequently, teams can exclude drafts, archived files, or highly sensitive documents from AI processing without disabling Copilot entirely. This selective exclusion addresses hallucination risks that arise when AI mixes approved content with redundant or obsolete material, while still allowing Copilot to assist where appropriate. At the same time, the approach requires careful labeling to avoid unintentionally hiding critical sources from the AI.

Tradeoffs, challenges, and practical guidance

While the combined Skills and the Copilot Off label improve accuracy and governance, they introduce tradeoffs that organizations must weigh. For example, labeling and maintaining the central SHAREPOINT.md and SKILL.md files add operational work, and overly broad labels can block useful material and reduce Copilot’s helpfulness. Therefore, teams must balance strict controls with the need for timely, accurate AI-assisted drafting.

Implementation also raises practical challenges, such as training site owners to apply sensitivity labels correctly and ensuring PowerShell configuration is consistent across tenants. Furthermore, the system can produce false positives if the verification prompts are too strict, or false negatives if prompt wording misses subtle compliance language. Thus, organizations should pilot the setup on a small set of documents, refine the prompts, and monitor the tracking list closely before rolling it out widely.

Overall, Daniel Anderson’s walkthrough offers a pragmatic path to reduce compliance risk using existing Microsoft tools and Copilot capabilities. By automating checks, leveraging site-level configuration files, and using selective exclusion through Purview, teams can significantly lower the chance that incomplete or outdated documents are used in advice. Nevertheless, success depends on clear policies, disciplined labeling, and ongoing review to keep the system aligned with business needs.

Keywords

copilot skill to stop bad documents, copilot document review, prevent sending bad documents, ai proofreading for documents, microsoft copilot for document quality, document compliance checks with copilot, copilot writing assistant for accuracy, catch errors before sending documents