Skip to content

Business knowledge preparation guide · By Yes AI

Fix SharePoint oversharing before AI search

Published 5 min read

AI search makes it easier to find anything a person already has access to, including files that were shared far too widely years ago. Fixing that before launch is mostly SharePoint administration, and Microsoft provides reports for it. This guide sets out the order of work for a small or mid-sized business.

Quick answer

Before connecting AI search to SharePoint, run Microsoft's data access governance reports to find sites open to all staff, sites with many sharing links and content shared with 'Everyone except external users'. Fix the highest-risk sites first, such as HR, payroll and finance. Hide sites still under review from discovery or leave them out of the index, then test from an ordinary staff account by asking for pay and disciplinary details.

IT administrator and HR manager reviewing site access permissions on a monitor in an Adelaide office
Start the permission review with HR, payroll and finance sites.

Why does AI search expose oversharing?

Microsoft states that Copilot only surfaces organisational data that a user has at least view permission for, and SharePoint search results are security trimmed in the same way. That protects files with correct permissions. It does nothing for files that were shared too widely in the first place, such as a salary spreadsheet in a site open to the whole company. Ordinary search needed the right keyword to find that file. An assistant answering 'what is the pay range for team leaders' can find it from a plain question.

Which reports show where the problems are?

In the SharePoint admin centre, the data access governance reports show where access is broadest. The site permissions report gives a snapshot across all SharePoint and OneDrive sites, including sites with thousands of users, external guests or 'Everyone except external users' permissions. The sharing links report shows the sites where people created the most links open to anyone, to everyone in the organisation or to specific people in the last 28 days. Another report lists the exact sites, folders and files shared through the special Everyone groups. Microsoft suggests running snapshot reports quarterly and activity reports monthly.

These reports sit within SharePoint Advanced Management, so check Microsoft's prerequisites for your licences. Microsoft notes that administrators with Microsoft 365 E5 can use data access governance reporting with limits: activity reports are available, while snapshot reports and remedial actions are not.

Which sites should you fix first?

Rank sites by what it would cost if the wrong person saw them. Human resources, payroll, finance, legal, board and executive sites come first, followed by sites holding customer or patient records. Within each, look for sites shared with everyone, folders with unique permissions that nobody remembers setting up, and old project sites with external guests. Ask each site owner to confirm who should have access, and record the decision. Sites with no identifiable owner need one assigned before anything else, because nobody else can say who should see them.

How do you fix access without breaking work?

Microsoft's remedial options include site access reviews, which ask site owners to review and update permissions, and restricted access control, which limits a site to a specific group. The change history report helps find recent permission changes that caused the exposure. Remove 'Everyone except external users' from sites that do not need it, replace organisation-wide links with links to named groups, and move sensitive files out of general team sites. Tell affected teams before you change access, because a broken link on a busy morning sends people looking for workarounds.

What if the review will take months?

Restricted Content Discovery hides selected SharePoint sites from organisation-wide search and Copilot while their permissions are reviewed. Microsoft describes it as a temporary control: it does not change permissions, people with access can still open the content directly, and it does not apply to OneDrive. Changes take time to spread through the index, and Microsoft says sites with more than 500,000 items can take over a week. It requires a Microsoft Copilot licence. For a custom knowledge search, the equivalent step is to leave those sites out of the indexing scope until they are fixed.

How do you test that it worked?

Create or borrow an account with ordinary staff access and ask the questions a curious employee might ask: pay rates, redundancies, disciplinary matters, bank details and board minutes. Every answer should come from a document that person is meant to see, or say nothing was found. Repeat the test after each batch of fixes. For a custom build, check that permission changes have reached the index. Microsoft notes that in Azure AI Search, source permission changes only appear in results after the next synchronisation, and some inherited SharePoint changes need an explicit refresh.

SharePoint reports and controls for oversharing

Report or controlWhat it doesChanges permissions?Licence note
Site permissions reportSnapshot of the broadest access across SharePoint and OneDrive sitesNoSharePoint Advanced Management
Sharing links reportSites with the most new sharing links in the last 28 daysNoSharePoint Advanced Management, or limited access on E5
'Everyone except external users' reportSites where content was shared with all internal staff in the last 28 daysNoSharePoint Advanced Management, or limited access on E5
Site access reviewAsks site owners to review and update accessYes, by the site ownerSharePoint Advanced Management
Restricted access controlLimits a site to a specific groupYesSharePoint Advanced Management
Restricted Content DiscoveryHides a site from organisation-wide search and CopilotNoNeeds a Microsoft Copilot licence
Site or library search settingStops a site or library appearing in SharePoint search resultsNoSite owner setting

Practical checklist

  • The site permissions report has been run, and every site open to all staff has an owner's decision recorded.
  • HR, payroll, finance, legal and board sites were checked first.
  • Organisation-wide sharing links on sensitive sites have been replaced or removed.
  • Sites still under review are hidden from discovery or left out of the index scope.
  • An ordinary staff account has asked for pay, disciplinary and bank details and received nothing it should not see.
  • For a custom build, the index has been re-synchronised after the permission fixes.
  • The reports are scheduled to run again after launch.

Common questions

Does AI search bypass SharePoint permissions?

Microsoft states that Copilot only shows content a user has at least view access to, and SharePoint search results are security trimmed. The exposure comes from permissions that are already too broad. A well-built custom search should apply the same principle, filtering passages by the asker's access before the model sees them.

We don't have SharePoint Advanced Management. What can we do?

Microsoft 365 E5 administrators can use the activity reports with limits. Organisations without SharePoint Advanced Management must turn on data collection first, and reports become available 24 hours later. Site owners can still review the members and sharing of their own sites, which is slower but covers the highest-risk sites if you start there.

Should we just exclude HR and finance from AI search?

For a first deployment, that is a sensible start. Leaving the most sensitive sites out of scope removes the largest risk while the permission review continues. It does not fix the underlying sharing, so people with access can still find those files through ordinary search, and the review still needs to finish.

How long do the activity reports keep data?

Activity reports cover the last 28 days. For organisations without SharePoint Advanced Management, Microsoft says collected data is stored for 28 days, reports only include data from when collection was enabled, and collection pauses if no reports are generated for three months, after which it must be re-enabled.

Does Restricted Content Discovery stop people opening the files?

No. Microsoft says it affects discoverability and leaves access permissions unchanged. People who already have access can still open the content directly, and the content stays in the search index, so eDiscovery and auto-labelling keep working. Use it to buy time for the permission review on high-risk sites, and remove it once each site is fixed.

Sources and further reading

Continue planning

Plan a useful knowledge-search pilot

Describe the documents your staff use, the questions they need answered and how uncertainty should be reviewed. We can discuss preparation and a bounded pilot. Keep source documents, customer records and confidential information out of this enquiry.