How to Clean Up a SharePoint Document Library with Copilot

Final_final_v3. Do Not Delete. A library full of files, and very little help working out what any of them actually are.

In this walkthrough, I start with a messy SharePoint document library, add the four metadata columns I use, then ask Copilot to help classify the documents, improve their names and record what needs attention.

Next, I create views for the clean-up, review the findings and save the process as a skill to run on another library. The useful part is making the work repeatable. We still check the findings and decide what should be renamed, updated, kept or archived.

This article follows the sequence in my SharePoint Library Clean-up with Copilot video. The six prompts below are transcribed and formatted for copying, with spoken punctuation and obvious transcription errors tidied. Replace the example document types and departments with the values you need before running Prompt 1. The transcript does not include the full skill-creation prompt, so I describe that part of the demonstration separately rather than inventing an exact prompt.

The process at a glance

1. Create the four metadata columns.
2. Scan and classify the documents.
3. Review and confirm natural-language filenames, preserving version information for this clean-up.
4. Create two temporary clean-up review columns without populating them.
5. Create seven clean-up views.
6. Populate the review outcomes and supporting evidence.
7. Check the findings, save the process as a site skill and test it on another library.

This is a demonstration of the process. Check your permissions and the Copilot capabilities available in your environment. Before using it with organisational content, agree the library purpose, the approved values and who can make decisions about the documents.

1. Create the four metadata columns

The first four columns tell us what the document is, where it belongs, who is responsible for it and whether it is ready to use.

Before copying: the video asks you to insert the document types and departments needed for your library. The bracketed lines below are those inputs, not extra choice values to create.

Prompt 1: Create the four columns

Prepare this SharePoint document library for document cleanup and ongoing information management. Create the following metadata columns exactly as specified.

1. Document Type
Column type: Choice.
Choices: [Insert the document types needed for this library, for example Policies and Procedures.]

2. Department
Column type: Choice.
Allow a single selection only.
Choices: [Insert the departments needed for this library, for example People and Culture, Marketing, Operations, Legal and Finance.]

3. Document Owner
Column type: Person or Group.
Allow selection of one person only.

4. Document Status
Column type: Choice.
Allow a single selection only.
Choices: Draft, Current, Superseded, Archived.

Use the column names and choice values exactly as provided above. Do not create any additional columns, choices or folders.

These columns will be used as a standard metadata structure across document libraries to support document cleanup, ownership, classification and life cycle management.

In the demonstration, Copilot creates the four columns and reports that it has verified them. It also reports that no formatting has been applied because none was requested. Check the resulting columns before moving on.

2. Scan and classify the documents

Now ask Copilot to populate the four columns using evidence from the documents. The important instruction is to leave uncertainty visible.

Prompt 2: Scan and classify

Review each document using its content, title, existing metadata and properties. Populate metadata only where it can be determined reliably.

Populate:
• Document Type
• Department
• Document Owner
• Document Status

Rules:
• Do not guess or invent information.
• If uncertain, leave the field blank for human review.
• Preserve existing metadata unless clearly incorrect.
• Do not change document content, file names or system properties.
• Do not delete, move or archive files.
• Only use the predefined metadata choices.

Goal: Automatically classify documents where there is reliable evidence, while leaving ambiguous documents for manual review.

Check the classifications against the actual documents. In the video, I look at the document types and departments Copilot has assigned before moving on. A populated field still needs to be right. The author or last editor is not automatically the accountable Document Owner, and a recent modified date does not establish that a document is Current.

3. Rename unclear files using natural language

I prefer names that tell people what a document is. Where a legal or controlled-document process requires a naming convention, keep those requirements. Otherwise, natural language is usually more helpful than a string of abbreviations and underscores.

There is an important clean-up detail here: in this video I ask Copilot to preserve existing version labels. We need that evidence to compare related files and decide which version is authoritative. Removing it before the review would make that decision harder.

Prompt 3: Rename documents

Review file names and rename any that are unclear, inconsistent or poorly formatted, so they are easy for people to understand.

• Use the existing file name, content, title and context to determine a meaningful name.
• Use natural language, proper spacing and title case where appropriate.
• Remove underscores, excessive hyphens, duplicate spaces, random characters, unnecessary abbreviations and redundant words.
• Keep the original file extension.
• Always preserve version information such as V1, V2, Draft 2, File V1 or Revision 2.
• Where possible, standardise simple version numbers to V1, V2, V3, etc.
• Give related versions consistent base names so they can easily be compared later.
• Do not assume Final means Current.
• Do not add information that is not supported by the document.
• If the intended name is unclear, leave it unchanged.
• Do not delete, move, archive or merge documents.

Goal: Create clear, consistent file names while preserving version information needed for manual clean-up and duplicate review.

In the demonstration, Copilot reviews 20 files and prepares content-based renames. I review the proposals, confirm them and then it renames the files. Check the proposed names before confirming your own run.

Keeping existing version labels at this stage is about preserving review evidence. Once the right working document has been confirmed, version history can manage subsequent changes. That later decision is separate from this naming prompt.

4. Add the two temporary clean-up columns

The four metadata columns describe the documents. These two columns record the clean-up work: what needs attention, why it was flagged and what somebody should check next.

At this stage, create the columns only. We create the views before asking Copilot to populate the review.

Prompt 4: Add the clean-up review columns

Create two columns to support the document cleanup process.

1. Clean-up Review Outcome
Column type: Choice.
Single selection.
No default.
Choices:
• Potential Current Source
• Needs Review and Update
• Potential Duplicate
• Potential Archive Candidate
• Out of Scope
• Needs Owner Decision

2. Clean-up Evidence
Column type: Multiple lines of plain text.
Record a short explanation supporting the review outcome.

Rules:
• Use the column names and choices exactly as listed.
• Do not add additional choices or columns.
• Do not populate the fields at this stage.

Goal: Capture the recommended clean-up action and the evidence behind it for later review.

Check that both columns exist and that the outcome choices match. A possible archive candidate is a suggestion to review, not approval to archive it.

5. Create the seven clean-up views

Views break the review into manageable groups. That matters when a library is too large to finish in one sitting.

Prompt 5: Create the clean-up views

Create SharePoint library views to manage document cleanup using the Clean-up Review Outcome column.

Create these views:

1. Clean-up All Documents
Show all documents.

2. Clean-up Needs Attention
Filter: Needs Review and Update OR Needs Owner Decision.

3. Clean-up Potential Duplicates
Filter: Potential Duplicate.

4. Clean-up Archive Candidates
Filter: Potential Archive Candidate.

5. Clean-up Potential Current Sources
Filter: Potential Current Source.

6. Clean-up Out of Scope
Filter: Out of Scope.

7. Clean-up Not Yet Reviewed
Show documents where Clean-up Review Outcome is blank.

Where available, show:
Name, Document Type, Department, Document Owner, Document Status, Clean-up Review Outcome, Clean-up Evidence and Modified.

Sort by Modified, newest first.
Apply colour formatting to Clean-up Review Outcome.
Keep the views simple and easy to scan.
Do not change metadata values or modify, move, archive or delete documents.

Colours are visual indicators only and do not represent final decisions.

Check the filters, displayed columns, newest-first sorting and colour formatting. The Needs Attention filter includes either of its two outcomes. Views organise what people see; they do not restrict access to the underlying files.

6. Populate the clean-up outcomes and evidence

Now the review columns and views are in place, ask Copilot to assess the documents. Each file gets one primary outcome and a short explanation of the evidence behind it.

Prompt 6: Populate the clean-up review

Review each document in this SharePoint library and populate the Clean-up Review Outcome and Clean-up Evidence columns.

Use the document’s content, file name, metadata, properties and relationship to other documents in the library.

Clean-up Review Outcome — select one:

Potential Current Source
Use when evidence suggests this may be the current, active or most relevant document.

Needs Review and Update
Use when the document appears relevant but may be outdated, incomplete or require updating.

Potential Duplicate
Use when another document appears to contain the same or substantially similar content, including different versions of the same document.

Potential Archive Candidate
Use when evidence suggests the document is outdated, superseded, historical or no longer actively required.

Out of Scope
Use when the document is not relevant to the purpose of this library or clean-up exercise.

Needs Owner Decision
Use when a clear clean-up decision cannot be made without business or document owner input.

Clean-up Evidence
Add a short, plain-English explanation of the evidence supporting the outcome. Include relevant indicators such as version information, similar documents, document dates, status, content or other observable evidence.

Rules:
• Do not guess.
• If evidence is insufficient, use Needs Owner Decision.
• Do not delete, archive, move, merge or modify document content.
• Do not assume Final means Current.
• Do not automatically decide which duplicate or version should be deleted.
• Treat the Clean-up Review Outcome as a recommendation for human review, not a final records management decision.

7. Review the results and decide what happens next

In the video, Copilot asks me to update the items after presenting its findings. I review the information, select Update and then work through the views.

One example needs an owner’s decision because the meeting notes refer to final files due Friday, but do not record a year, approval status or a link to the resulting files. The useful part is having the reason beside the outcome.

I look at All Documents, Archive Candidates, Needs Attention and Potential Current Sources. The Potential Duplicates view is empty in that demonstration library. That is the result of that particular run, not a reason to assume another library has no duplicates.

Check the underlying content, related documents and version information yourself. Confirm ownership, business requirements and any retention obligations before acting on a finding. A potential current source is not automatically approved content, and an archive candidate is not an instruction to archive.

8. Save the process as a skill

Once I have checked the work, I ask Copilot to create a skill from the process. It asks me to confirm the name and scope, then whether it should be a site skill or a personal skill. In this demonstration I choose a site skill, named Replicate Document Cleanup Library.

The full skill-creation prompt is not spoken in the supplied transcript. The sequence shown is: ask Copilot to turn the process into a skill, review its proposed name and scope, choose Site skill and confirm the save. I have not added an invented “exact prompt” here.

A site skill is scoped to its associated site and remains subject to SharePoint permissions. Choosing Site skill does not make it available to every person or every site in the organisation.

9. Run the skill on another library

I then open another messy library with unclear filenames and no metadata, and ask Copilot to run the Replicate Document Cleanup Library skill on the Marketing Hub library.

The demonstration brings the earlier steps together: the four main columns, two clean-up columns, views, formatting and document assessment. Copilot presents proposed metadata and review outcomes for 20 files, then asks for confirmation before applying the changes.

I check the proposals, confirm them and select Update when prompted. The metadata, Clean-up Review Outcome and Clean-up Evidence are then populated through the reusable process.

The confirmation steps matter. A saved skill makes the review easier to repeat; we still inspect its proposals and check what it actually changed. The scope, supported actions and confirmation experience depend on the environment and the skill being run.

Before you scale it, design it

This was a demonstration of the process. If I were building it for an organisation, I would agree the document types, departments, status definitions and ownership rules first.

I would then create reusable columns and content types, using managed metadata linked to the Term Store where centrally managed values are needed. For Document Owner, I would use a Person column for a named individual, or managed metadata where we are recording an owning team or role.

When creating new libraries, I would build that approved structure in from the beginning. Copilot can help create the library, but I would check that it is using the approved columns and term sets rather than creating another local variation.

Then I would apply the clean-up process. Keep the usual site and library design steps in place. Copilot helps us carry out the work; we still decide how the information should be organised.

Start with one library

The result we want is a library people can understand, with visible review findings and clear decisions about what happens next.

There is no AI without IA. Good information design gives the process something useful to work with.

Not sure where to begin? Use the FREE SharePoint Ready Scorecard to identify what needs attention first.

Liza Tinker · SharePoint, simply sorted.

Watch: Clean up your SharePoint document library with Copilot

Follow the complete walkthrough, from adding the four metadata columns to reviewing your documents, creating clean-up views and saving the process as a reusable skill.

Watch the SharePoint library clean-up walkthrough on YouTube.

Microsoft references

• Create Autofill columns
• Create views in a SharePoint document library
• Extend Copilot in SharePoint with skills