Clean Files Before Sharing

A file that leaves your team carries more than its visible text: author and company names, reviewer comments, tracked changes, speaker notes, custom properties, the location where a photo was taken, and Content Credentials or AI labels on its pictures. PSWriteOffice finds these items, writes a cleaned copy and tells you what happened to each one. The original file is never changed, and neither Office nor any other application needs to be installed.

Get-OfficeHiddenData and Remove-OfficeHiddenData work on Word, Excel, PowerPoint, Visio, OpenDocument, EPUB and PDF files, and on JPEG, PNG, WebP, GIF and TIFF images. They are the PowerShell form of officeimo hidden-data inspect and officeimo hidden-data remove, and they use the same engine.

The outputs below come from a folder with four files: a Word proposal with a reviewer comment, a custom property and a picture labelled as AI-generated; an Excel budget with a cell comment; a PowerPoint briefing with speaker notes; and a PDF summary with an attached cost model. The clean-before-sharing recipe builds similar files and runs the whole check.

See what would leave with a file

PS> Get-ChildItem -File | Get-OfficeHiddenData

Name          Format     Findings Changed Output
----          ------     -------- ------- ------
Briefing.pptx PowerPoint 4        False
Budget.xlsx   Excel      3        False
Proposal.docx Word       9        False
Summary.pdf   PDF        2        False

Each result has the file Path, its Format, the categories found in Kinds, and a Findings list:

PS> (Get-OfficeHiddenData -Path .\Proposal.docx).Findings | Select-Object Kind, Description, Value

               Kind Description                                  Value
               ---- -----------                                  -----
           Comments Comment by Jane Reviewer                     Check the price before sending.
 PersonalIdentities Author name on 1 comment or revision marks   Jane Reviewer
 DocumentProperties Author                                       Jane Reviewer
   CustomProperties Custom property 'ClientCode'                 CONTOSO-7731
RevisionIdentifiers 1 revision save identifiers
RevisionIdentifiers 6 revision save identifiers
RevisionIdentifiers 17 revision save identifiers
RevisionIdentifiers 1 revision save identifiers
         Provenance Digital source type: TrainedAlgorithmicMedia http://cv.iptc.org/newscodes/digitalsourcety…

PS> (Get-OfficeHiddenData -Path .\Briefing.pptx).Findings | Select-Object Kind, Description, Value, Location

              Kind Description              Value                          Location
              ---- -----------              -----                          --------
      SpeakerNotes Speaker notes            Do not mention the merger yet. Slide 1
DocumentProperties Author                   Jane Reviewer                  /docProps/core.xml
DocumentProperties Last modified by         OfficeIMO                      /docProps/core.xml
         Thumbnail Document thumbnail image                                /docProps/thumbnail.jpeg

Location says where an item lives: a package part, a slide, a sheet cell or a comment. Value is a short preview, such as the author name or the first words of a comment. A file that Get-OfficeHiddenData can't read, for example a .txt file in the same folder, writes a non-terminating error and the rest of the folder is still checked.

Write a cleaned copy

PS> $report = Remove-OfficeHiddenData -Path .\Proposal.docx
PS> $report

Name          Format Findings Changed Output
----          ------ -------- ------- ------
Proposal.docx Word   8        True    Proposal.cleaned.docx

PS> $report.Findings | Select-Object Outcome, Kind, Description, Reason

 Outcome                Kind Description                                       Reason
 -------                ---- -----------                                       ------
 Removed            Comments Comment by Jane Reviewer                            None
 Removed  DocumentProperties Author                                              None
 Removed    CustomProperties Custom property 'ClientCode'                        None
 Removed RevisionIdentifiers 1 revision save identifiers                         None
 Removed RevisionIdentifiers 6 revision save identifiers                         None
 Removed RevisionIdentifiers 17 revision save identifiers                        None
 Removed RevisionIdentifiers 1 revision save identifiers                         None
Retained          Provenance Digital source type: TrainedAlgorithmicMedia NotSelected

PS> ($report.Findings | Where-Object Outcome -EQ Retained).Note
Kept because this category wasn't chosen for removal.

The copy is written as <name>.cleaned<ext> beside the input. Use -OutputPath to name it, or -OutputDirectory to write every copy into one folder under its original name. Running the command again replaces an earlier cleaned copy, the same as Convert-OfficeDocument and Optimize-OfficePdf. Add -NoClobber to keep an existing copy instead; the command then reports "The output already exists" for that file and moves on. -WhatIf shows which files would be written without reading or writing anything.

Every finding has an Outcome: Removed, Anonymized, Accepted, Rejected or Retained. A retained finding has a Reason code and a Note sentence. Here the AI label on the picture stayed because provenance is not one of the default categories. Inspecting the copy shows what is left:

PS> (Get-OfficeHiddenData -Path .\Proposal.cleaned.docx).Findings | Select-Object Outcome, Kind, Description

Outcome       Kind Description
-------       ---- -----------
  Found Provenance Digital source type: TrainedAlgorithmicMedia

Choose what to remove

With no options, Remove-OfficeHiddenData removes the categories that never change what readers see: document and custom properties, comments, author names, speaker notes, the thumbnail, custom XML, image metadata such as GPS locations, printer settings, external references, Word document variables and revision identifiers, and scripts and other active content. These stay until you select them:

• Tracked changes. Accepting or rejecting them changes the text. Use -TrackedChanges AcceptAll or RejectAll.
• Hidden text, rows, columns, sheets and slides. Readers would see different content, and deleting a hidden sheet can break formulas. Add HiddenContent.
• Attachments and PDF bookmarks. The reader may need them. Add Attachments or Bookmarks.
• Content Credentials and AI labels. They tell the reader how content was made. Add Provenance.

New-OfficeHiddenDataOptions builds the choice. -Kind takes Default, All or category names, and combines them:

PS> $options = New-OfficeHiddenDataOptions -Kind Default, Provenance
PS> $options | Format-List Kinds, TrackedChanges, AnonymousAuthor

Kinds           : Provenance, Default
TrackedChanges  : Keep
AnonymousAuthor : Author

-KeepProvenance keeps some provenance records when Provenance is selected: C2pa keeps Content Credentials, External keeps links to them, and AiSource keeps AI-generated labels. New-OfficeHiddenDataOptions -Kind Default, Provenance -KeepProvenance C2pa, External removes only the AI labels. -AnonymousAuthor sets the name that replaces authors on comments and revisions you keep; the default is Author.

Remove-OfficeHiddenData refuses to save a change that would break an Office package signature. Add -RemoveInvalidatedSignatures to remove those signatures and sign the cleaned copy again afterwards. Signed PDFs are always refused.

Content Credentials and AI labels

Get-OfficeDocumentProvenance reports provenance in more detail than the hidden-data list, including signatures and, with a c2patool you supply, cryptographic verification:

PS> (Get-OfficeDocumentProvenance -Path .\Proposal.docx).Structural | Select-Object Format, HasC2paManifest, HasGenerativeAiDeclaration

    Format HasC2paManifest HasGenerativeAiDeclaration
    ------ --------------- --------------------------
ZipPackage           False                       True

Selecting Provenance removes the label from the picture inside the proposal:

PS> $report = Remove-OfficeHiddenData -Path .\Proposal.docx -Options $options
PS> $report.Findings | Where-Object Kind -EQ Provenance | Select-Object Outcome, Description, Part

Outcome Description                                  Part
------- -----------                                  ----
Removed Digital source type: TrainedAlgorithmicMedia /media/image.png

PS> (Get-OfficeDocumentProvenance -Path .\Proposal.cleaned.docx).Structural.HasGenerativeAiDeclaration
False

Content Credentials that cover the whole file, such as those in a PDF or in an Office package's own credential part, are always removed, because any change to the file makes them invalid. Remove provenance only when you mean to: a label that says a picture was generated is often something the reader should see.

Clean a folder and check the result

PS> Get-ChildItem -File | Where-Object Name -NotLike '*.cleaned.*' |
        Remove-OfficeHiddenData -OutputDirectory ..\Share -Options $options

Name          Format     Findings Changed Output
----          ------     -------- ------- ------
Briefing.pptx PowerPoint 4        True    Briefing.pptx
Budget.xlsx   Excel      2        True    Budget.xlsx
Proposal.docx Word       8        True    Proposal.docx
Summary.pdf   PDF        3        True    Summary.pdf

PS> Get-ChildItem ..\Share | Get-OfficeHiddenData

Name          Format     Findings Changed Output
----          ------     -------- ------- ------
Briefing.pptx PowerPoint 0        False
Budget.xlsx   Excel      0        False
Proposal.docx Word       0        False
Summary.pdf   PDF        1        False

The finding left in Summary.pdf is the attached cost model, because attachments are opt-in. A scheduled job can stop on anything that remains:

$left = Get-ChildItem .\Share -File | Get-OfficeHiddenData | Where-Object { $_.Findings.Count -gt 0 }
if ($left) { throw "Hidden data remains in: $($left.Path -join ', ')" }

PDF files

Remove-OfficeHiddenData cleans PDFs with the same categories as Office files. For a PDF the defaults remove the author-written document information and XMP metadata, comments, scripts and unsafe actions, and keep attachments, bookmarks and layers. The title always stays, as it does for Word, Excel and PowerPoint, and so does PDF/A, PDF/UA, PDF/X and Factur-X identification, so a cleaned file keeps its conformance. Inspection doesn't list what always stays:

PS> (Get-OfficeHiddenData -Path .\Summary.pdf).Findings | Select-Object Kind, Description, Value

              Kind Description                         Value
              ---- -----------                         -----
DocumentProperties Document information Author         Jane Reviewer
       Attachments Attached file (Names/EmbeddedFiles) cost-model.txt

The removal report does, with the reason:

PS> (Remove-OfficeHiddenData -Path .\Summary.pdf).Findings | Select-Object Outcome, Kind, Description

 Outcome               Kind Description
 -------               ---- -----------
Retained DocumentProperties Document information Title
 Removed DocumentProperties Document information Author
Retained        Attachments Attached file (Names/EmbeddedFiles)

Optimize-OfficePdf -Sanitize is the PDF route for removing what can run or travel with the file. Its default is attachments and active content, and it leaves the document information alone:

PS> $result = Optimize-OfficePdf -Path .\Summary.pdf -OutputPath .\Summary-sanitized.pdf -Sanitize -PassThru
PS> $result.Report.Findings | Select-Object Outcome, Kind, Description

 Outcome               Kind Description
 -------               ---- -----------
Retained DocumentProperties Document information Title
Retained DocumentProperties Document information Author
 Removed        Attachments Attached file (Names/EmbeddedFiles)

Both accept the options from New-OfficeHiddenDataOptions; Optimize-OfficePdf takes them through -HiddenDataOptions. Use Redact, sanitize, and optimize when visible text has to go too: removing hidden data doesn't redact page content.

Related

• Prepare a document for sharing: the same checks from .NET, the browser and the command line, with accessibility.
• Remove hidden data and metadata: every category and what each format reports.
• Content Credentials and AI labels
• Before sharing from the CLI: officeimo hidden-data and provenance checks in CI.
• Command reference: Get-OfficeHiddenData, Remove-OfficeHiddenData, New-OfficeHiddenDataOptions and Get-OfficeDocumentProvenance.