Back to all articles

How to remove personal data from PDFs for an AI assistant

A black box in a PDF can leave readable text underneath. Choose what your assistant needs and check the exported file before uploading it.

Informio··4 min read

Hands hold a sheet with black bars above a blue folder on a wooden table.

A local council wants an assistant to explain what to attach to an application. The team has a completed form to hand. The instructions are useful, but the applicant’s name and address do nothing to help answer the question. Before uploading a file, pick a smaller source and check that the resulting PDF contains no details from the original case.

Choose a source that answers the question

Write down the question the assistant should answer, then mark the paragraphs it needs. Under the GDPR principle of data minimisation, the scope of personal data depends on what the specific purpose requires. Article 5 of the GDPR.

  • If residents ask which documents to attach to an application, use a blank form and the approved list of attachments.
  • If customers ask where to send a warranty claim, prepare instructions with the current contact and procedure. Keep completed claim forms in the case records.

For a public assistant, ask whether you would make the chosen material available to the same visitors directly on the website. Sometimes a short new guide is enough. Your team can check the current conditions and links as they write it.

Check what sits underneath the black box

A rectangle drawn in an editor can cover a name on screen while leaving it in the file. It may still be possible to copy the original text. UK ICO guidance on redacting information.

A scan can contain a text layer as well as an image. Optical character recognition, or OCR, adds that layer to make the document searchable. Adobe documentation on recognising text in scans.

Edit a copy and save a new file

Use a copy and a tool approved by your organisation. Apply the feature that permanently removes the marked text or images, confirm the changes and save the result as a new file. Use the PDF sanitisation feature to remove unnecessary comments, attachments, hidden layers and personal details in the document properties. Adobe documentation on PDF redaction and sanitisation.

Read the whole export and consider details in combination. After removing a name, a case number and an exact address may still identify the person. Choose a neutral filename such as warranty-claim-guide.pdf and retain information the assistant needs, including current contact details.

Open the exported file and check it before uploading

  1. Review every page, including scans, headers and footers. Check that the instructions remain readable and their attachments are current.
  2. Search for some of the removed details and try copying text from an edited area. This catches some mistakes. Assess the file as a whole as well.
  3. Check the export for unnecessary comments, attachments, hidden layers and identifying file properties. Ask a colleague to review material before adding it to a public assistant.

If the export still contains case details, stop the upload and fix the copy. For a complex attachment, it may be easier to create new material from verified instructions alone.

Test the assistant with the prepared material

You can practise on a fictional form. The document anonymiser suggests areas for removal so you can review them before saving. Document Anonymiser.

Add a short, approved source to a test assistant. Ask “What should I attach to the application?” and compare the reply with the instructions. Also ask about a specific applicant. When using fictional details for the test, check that the removed details stay out of the reply and that the assistant can say when its sources do not have the answer. Try an Informio assistant with your own approved material.

What takes up too much time in your business?

Tell us about a process you want to improve. Together, we will choose a suitable use for AI and agree on a first step we can test.

AI chat: 100 free answers, no payment card