How-to
How to remove client names and details before using AI
Replace every name and identifying detail with a consistent placeholder before the document reaches the AI, for example [PERSON_001] for the same client everywhere. Remove more than names: addresses, emails, phone numbers, dates of birth, account, case and government ID numbers. Then read the whole copy yourself before pasting it, because automated detection and find-and-replace both miss things.
Key takeaway: Consistent placeholders keep the document readable to the AI; a final human read is still necessary.
Last reviewed · 4 sources
Why regulators tell professionals to do this
The Law Society of Alberta’s generative AI playbook says: “Never include confidential or potentially identifying information in prompts.” The Law Society of British Columbia says client confidential information, “including any information identifying the client, would be omitted” from anything supplied to a generative AI tool.
The Law Society of Ontario adds a caution worth keeping in mind: “Anonymizing is not perfect.” Facts that remain can sometimes identify a person, so removing names is the minimum, not the finish line.
What to remove
Identifying details are broader than names. Work through the document for:
- Names of clients, family members, employers, opposing parties and witnesses
- Street addresses, postal codes and small towns
- Email addresses, phone numbers and usernames
- Dates of birth and other exact personal dates
- Government numbers: SIN or SSN, passport, driver’s licence, immigration file numbers
- Bank, credit card, policy, case and account numbers
- Unusual facts that point to one person, such as a rare job title in a small company
Keep the document useful to the AI
Blacking text out or deleting it makes the AI’s job harder: it cannot tell whether two redactions are the same person. Use consistent, typed placeholders instead, such as [PERSON_001] for the client and [PERSON_002] for their employer, and use the same placeholder every time that person appears.
Keep the details the task depends on. A lease review still needs the rent and dates; it rarely needs the tenant’s name.
Check the copy in the format you will actually share
Choose the smallest useful input for the task. A clause review may need one paragraph rather than a complete client file. Preserve the roles and facts needed for the question, then check whether the remaining combination still points to one person.
Review the final artifact, not only the replacement list: text, filename, headers, tables, comments and any included images. A clean text copy does not prove the source PDF or Word file is clean. If you attach the source instead, you disclose a different artifact.
Doing it by hand versus with a tool
Find-and-replace works for short documents but misses variations such as initials, possessives, OCR errors in scans and names inside email addresses. A detection tool can help locate candidates, but it can also miss details or flag text that should remain, so the final read stays your job either way.
Plan for the return trip too. If the AI writes a letter or summary with placeholders, restore the details locally and verify who each sentence refers to before using it.
Where PiBye fits
How PiBye handles this
PiBye looks for names, addresses, emails, phone numbers, dates and identifying numbers on your Mac, in English and French documents, including scanned PDFs. It proposes consistent stand-ins like [PERSON_001] for your review, and you choose what should stay readable. Scans and complex layouts can need additional checking.
When the AI’s work comes back, PiBye puts the real details back into the document locally and asks you about any stand-in the AI changed or dropped. Your original documents never leave your Mac. Review the complete approved copy before you choose to share it with an assistant; detection and replacement do not establish anonymity.
Version 1.0.16 · Apple Silicon (M1 or later) · macOS 14.8.5 or later · 1.1 GB
Frequently asked questions
Is removing names enough to protect a client?
Not always. The Law Society of Ontario notes that anonymizing is not perfect and remaining facts can identify someone. Remove other identifiers and read the full copy before sharing.
Should I use placeholders or black boxes?
Placeholders, when the goal is AI help. Consistent placeholders let the AI track who is who; black boxes suit documents filed or published without the AI step.
Does this replace client consent?
No. Guidance such as the Law Society of Ontario’s says to explain risks and obtain informed consent if confidentiality cannot be adequately protected by anonymizing.
Sources
- The Generative AI Playbook, Law Society of Alberta, updated February 2026. Checked 17 September 2026.
- Guidance on Professional Responsibility and Generative AI, Law Society of British Columbia, October 2023. Checked 17 September 2026.
- Licensee use of generative artificial intelligence (white paper), Law Society of Ontario, April 2024. Checked 17 September 2026.
- Generative AI: Your professional obligations, Law Society of Ontario, April 10, 2024. Checked 17 September 2026.