How to Redact Contracts Before Using ChatGPT, Copilot or Claude

AI can be useful for contract work.

It can summarise long documents, explain clauses, compare terms, extract obligations, draft responses, prepare issue lists and help commercial teams work faster.

But there is a problem.

Most contracts and commercial documents contain information that should not be copied straight into ChatGPT, Copilot, Claude, Gemini or any other AI tool.

Client names. Project names. Contract values. Site addresses. Personal details. Claim references. Commercial positions. Internal correspondence.

Before a document goes into AI, it should be cleaned.

That does not mean making the document useless.

It means removing or replacing confidential information while keeping enough structure for the AI tool to understand what the document is about.

This guide explains how to redact and anonymise contracts before using AI.

Why contracts need cleaning before AI

Contracts are not just legal documents.

They often contain a full commercial picture of a project, relationship or dispute.

A typical contract, tender, claim or project document may include:

• company names
• client names
• subcontractor names
• project names
• site addresses
• contract values
• payment terms
• programme dates
• named individuals
• email addresses
• dispute positions
• claim values
• confidential technical information
• internal comments or negotiation history

AI tools do not need all of that information to be useful.

For example, if you want AI to summarise a clause, explain a risk, compare two positions or produce a list of obligations, it usually does not need the real client name, exact site address or actual contract value.

The goal is simple:

Use AI on the document structure, not on the confidential identity of the project.

Redaction vs anonymisation

People often use "redaction" and "anonymisation" to mean the same thing.

For AI workflows, there is an important difference.

Redaction removes information

Traditional redaction blocks out or removes sensitive text.

Example:

The contractor shall complete the works at █████ by █████ for the sum of █████.

This protects information, but it also removes context.

The AI tool may no longer understand the relationship between the parties, the project, the site, the value or the obligation.

Anonymisation replaces information with useful placeholders

Anonymisation keeps the structure of the document while replacing confidential details.

Example:

The contractor shall complete the works at [SITE ADDRESS] by [COMPLETION DATE] for the sum of [CONTRACT VALUE].

This is usually more useful for AI.

The confidential information is removed, but the document still makes sense.

What to remove before using AI on a contract

Before uploading, pasting or processing a contract with AI, check for the following categories.

1. Company and client names

Replace real organisations with placeholders. Examples:

• [COMPANY A]
• [COMPANY B]
• [CLIENT]
• [CONTRACTOR]
• [SUBCONTRACTOR]
• [SUPPLIER]

This lets AI understand the roles without exposing the real parties.

2. Project names

Project names can be highly sensitive, especially in construction, infrastructure, engineering, energy, public sector and commercial development. Replace them with:

• [PROJECT NAME]
• [PROJECT A]
• [DEVELOPMENT]
• [WORKS PACKAGE]

3. Site addresses and locations

Addresses can identify a project even when company names are removed. Replace them with:

• [SITE ADDRESS]
• [PROJECT LOCATION]
• [FACILITY]
• [REGION]

You may keep broad geography if needed. For example:

• [UK SITE]
• [NORDIC PROJECT]
• [LONDON PROJECT]
• [EU FACILITY]

4. Personal names and contact details

Remove names, email addresses, phone numbers and signatures unless they are genuinely needed. Replace with:

• [PERSON A]
• [PROJECT MANAGER]
• [COMMERCIAL MANAGER]
• [LEGAL COUNSEL]
• [EMAIL ADDRESS]
• [PHONE NUMBER]

5. Contract values and rates

Financial information is often one of the most commercially sensitive parts of a document. Replace exact figures with placeholders. Examples:

• [CONTRACT VALUE]
• [CLAIM VALUE]
• [DAY RATE]
• [HOURLY RATE]
• [PAYMENT AMOUNT]
• [RETENTION PERCENTAGE]

In some cases, you may want to preserve the scale without exposing the exact amount. Example:

• [LOW SIX-FIGURE VALUE]
• [MULTI-MILLION CONTRACT VALUE]
• [HIGH-VALUE CLAIM]

6. Dates and programme information

Dates can reveal project timelines, delay positions and commercial leverage. Replace with:

• [START DATE]
• [COMPLETION DATE]
• [PAYMENT DATE]
• [NOTICE DATE]
• [DELAY PERIOD]
• [PROGRAMME MILESTONE]

If the timing relationship matters, keep the structure. Example:

Notice was issued [NUMBER] days after [EVENT].

7. Claim and dispute details

Claims, variations, extensions of time and dispute documents often contain sensitive positions. Replace:

• claim numbers
• dispute references
• settlement values
• named decision makers
• negotiation positions
• internal legal comments

Use placeholders such as:

• [CLAIM REFERENCE]
• [VARIATION NUMBER]
• [DISPUTE REFERENCE]
• [SETTLEMENT POSITION]
• [WITHOUT PREJUDICE COMMENT]

8. Unique technical or commercial identifiers

Some information can identify a project even if names are removed. Check for:

• drawing numbers
• package references
• asset numbers
• internal cost codes
• procurement references
• tender IDs
• document control numbers
• file names
• metadata

Replace with:

• [DRAWING REFERENCE]
• [PACKAGE NUMBER]
• [ASSET ID]
• [DOCUMENT REFERENCE]
• [TENDER REFERENCE]

Bad redaction vs AI-ready redaction

The best AI preparation protects confidential information while keeping the document useful.

Poor redaction

The contractor submitted a claim for █████ against █████ due to delays at █████.

This removes the sensitive details, but it also removes meaning.

Better redaction

[COMPANY A] submitted a claim for [CLAIM VALUE] against [COMPANY B] due to delays at [PROJECT NAME].

This is much more useful. The AI tool can still understand:

• who is making the claim
• who the claim is against
• that the issue relates to delay
• that a claim value exists
• that the project identity has been removed

That is the goal.

A practical workflow for using contracts with AI

Use this workflow before using AI on contracts, tenders, claims or commercial documents.

Step 1: Work from a copy

Never clean the only version of the document. Create a duplicate and label it clearly.

Example:

Original: Main Contract - Project Name - Signed.pdf
AI-ready copy: AI-ready contract - anonymised version.docx

Keep the original separate.

Step 2: Remove obvious identifiers

Start with the easy items:

• company names
• project names
• addresses
• personal names
• email addresses
• phone numbers
• signatures

Replace them with consistent placeholders. Do not use a different placeholder every time the same party appears.

If the same company appears throughout the contract, use the same placeholder throughout.

Use [COMPANY A] every time, not [COMPANY A], [CLIENT], [PARTY 1] and [EMPLOYER] interchangeably.

Step 3: Replace commercial values

Search for currency symbols, numbers and payment language. Look for:

• £, $, €, SEK
• contract sum
• rates
• daywork
• variation
• claim
• damages
• retention
• payment schedule

Replace exact values with useful placeholders. Example:

The contract sum is [CONTRACT VALUE].

Step 4: Preserve the structure

Do not over-clean the document. AI tools need context.

A useful AI-ready document should still show:

• the roles of the parties
• the type of agreement
• the obligations
• the sequence of events
• the commercial logic
• the risk allocation
• the issue being reviewed

The goal is not to destroy the document. The goal is to remove unnecessary exposure.

Step 5: Check document metadata

Confidential information may sit outside the main text. Check:

• file name
• comments
• tracked changes
• document properties
• author names
• hidden text
• headers and footers
• embedded attachments
• screenshots
• copied email chains

These are easy to miss.

Step 6: Use AI on the cleaned version

Once the document has been cleaned, use the AI-ready version for prompts such as:

• "Summarise the main obligations in this contract."
• "Identify the key commercial risks."
• "Create a list of notice requirements."
• "Explain this clause in plain English."
• "Compare these two positions."
• "Draft questions for a commercial review."
• "Extract deadlines, payment triggers and approval steps."

Do not paste the original confidential document if a cleaned version would work.

Example AI-ready prompt

After cleaning the document, you could use a prompt like this:

Review the anonymised contract below. Summarise the key commercial obligations, notice requirements, payment triggers, delay risks and any clauses that may need legal or commercial review. The placeholders such as [COMPANY A], [PROJECT NAME] and [CONTRACT VALUE] represent removed confidential information.

This tells the AI tool how to treat the placeholders.

It also keeps the review focused on the structure and meaning of the document.

Contract types that should be cleaned before AI

This approach is useful for more than signed contracts. It can also apply to:

• tenders
• subcontract agreements
• supply agreements
• framework agreements
• scopes of work
• payment notices
• variation claims
• extension of time claims
• correspondence
• meeting minutes
• commercial reports
• procurement documents
• settlement drafts
• contract summaries
• board or project papers

Any document that combines commercial value, project detail and named parties should be reviewed before being used with AI.

Common mistakes to avoid

Mistake 1: only removing company names

Company names are not the only identifiers. A project can often be identified from its address, contract value, programme dates, document references or technical scope.

Mistake 2: redacting so much the document becomes useless

If all context is removed, AI cannot help properly. Use structured placeholders instead.

Mistake 3: forgetting comments and tracked changes

Comments, edits and metadata can contain sensitive information. Clean the document itself, not just the visible body text.

Mistake 4: mixing placeholders

Be consistent. If [COMPANY A] is the contractor, keep [COMPANY A] as the contractor throughout the document.

Mistake 5: assuming every AI use case needs the full document

Often, the AI task only needs a clause, schedule, extract or correspondence chain. Use the smallest useful section.

AI-ready contract checklist

Before using a contract or commercial document with AI, check:

• Have company names been replaced?
• Have project names been replaced?
• Have site addresses been removed or generalised?
• Have personal names and contact details been removed?
• Have contract values and rates been replaced?
• Have claim values and dispute references been removed?
• Have document references and drawing numbers been checked?
• Have comments, tracked changes and metadata been reviewed?
• Are placeholders consistent throughout?
• Does the cleaned document still make sense?
• Are you using only the section needed for the task?

If the answer is yes, the document is much closer to being AI-ready.

Where OP Redact fits

OP Redact is being built for people who want to use AI with contracts, tenders, claims and commercial documents without unnecessarily exposing confidential information.

It creates clean, AI-ready versions of sensitive documents by replacing confidential details with useful placeholders such as:

• [COMPANY A]
• [PROJECT NAME]
• [CONTRACT VALUE]
• [SITE ADDRESS]
• [PERSON A]
• [CLAIM REFERENCE]

The aim is not to make the document unreadable.

The aim is to make it safer to use with AI while keeping the structure useful.

Documents are processed locally. No contract upload is required.

FAQs