What Information Should You Not Put Into AI?

AI produces better answers when it understands your business.

That does not mean you should give it every document, conversation or piece of data you have.

A team member may paste a client email into AI to draft a response.

An owner may upload a contract for review.

A manager may enter employee performance information to prepare for a difficult conversation.

The intention is usually practical.

Get the work done faster.

But before information enters an AI tool, your business needs to ask:

Should this information be shared with this tool, through this account, for this purpose?

What information should you not put into AI?

Do not enter personal, confidential, security-sensitive, legally protected or proprietary information into an AI tool unless:

• The tool and account are approved for that type of information
• You have the authority to use the information
• The information is necessary for the task
• The provider’s privacy, security and data-use practices have been reviewed
• Appropriate safeguards are in place
• A person remains responsible for the work and its outcome

The safest starting point is simple:

Give AI the minimum information it needs to support the work.

Do not give it every piece of information you have.

A question I hear from business owners

In client conversations, I often hear a version of the same question:

“Can I just upload everything and let AI sort it out?”

My answer is usually:

Not yet.

Before you upload a folder, transcript or document, you need to know:

• What information it contains
• Who owns that information
• Whether you have permission to use it
• Whether the tool is approved for that information
• What the tool may do with the information
• Who will review the output

One client wanted to use AI as an external brain for the business.

That made sense.

The business had years of documents, operating information and client knowledge spread across different folders and systems.

But the first step was not uploading everything.

We first had to separate general business context from confidential client information, employee information, login details and documents that required additional care.

That structure made the AI more useful.

It also reduced the risk of sharing information that the AI did not need.

Structure first. Then AI.

Your prompt is business data

Many people treat a prompt like a casual question.

In business, it may contain:

• Customer information
• Employee details
• Financial results
• Internal decisions
• Intellectual property
• Strategic plans
• Confidential conversations

The prompt is not separate from your business data.

The prompt is business data.

The same applies to anything you:

• Paste
• Upload
• Connect
• Record
• Transcribe
• Import
• Ask an AI agent to access

The fact that information can be uploaded does not mean it should be.

Why can entering information into AI create risk?

When you put information into an AI tool, it may move outside the system where it was originally collected and protected.

Depending on the provider, product, account and settings, the information may be:

• Stored for a period of time
• Processed in another jurisdiction
• Retained in conversation history or system logs
• Accessible to service providers or subcontractors
• Used to improve the service
• Connected to other systems or applications
• Exposed through weak account security
• Used in a way that conflicts with a client agreement

The Government of Canada warns that some AI suppliers may inspect input data or use it to further train their models. It also notes that information may be stored on systems outside the organization’s control or retained longer than necessary.

This does not mean your business cannot use AI with internal information.

It means you need to understand the tool, account and purpose before sharing it.

1. Personal and identifying information

Do not put identifiable personal information into an unapproved AI tool.

This may include:

• Full names
• Personal email addresses
• Phone numbers
• Home addresses
• Dates of birth
• Government identification numbers
• Driver’s licence or passport details
• Customer account numbers
• Photographs of identifiable people
• Information about children
• Details that could identify someone when combined

Removing a person’s name does not always make the information anonymous.

A job title, location, employer, age and description of an incident may still make the person identifiable.

Canadian privacy regulators recommend using anonymized, synthetic or de-identified information when personal information is not necessary for the task.

Ask:

Does the AI need to know who this person is to complete the task?

In many cases, it does not.

2. Employee and human resources information

Employee information needs a higher level of care.

Do not paste raw employee records or sensitive workplace details into an unapproved tool.

This includes:

• Performance reviews
• Disciplinary records
• Compensation information
• Payroll details
• Medical information
• Accommodation requests
• Workplace complaints
• Investigation notes
• Background checks
• Résumés containing personal information
• Termination discussions
• Individual productivity data

AI can help you create a general performance-review template.

It can help you prepare questions for a difficult conversation.

It does not need an employee’s full file to do that.

Instead of entering:

“Help me respond to Jane Smith, who is 54, has a medical accommodation and has missed six deadlines.”

You could write:

“A manager in a small professional services firm needs to address repeated missed deadlines while following an existing workplace accommodation. Help prepare respectful questions for the conversation. Do not make legal or HR conclusions.”

The second version provides relevant context without identifying the employee.

You still need a qualified person to review decisions involving employment, discipline, accommodation or termination.

3. Confidential client and customer information

Your clients may have trusted you with information they do not expect to be shared with an external AI provider.

That information can include:

• Client records
• Meeting transcripts
• Discovery notes
• Customer complaints
• Internal business challenges
• Business plans
• Pricing information
• Financial results
• Contracts
• Legal correspondence
• Proprietary processes
• Unpublished product information
• Login or system-access details

Before putting client information into AI, check:

• Your contract
• Confidentiality clauses
• Privacy commitments
• Data-processing requirements
• Industry obligations
• The client’s expectations
• The tool’s terms and settings

A client may have given you information so you can deliver a service.

That does not automatically mean you have permission to provide it to another technology provider for a different purpose.

When possible, summarize the issue without naming the client.

For example:

“A five-person consulting firm is experiencing a decline in repeat business. Here are its anonymized revenue percentages by service category.”

The AI may not need the client’s name, customer list, bank statements or complete CRM export.

4. Passwords and security information

Never enter passwords, access codes or live security credentials into a general AI prompt.

This includes:

• Passwords
• PINs
• Security-question answers
• API keys
• Authentication tokens
• Private encryption keys
• Recovery codes
• Database credentials
• Network configurations
• Detailed security vulnerabilities
• Complete system-access instructions

AI can help explain how an integration works.

It does not need the live credentials required to access your system.

Replace real values with placeholders such as:

• [API KEY]
• [CLIENT ID]
• [PASSWORD]
• [PRIVATE URL]

If a real credential is accidentally entered into a tool, replacing or revoking that credential is safer than assuming that deleting the conversation has removed the risk.

5. Financial and payment information

Avoid entering financial information that could expose your business, an employee or a customer.

This includes:

• Bank account numbers
• Credit card numbers
• Payment details
• Tax identification numbers
• Payroll files
• Personal tax records
• Loan documentation
• Investor information
• Customer-level financial records
• Unpublished financial results where confidentiality matters

AI may help you identify trends in:

• Revenue
• Expenses
• Gross margin
• Cash flow
• Sales performance
• Service profitability

It often does not need every transaction, customer name or account number.

Use totals, categories, ranges and anonymized data where possible.

For example, provide monthly revenue by service line rather than a complete transaction file connected to individual customers.

6. Legal, privileged or regulated information

Use extra care with information connected to:

• Legal advice
• Active disputes
• Litigation
• Insurance claims
• Regulatory investigations
• Health records
• Financial services
• Education records
• Government contracts
• Protected research
• Professional confidentiality requirements

The question is not whether AI can summarize the document.

The question is whether you have the authority to share the document with that provider.

Uploading legally sensitive information to an external tool could affect confidentiality, contractual obligations or professional responsibilities.

Confirm the permitted use before uploading it.

7. Trade secrets and proprietary business information

AI can support innovation without receiving your entire competitive advantage.

Protect information such as:

• Unpublished inventions
• Product formulas
• Source code
• Algorithms
• Pricing models
• Acquisition plans
• Bid strategies
• Supplier terms
• Proprietary methodologies
• Detailed customer lists
• Unreleased research
• Merger or investment discussions

Ask what the AI needs to complete the task.

It may need a description of the problem.

It may not need the complete formula, code base, methodology or commercial strategy behind it.

8. Information you do not own

Information does not need to be personal to create risk.

Be cautious about uploading:

• Licensed reports
• Paid research
• Copyrighted books
• Training materials
• Supplier documentation
• Partner resources
• Purchased templates
• Other people’s intellectual property

Having access to information does not always give you the right to upload, reproduce or transform it using another provider.

Check the licence, agreement or permission attached to the material.

Is it safe to use confidential information in a paid AI account?

Not automatically.

A paid or business plan may offer stronger privacy, security and administrative controls than a free public account.

But paying for a subscription does not automatically make every type of information safe to use.

Before approving a tool for business information, review:

• Whether prompts and files are used to train or improve models
• How long information is retained
• Whether users can delete conversations and files
• Who can access the information
• Where data is stored or processed
• Whether the business controls the account
• Whether administrative controls are available
• Whether users can connect email, cloud storage or other systems
• Whether the provider offers an appropriate business agreement
• Whether the tool meets client, industry and legal requirements

Do not assume the default settings match your business requirements.

Confirm them.

Does deleting an AI conversation remove the risk?

Deleting a conversation may remove it from your visible history, but it does not undo the original disclosure.

Provider retention and deletion practices vary.

The better approach is to avoid entering unnecessary sensitive information in the first place.

If information has been entered accidentally:

• Record what was shared
• Identify the tool and account used
• Delete the conversation or file where possible
• Change any exposed passwords or credentials
• Review the provider’s retention practices
• Follow your privacy or incident-response process
• Determine whether a client, employee or other party needs to be informed

Your business should have a simple process for reporting AI-related mistakes.

Employees are more likely to report a problem early when they know what to do and are not afraid to ask for help.

Is anonymizing information enough?

Sometimes.

Good anonymization can reduce risk.

Replacing a name with “Client A” is not always enough.

Consider this example:

“Client A is the only veterinary clinic in a specific rural community, has 14 employees and is preparing to acquire its local competitor.”

The client’s name is gone.

The client may still be obvious.

When preparing information for AI:

• Remove names and contact details
• Remove account and identification numbers
• Remove unnecessary dates and locations
• Replace exact figures with ranges when precision is not required
• Summarize instead of uploading complete records
• Use generic job titles
• Use fictional or synthetic examples where possible
• Share only the section needed for the task

Anonymization should preserve what AI needs to understand while removing what it does not need to know.

Use an Allowed, Caution and Never framework

A small business does not need a complicated data-classification system to get started.

Use three categories.

Allowed

Information employees may use in approved AI tools.

Examples include:

• Public website content
• Published marketing materials
• General industry information
• Approved brand guidelines
• Generic templates
• Public job descriptions
• Non-sensitive business processes
• Fictional examples
• Synthetic data

Caution

Information that requires an approved tool, a clear purpose and appropriate safeguards.

Examples include:

• Anonymized client scenarios
• Internal procedures
• Financial trends
• Draft contracts
• Employee situations with identifying details removed
• Internal performance measures
• Unpublished marketing plans
• Business forecasts
• Internal meeting summaries

Never

Information that should not be entered unless a specific secure use has been reviewed and authorized.

Examples include:

• Passwords and access credentials
• Payment card information
• Government identification numbers
• Sensitive medical information
• Detailed employee records
• Confidential client records
• Legally privileged documents
• Trade secrets
• Information prohibited by a contract
• Personal information you are not authorized to use

Your categories may differ based on your business.

The important part is that your team knows the difference before opening the AI tool.

A practical small-business example

Consider a consulting firm that wants AI to summarize a client discovery meeting.

The fastest approach is to upload the entire transcript.

The better approach is to ask:

• Did the client know the meeting was being recorded?
• Does the transcript contain personal or confidential information?
• Is the AI tool approved for client information?
• Is the account controlled by the business?
• Does the full transcript need to be uploaded?
• Can identifying information be removed?
• Who will review the summary?
• When should the transcript and AI output be deleted?

The business may decide to remove:

• Names
• Contact details
• Customer information
• Financial account details
• Employee information
• Unrelated personal conversation

It may then upload only the section needed for the task.

That takes more thought than clicking “upload.”

It also creates a better business process.

A Canadian small-business consideration

Canadian businesses remain responsible for how they collect, use, disclose and protect personal information when using AI.

The Office of the Privacy Commissioner of Canada advises organizations using generative AI to establish a lawful basis for using personal information, provide transparency, limit the sharing of sensitive information and build privacy safeguards into their processes.

PIPEDA also requires organizations to identify the purposes for collecting personal information and appoint someone accountable for privacy compliance. Provincial privacy requirements may also apply depending on your location, industry and activities.

The practical business lesson is:

Your responsibility does not disappear because the work was completed through AI.

Five questions to ask before putting information into AI

Before you paste, upload or connect information, ask:

1. Is this information necessary?

Could the AI complete the task with less information?

2. Do we have the right to use it?

Does the information belong to the business, a client, an employee, a supplier or another party?

3. Is this tool approved?

Have you reviewed the account, settings, terms, storage and data-use practices?

4. Can we remove identifying or sensitive details?

Could you use a summary, range, placeholder or fictional example?

5. Who is responsible for the result?

Who will review the output and decide how it will be used?

If your team cannot answer those questions, stop before uploading the information.

A simple AI information rule for your team

You can use this as a starting point:

Before using AI, remove personal, confidential, financial, security-sensitive and client-identifying information unless the tool and use have been specifically approved. Use only the minimum information required for the task. A person must review the output before it is used or shared.

A policy only works when employees understand how to apply it.

Include examples based on the work your team completes every day.

Quick self-assessment

Ask yourself:

• Do employees know which AI tools they may use?
• Does the business control the accounts being used?
• Have you reviewed the tools’ privacy and data-use settings?
• Do employees know what information is prohibited?
• Can employees identify confidential client information?
• Do you have rules for employee and HR information?
• Do you have a process for anonymizing information?
• Does someone review new AI tools before use?
• Can employees report a mistake without hiding it?
• Do your client agreements address the use of third-party technology?
• Have you reviewed which systems AI tools can access?
• Is someone responsible for reviewing these rules?

If several answers are “no” or “I am not sure,” your next step is not another AI tool.

Your next step is to establish the rules.

AI prompt: Create your Allowed, Caution and Never information guide

Do not include confidential or identifying information when using this prompt.

Copy and paste:

You are helping a small business create practical rules for what information employees may put into approved AI tools.

Business overview:
[Describe the business, services, team size and industry without including confidential information.]

Common AI uses:
[List how the team currently uses or wants to use AI.]

Types of information handled:
[List general categories such as client emails, employee records, financial reports, contracts, marketing content and meeting notes. Do not include actual records or identifying information.]

Approved AI tools and accounts:
[List tools and account types, or state that this has not yet been confirmed.]

Create an Allowed, Caution and Never information guide for this business.

For each category:

1. Provide specific examples relevant to the business.
2. Explain why the information belongs in that category.
3. Identify any conditions required before the information can be used.
4. Suggest how sensitive information could be removed, summarized or replaced.
5. Identify situations that require human, privacy, legal, security or client approval.
6. Flag any missing information you would need before making a reliable recommendation.

Do not assume a paid AI account is automatically approved for confidential information.

Do not make legal conclusions.

Use plain language that a small team can apply during everyday work.

Finish with:

• A one-paragraph employee rule
• A five-question pre-upload checklist
• A process for reporting information entered by mistake
• Questions the business owner still needs to answer

Frequently asked questions

What should you never put into ChatGPT or another AI tool?

Do not put passwords, security credentials, payment information, government identification numbers, sensitive employee information, confidential client records, trade secrets or legally protected documents into an unapproved AI tool.

Can I paste a client email into AI?

Only when you have confirmed that the tool and account are approved for the information and that using it does not conflict with your client obligations. In many cases, you can remove the client’s name and other identifying details first.

Can I upload a contract to AI?

AI can help summarize or organize contract language, but first confirm that you have permission to share the document and that the tool is approved for confidential or legally sensitive information. A qualified person should review any legal interpretation or decision.

Can I put employee information into AI?

Avoid entering identifiable employee information into a general AI tool. Use anonymized scenarios where possible and maintain human oversight for employment decisions.

Is a business AI account safer than a free account?

It may provide stronger controls, but you still need to review the specific product, settings, contract, retention practices and permitted uses.

Can AI use public information about a person?

Publicly available information can still be personal information. Its availability does not automatically remove your privacy, consent or fairness responsibilities.

Who should decide what information employees can use?

The business owner or a named senior leader should approve the rules with input from the people responsible for privacy, security, HR, legal obligations and client relationships.

The question is not only what AI can do

AI can summarize a contract.

It can analyze a spreadsheet.

It can draft an employee communication.

It can organize years of business information.

That does not mean it should receive every detail contained in those materials.

Before you ask what AI can do with your information, ask:

Does AI need this information to do the job?

Do we have the right to share it?

Is this the right tool and account?

Who will remain responsible?

Better AI use does not begin with uploading everything.

It begins with knowing what not to share.

Structure first. Then AI.