Microsoft Copilot Privacy vs ChatGPT: What Are the Differences?

Artificial intelligence is becoming embedded in everyday business processes. Employees are using it to draft emails, summarise meetings, analyse documents, research topics, create presentations and solve technical problems.

Two of the best-known AI platforms are Microsoft Copilot and ChatGPT. Although they can perform many similar tasks, the way they access, process and govern company information can be significantly different.

However, comparisons between the two platforms are often oversimplified. Microsoft offers several products under the Copilot name, while ChatGPT is available through personal, Business and Enterprise accounts. The privacy protections you receive depend on the exact product, licence, account and configuration being used.

For this article, Microsoft Copilot primarily means Microsoft 365 Copilot and Microsoft 365 Copilot Chat used through a business Microsoft 365 account.

The Most Important Difference: Which Version Are You Using?

It is misleading to say that Microsoft Copilot is private while ChatGPT is not.

A more accurate comparison is:

• Microsoft 365 Copilot used through a managed business account.
• ChatGPT Free, Plus or Pro used through a personal workspace.
• ChatGPT Business or ChatGPT Enterprise used through an organisation-managed workspace.

These are not equivalent environments.

Microsoft 365 Copilot includes enterprise data protection when used through an eligible work account. Microsoft says prompts, responses and information accessed through Microsoft Graph are not used to train its foundation models. (Microsoft Learn⁠)

OpenAI also states that it does not use inputs and outputs from ChatGPT Business or ChatGPT Enterprise to train its models by default. (OpenAI⁠)

The larger privacy difference therefore usually appears when employees use personal ChatGPT accounts for business information.

Microsoft Copilot vs ChatGPT at a Glance

The exact features available vary by subscription and configuration.

How Microsoft 365 Copilot Handles Business Information

Microsoft 365 Copilot is designed to operate within the Microsoft 365 environment.

It can use Microsoft Graph to identify information relevant to the user, including permitted emails, documents, calendar appointments, Teams conversations and meetings. It can then use that information to produce a more contextually relevant answer.

Microsoft states that Microsoft 365 Copilot only surfaces organisational information that the individual user already has permission to view. The same identity and access boundaries used elsewhere in Microsoft 365 continue to apply. (Microsoft Learn⁠)

This means Copilot does not give an employee permission to open a restricted document simply because they ask for it.

However, Copilot may make information that the employee can technically access much easier to discover.

For example, an employee may have been given access to an old SharePoint site several years ago and forgotten about it. If the permission was never removed, Copilot could potentially use information from that site because the employee remains authorised to view it.

This is why businesses should review SharePoint, Teams and OneDrive permissions before deploying Copilot. Microsoft specifically recommends identifying overshared, inactive, ownerless or sensitive sites and correcting excessive access. (Microsoft Learn⁠)

Are Microsoft Copilot Prompts Used for AI Training?

Microsoft says that prompts, responses and information accessed through Microsoft Graph are not used to train the foundation large language models used by Microsoft 365 Copilot. (Microsoft Learn⁠)

Copilot interaction information is still stored. This can include the user’s prompt, Copilot’s response and references used to ground the response.

The stored information supports functions such as Copilot activity history, auditing, compliance investigations and retention. Microsoft says it is encrypted while stored and can be managed by administrators using tools such as Microsoft Purview. (Microsoft Learn⁠)

Therefore, “not used for model training” does not mean “nothing is recorded”.

Businesses should understand what is retained, who can investigate it and which retention policies apply.

How Personal ChatGPT Accounts Handle Information

Personal ChatGPT accounts offer users controls over whether their conversations can be used to improve OpenAI’s models.

Users can turn off Improve the model for everyone through ChatGPT’s Data Controls. Once disabled, conversations can remain in the user’s history but will not be used to train ChatGPT. (OpenAI Help Center⁠)

Temporary Chat provides another option. OpenAI states that Temporary Chats:

• Do not appear in chat history.
• Do not create memories.
• Are not used to train its models.
• Are deleted from OpenAI’s systems after 30 days.
• May be reviewed for abuse monitoring. (OpenAI Help Center⁠)

These are useful personal privacy controls, but they are not the same as centralised business governance.

An employer may have no reliable way to determine whether an employee has changed their training setting, enabled a Temporary Chat, deleted a conversation or uploaded company information to a personal account.

That lack of central control is one of the main reasons businesses should not rely on employees’ personal AI accounts for confidential work.

How ChatGPT Business and Enterprise Differ

ChatGPT Business and ChatGPT Enterprise are designed for organisational use.

OpenAI states that data from its business products—including prompts, uploaded files and outputs—is not used to train its models by default. Business data is also encrypted in transit and at rest. (OpenAI Help Center⁠)

ChatGPT Business provides a managed workspace with administration controls, while Enterprise adds more advanced governance capabilities. Depending on the plan and configuration, these can include:

• Identity and access management.
• Domain verification.
• Role-based administration.
• User provisioning.
• Retention controls.
• Compliance capabilities.
• Enterprise encryption controls.
• Workspace analytics and auditing.

Some advanced controls, such as SCIM provisioning, custom retention and Enterprise Key Management, are associated with eligible Enterprise arrangements rather than every ChatGPT subscription. (OpenAI Help Center⁠)

Businesses should therefore check the exact capabilities included with their chosen plan rather than assuming every paid ChatGPT subscription provides enterprise governance.

Integration with Company Information

One of Microsoft 365 Copilot’s main differences is its native connection with the Microsoft 365 environment.

Copilot can work with content from:

• Exchange Online.
• Microsoft Teams.
• SharePoint.
• OneDrive.
• Outlook calendars.
• Microsoft 365 applications.
• Approved third-party information sources.

It applies the user’s existing Microsoft identity and content permissions when retrieving that information. (Microsoft Learn⁠)

ChatGPT can also connect to organisational information through approved apps and connectors. These may include services such as SharePoint, Google Drive, GitHub and other workplace platforms.

OpenAI states that connected apps respect the organisation’s existing permissions and require users to authenticate with the connected application. Information accessed from apps is not used to train OpenAI models by default for Business and Enterprise customers. (OpenAI⁠)

The difference is therefore becoming less about whether the platform can access business systems and more about:

• Which systems the organisation already uses.
• How tightly integrated the AI needs to be.
• Which controls are available under the chosen subscription.
• How applications and connectors are approved.
• Whether permissions have been correctly configured.

Microsoft Copilot Inherits Microsoft 365 Governance

Microsoft 365 Copilot can inherit several controls already used within Microsoft 365, depending on the organisation’s licences and configuration.

These can include:

• Microsoft Entra identity controls.
• Multifactor authentication.
• Conditional Access.
• SharePoint and OneDrive permissions.
• Microsoft Purview sensitivity labels.
• Data loss prevention policies.
• Retention policies.
• eDiscovery.
• Audit and content-search capabilities.

Microsoft says Copilot respects the organisation’s identity model and permissions, applies retention policies, inherits sensitivity labels and supports auditing of user interactions. (Microsoft Learn⁠)

For businesses already heavily invested in Microsoft 365, this can provide a more consistent governance experience.

However, these protections must be configured correctly. Buying a Copilot licence does not automatically repair weak permissions, missing labels or poorly managed information.

What Happens When Web Search Is Used?

Microsoft 365 Copilot can use web search to improve responses involving current or public information.

When this occurs, Copilot creates a shortened web query from the user’s prompt and sends it to Bing. Microsoft says user and tenant identifiers are removed from these queries, and the queries are not shared with advertisers or used to train foundation models. (Microsoft Learn⁠)

However, web search is a separate connected experience with its own data-handling arrangements. Microsoft also explains that Bing operates separately from Microsoft 365 for these queries.

Employees should avoid assuming that every external search, agent or third-party integration is governed in exactly the same way as information processed entirely within Microsoft 365.

The same general principle applies to ChatGPT. When an external app, website, action or third-party service is used, businesses should review that third party’s privacy terms, data access and retention arrangements.

Apps, Connectors and Agents Create Additional Risk

Both platforms can extend their capabilities through apps, connectors, agents and other integrations.

These additions can improve productivity, but they can also introduce new data pathways.

For Microsoft 365 Copilot, administrators can review an agent’s required permissions, terms of use and privacy statement before allowing it within the organisation. Microsoft advises customers to examine how each agent handles company information. (Microsoft Learn⁠)

ChatGPT administrators can also control which apps are enabled. OpenAI states that apps are enabled by default in ChatGPT Business workspaces, while they are disabled by default in Enterprise and Edu until approved by an administrator. (OpenAI Help Center⁠)

This is an important configuration difference. A business adopting ChatGPT Business should actively review its app settings instead of assuming that every connector begins disabled.

Before enabling an integration, the organisation should determine:

• What information the integration can access.
• Whether it can write or change data.
• Where information is processed.
• How long information is retained.
• Whether the provider uses it for another purpose.
• What happens when an employee leaves.
• Whether access can be centrally withdrawn.
• Whether activity is recorded for auditing.

Can Administrators See Employees’ AI Conversations?

Employees should not assume that conversations within a managed business account are necessarily private from their employer.

In ChatGPT Business, individual users have their own histories and other ordinary workspace members do not automatically see their conversations. However, organisation-managed accounts remain subject to the organisation’s controls and policies. OpenAI’s privacy information states that business-account administrators may be able to access and control managed accounts, including account content. (OpenAI Help Center⁠)

ChatGPT Enterprise also offers compliance and audit capabilities for authorised administrators. (OpenAI⁠)

Similarly, Microsoft 365 Copilot interactions can be retained and searched using Microsoft compliance tools, subject to the organisation’s licensing, policies and permissions. (Microsoft Learn⁠)

Businesses should communicate this clearly through their employee privacy notices and acceptable-use policies.

Which Platform Is More Private?

There is no universally correct answer.

For a business already using Microsoft 365, Microsoft 365 Copilot often offers the most natural governance model because it operates within the existing Microsoft identity, permissions, security and compliance environment.

It may be particularly appropriate when the main objective is to work with:

• Outlook emails.
• Teams meetings.
• Word documents.
• Excel workbooks.
• SharePoint information.
• OneDrive files.
• Existing Microsoft security and compliance policies.

ChatGPT Business and Enterprise can also provide strong business privacy commitments. They may be attractive when an organisation wants a broader standalone AI workspace, advanced reasoning capabilities, research tools, coding functions or connections across multiple technology platforms.

The more important comparison is not simply Copilot versus ChatGPT. It is:

Which precise product, subscription, configuration and governance model is suitable for the information being processed?

Should Employees Use Personal ChatGPT Accounts for Work?

Businesses should be extremely cautious about allowing employees to enter confidential information into unmanaged personal AI accounts.

Even where the user has disabled model training, the organisation may still lack:

• Central account ownership.
• Enforced retention settings.
• Employee offboarding controls.
• Consistent access policies.
• Approved connector controls.
• Compliance auditing.
• Legal-hold capabilities.
• Visibility of company information being uploaded.
• The ability to apply company-wide security settings.

Personal accounts may still be suitable for low-risk, public or non-confidential activities when permitted by company policy. Examples could include generating generic ideas or rewriting text that contains no personal, customer or commercially sensitive information.

Employees should never be expected to decide alone what is safe to upload. The organisation should establish clear classifications and examples.

Neither Platform Prevents Human Error

Microsoft Copilot and ChatGPT can both be used securely, but neither can prevent every poor decision.

An employee could still:

• Include unnecessary personal information in a prompt.
• Upload the wrong customer document.
• Share an AI conversation with an unauthorised person.
• Connect an unapproved application.
• Accept an inaccurate AI-generated answer.
• Expose information through excessive permissions.
• Copy sensitive output into an insecure system.
• Use a personal account instead of the approved business platform.

Privacy therefore depends on a combination of technology, configuration, policy, training and human review.

Seven Steps for Safer Business AI

1. Approve Specific AI Products

Define exactly which AI platforms, plans and accounts employees are authorised to use.

“Employees may use ChatGPT” is not specific enough. The policy should identify whether this means ChatGPT Business, Enterprise or a restricted use of personal accounts.

2. Classify Your Information

Employees need to understand the difference between public, internal, confidential and highly restricted information.

The organisation should decide which categories can be used with AI and under what conditions.

3. Review Microsoft 365 Permissions

Before deploying Microsoft 365 Copilot, review SharePoint, Teams and OneDrive access. Remove company-wide sharing links, historic access and permissions that are no longer required.

4. Control Apps and Connectors

Only approve integrations that have been reviewed for data access, security, privacy and business value.

5. Configure Retention and Auditing

Decide how long AI conversations should be kept, who can investigate them and how they will be handled during legal, regulatory or disciplinary processes.

6. Train Employees

Training should cover privacy as well as prompt writing. Staff should know what they may enter, what must remain excluded and how to check AI-generated output.

7. Complete a Risk Assessment

Businesses should consider conducting a data-protection impact assessment or equivalent AI risk assessment, particularly where personal, regulated or highly confidential information will be processed.

The Final Verdict

Microsoft 365 Copilot has a significant privacy advantage for organisations that want AI deeply integrated with an existing and correctly governed Microsoft 365 environment.

It can apply Microsoft identities, permissions, sensitivity labels, retention controls and auditing while keeping prompts, responses and Microsoft Graph data out of foundation-model training.

ChatGPT Business and Enterprise also provide strong commercial privacy commitments, including no training on business data by default, encryption and organisational controls.

The greatest concern is not necessarily ChatGPT itself. It is the unmanaged use of personal ChatGPT accounts for company information.

A secure AI strategy should therefore avoid simplistic claims that one brand is safe and another is unsafe. Businesses must compare the correct product versions, review their contractual protections, configure the available controls and establish clear rules for employees.

How Hamilton Group Can Help

Hamilton Group can help your organisation adopt AI tools without losing control of sensitive company information.

We can review your Microsoft 365 permissions, assess Copilot readiness, improve SharePoint governance, configure Microsoft Purview protections and help you compare Microsoft 365 Copilot with ChatGPT Business or Enterprise.

We can also help develop an acceptable-use policy, identify appropriate AI use cases and train employees to use generative AI safely and responsibly.

To discuss Microsoft Copilot, ChatGPT or your wider business AI strategy, contact Hamilton Group on 0330 043 0069.