The AI AdoptionFree AI pre-audit

Microsoft Copilot or Claude,
for a business on Microsoft 365?

YOU

“What did we agree with the supplier about the price rise, and where is the latest quote?” On the blog, our guides on AI adoption training for your team and what an AI adoption consultant does.

One question, asked to each assistant

Microsoft CopilotWith a Microsoft Copilot licence

Looks across, through Microsoft Graph

  • OutlookEmail, calendar
  • TeamsChats, meetings
  • SharePointSites, files
  • OneDriveFiles

Without the licence, Copilot Chat answers from the web and what you paste or upload, and in Outlook from your mail and calendar

Works inside, with the licenceWord Excel PowerPoint

ClaudeWith the Microsoft 365 connector

Looks across, as you, when you ask

  • OutlookEmail, calendar
  • TeamsChats, channels, transcripts
  • SharePointSites, files
  • OneDriveFiles

Read-only by default; sending and filing need your administrator

Add-ins, in the file or email you have openWord Excel PowerPoint Outlook

Both stop at the same line: what this person can already open

  • Supplier contractsOpen to this person
  • QuotesOpen to this person
  • PayrollNot open to them
  • Board papersNot open to them

Our drawing, with a fictional question, from Microsoft’s and Anthropic’s documentation read on 25 October 2026. It shows neither product’s interface.

For a business that already runs on Microsoft 365, the choice between Microsoft Copilot and Claude rarely comes down to which one writes better. It comes down to what each can reach in your email, files and meetings, what it may change there, who in the business switches it on and where your data goes. Microsoft also offers Claude models inside Copilot, switched on by default in Australia and New Zealand and off by default in the United Kingdom. This page compares the two from Microsoft’s and Anthropic’s own documentation, read on 25 October 2026.

By The AI Adoption · Published on 25 October 2026 · Sources: Microsoft and Anthropic documentation

What Copilot reaches in Microsoft 365

Microsoft has renamed its assistant: Microsoft 365 Copilot is now Microsoft Copilot, and Microsoft 365 Copilot Chat is now Microsoft Copilot Chat. They are two levels of the same product. Copilot Chat comes with eligible Microsoft 365 licences and answers mainly from the web, from files you upload and from what you paste in. Inside Outlook it goes further and can answer from your inbox, calendar, meetings and chats.

A Microsoft Copilot licence, assigned to each person who needs it, extends Copilot to the rest of your organisation’s data. Copilot then reasons across emails, meetings, calendars, chats, documents, sites and files through Microsoft Graph and a semantic index of your content, works with its full experience inside Word, Excel, PowerPoint and OneNote, and gives access to the more advanced agents. For businesses on Microsoft 365 Business plans, Microsoft sells this licence as an add-on called Microsoft Copilot Business, on an annual commitment.

Copilot combines that content with what you are doing at the time, such as the meeting you are in or last week’s chats on a topic. Microsoft says the prompts, the responses and the data Copilot reads through Microsoft Graph are not used to train the foundation models, including the ones Copilot uses.

What Claude reaches in Microsoft 365

Claude reaches Microsoft 365 in two ways. The first is the Microsoft 365 connector, available on every Claude plan, which lets Claude search SharePoint, OneDrive, Outlook, Teams chats and channels, calendars, and meeting transcripts and recordings when someone asks a question that needs them. It is read-only by default and only works with a work account on a Microsoft Business plan; personal Outlook and Hotmail addresses cannot connect.

An administrator can add write tools, which let Claude send and draft email, manage calendar events, create and update files in OneDrive and SharePoint and post in Teams. For Teams, each send tool can be set so that the person confirms every message, and the emails Claude sends carry a header that marks them as sent by an agent. Anthropic notes that changes to files and calendars and Teams messages are not tagged in the same way, and that creating or changing a calendar event sends the attendees Microsoft’s usual invitation or update.

The second way is the Claude add-ins, which work inside the apps your team already opens. Claude for Excel, Word and PowerPoint are generally available on the Pro, Max, Team and Enterprise plans: Claude works in the open workbook while keeping its formulas intact, edits a document as tracked changes you accept or reject, and builds slides from your own template. Claude for Outlook is in beta. It sorts your unread mail, drafts replies in your voice and prepares a brief for your next meeting, and it never sends mail itself, because every draft lands unsent in Outlook and you press Send. The add-ins only see the files and the email you have open, and they share that context with each other within one conversation.

The same line for both: what each person can open

On the question that matters most to an owner, the two products give the same answer. Microsoft writes that Copilot only surfaces organisational data to which each person has at least view permission. Anthropic writes that Claude mirrors each person’s existing Microsoft 365 permissions, so that nobody can reach anything through Claude that they could not already open in Microsoft 365.

In practice, whichever assistant you choose will show each person what your SharePoint sites, shared folders and mailboxes already let them open, including a folder shared with everyone years ago, if nobody has changed its access since. Microsoft itself asks its customers to use the permission settings of SharePoint and its other services so that the right people reach the right content. Tidying who can open what therefore comes before switching either one on, and it is the same work for both. We cover it in answers your team can find without asking you and in adopting AI securely.

Claude can already run inside Copilot

The choice is not always one or the other. Anthropic works as a Microsoft subprocessor, and Microsoft offers Claude models inside Microsoft Copilot, Researcher, Copilot Studio, Power Platform and Copilot in the Microsoft 365 apps. People can select Claude in Researcher and in Edit with Copilot, the person building an agent in Copilot Studio chooses its model, and Copilot shows an indicator when a Claude model is in use. Microsoft’s Product Terms and Data Protection Addendum apply to that use, except for the models labelled as keeping data with Anthropic.

Where your business is decides the starting point. Microsoft turns Anthropic models on by default for most customers in its commercial cloud, and lists the European Union, the European Free Trade Association and the United Kingdom as the regions where they start switched off. Australia and New Zealand are not on that list. A UK business that wants Claude inside Copilot needs an AI Administrator or a Global Administrator to switch it on in the Microsoft 365 admin centre and choose which people or groups get it; an Australian business that does not want it switches it off in the same place.

Claude inside Microsoft Copilot, by default

Microsoft 365 admin centre, Copilot settings: AI providers operating as Microsoft subprocessors, Anthropic
  • AustraliaCommercial cloudOn by defaultAn administrator can turn it off
  • New ZealandCommercial cloudOn by defaultNot on Microsoft’s list of exceptions
  • United StatesCommercial cloudOn by defaultFor most commercial customers
  • United KingdomLike the EU and EFTAOff by defaultAn administrator opts in, for chosen people

Where people can pick Claude

  • ResearcherThe person selects Claude
  • Edit with CopilotIn the Microsoft 365 apps
  • Copilot StudioThe person building an agent chooses its model

Copilot shows an indicator when a Claude model is in use

Our drawing of the defaults on Microsoft’s page “Anthropic models in Microsoft Online Services”, last updated on 19 September 2026. It is not the admin centre’s interface.

Two further details sit on the same page. Anthropic models in Copilot are excluded from Microsoft’s EU Data Boundary and, where they apply, from its in-country processing commitments. And some newer Anthropic models come with Anthropic’s own commercial terms and data retention: they stay off everywhere until an administrator opts in to them separately, and Anthropic, rather than Microsoft, then keeps most inputs and outputs for up to 30 days.

Who switches each one on

In both cases your Microsoft administrator is part of the decision, and the two products call on different people at different steps.

Who switches each one on

Each step, and the person who takes it

Microsoft Copilot

  1. 1
    Copilot ChatComes with eligible Microsoft 365 licences
  2. 2
    A Copilot licence for each person who needs itMicrosoft 365 administrator, in the admin centre
  3. 3
    Claude models inside CopilotAI Administrator or Global Administrator
  4. 4
    The agents people may useMicrosoft 365 administrator, in Integrated apps

Claude

  1. 1
    The Microsoft 365 connector, for the organisationClaude organisation owner, on Team and Enterprise
  2. 2
    One consent for your whole tenantMicrosoft Entra Global Administrator
  3. 3
    Write tools, if Claude should send or fileEntra administrator approves, then the Claude owner enables
  4. 4
    The add-ins for Excel, Word, PowerPoint and OutlookMicrosoft 365 administrator; Outlook also needs a Graph consent

Our drawing of the steps in Microsoft’s and Anthropic’s documentation, read on 25 October 2026.

Each side also lets you limit who gets what. Copilot licences are assigned person by person, and access to Anthropic models inside Copilot can be limited to chosen users or groups. For the Claude connector, an Entra administrator can require assignment, so that only named people or groups can sign in to it, and can withdraw individual permissions from the connector as a whole, such as reading Teams chats.

Where your data goes

Copilot stores your prompts and its responses with your other Microsoft 365 content, under the same contractual commitments. Administrators can search them and set how long they are kept through Microsoft Purview, and each person can delete their own Copilot history. For customers outside the European Union, Microsoft says queries may be processed in the United States, in the EU or in other regions.

Anthropic does not use the inputs or outputs of its commercial products, which include Claude for Work, its Team and Enterprise plans, to train its models by default. The consumer plans, Free, Pro and Max, are covered by a separate policy, so a team on business accounts and a team on personal accounts do not work under the same terms. For the Claude add-ins, Anthropic deletes inputs and outputs from its servers within 30 days, keeps the chat history in the browser rather than on its servers, and does not apply the custom retention settings an organisation may have set in Claude. The Outlook add-in reads your mailbox from your browser, the Microsoft access token stays there, and the mail Claude reads is sent to Anthropic as part of each request.

The questions that decide it

No verdict fits every business. These are the questions we would ask you before recommending one, the other or both.

  • Where does your team spend its day: inside Outlook, Word, Excel and Teams, or in a separate AI app where it keeps its projects and instructions?
  • Do your SharePoint sites, shared folders and shared mailboxes reflect who should see what? Both assistants will show each person what they can already open.
  • Who is your Microsoft administrator, and can they give the consents each option needs?
  • Which tasks only need the AI to read and draft, and which need it to send, file or change something?
  • In the UK, has your administrator decided whether Claude models may run inside Copilot? In Australia and New Zealand, do you know they already can?
  • Are your people on business plans? Anthropic’s commercial terms cover its Team and Enterprise plans, and its consumer terms cover Pro and Max.

Where we stand, and how we start

We work in Claude and set it up for the businesses we work with. When a team runs on Microsoft 365, we start from its tenant: the Microsoft 365 connector and the add-ins, the permissions each person has, the consents your administrator gives and the points where a person approves before anything is sent. When a team already works in another AI environment, Copilot included, we start from what it has. If you want to talk through which one fits your business, the first step with us is a free AI pre-audit: 45 minutes on a call about how you and your team use AI today and which work to hand over first. You leave with a clear next step, with or without us.

How we help, who we are, how we handle security, and our earlier comparisons: ChatGPT or Claude for a business and which AI agent a business should use. Also today: Claude Duet, Anthropic’s agent that asks before it acts.

Sources

Every fact on this page comes from Microsoft’s and Anthropic’s own documentation, read on the day of publication. Both products change often and details may change after publication.

  1. Microsoft Learn: Overview of Microsoft Copilot Chat, updated 11 September 2026
  2. Microsoft Learn: Data, Privacy, and Security for Microsoft Copilot
  3. Microsoft Learn: Licence options for Microsoft Copilot, updated 20 October 2026
  4. Microsoft Learn: Anthropic models in Microsoft Online Services, updated 19 September 2026
  5. Claude help centre: Set up the Microsoft 365 connector
  6. Claude docs: Use Claude for Excel
  7. Claude docs: Use Claude for Word
  8. Claude docs: Use Claude for PowerPoint
  9. Claude docs: Use Claude for Outlook
  10. Claude docs: Work across M365 apps
  11. Anthropic privacy centre: Is my data used for model training?

Bring your questions.

Free AI pre-audit