Copilot, ChatGPT, Gemini or Claude: how to choose an assistant for a team

The default: the assistant inside the suite you already have

If your organisation runs on Microsoft 365, the default is Microsoft 365 Copilot. If it runs on Google Workspace, the default is Gemini. Not because either is the strongest model on any given day, but because they sit inside Outlook or Gmail, Word or Docs, Teams or Meet, with your permissions already applied and your documents already in reach. Microsoft states that Copilot only surfaces content a user already has permission to see and honours sensitivity labels; Google states the same for Gemini, and both say business data is not used to train their models by default. The links are in the sources.

The practical effect is that the tool meets people where the work already is. Nobody has to copy a document out of one system and paste it into another, which is the step where both effort and data control are lost. For a team of twelve doing ordinary knowledge work, this is the difference between a habit and a novelty.

Five questions that settle it

1. 1

   Where do the documents live?

   If ninety per cent of what the team reads and writes is in SharePoint or Drive, the suite assistant wins on reach alone. If the material arrives as files from outside, PDFs, contracts, data rooms, a standalone tool that takes attachments and long documents well (Claude and ChatGPT both do) may fit the actual work better.

2. 2

   What do the contracts and the regulator say?

   Read your client engagement letters, framework agreements and any regulator guidance before you read a vendor comparison. Some will name approved tools or require disclosure. All four vendors publish business-tier commitments on training and data handling; your DPO should read the actual pages, linked below, rather than a summary.

3. 3

   Long documents or short exchanges?

   Teams that live in email and meetings get the most from an assistant inside those applications. Teams that spend their days inside forty-page documents, due diligence, policy drafting, research, benefit from tools built around uploading and interrogating long material, and from features such as Claude Projects or ChatGPT custom workspaces that hold shared context.

4. 4

   Who administers it?

   Suite assistants are administered where your IT team already works, with the same identities, audit logs and retention controls. A standalone tool means another admin console, another SSO setup and another data processing agreement. That is manageable, and it is a cost; count it.

5. 5

   What will people actually open?

   Run a two-week trial with one team and count real use, not licences. The tool people open without being reminded is the one to standardise on. Enthusiasm in the demo predicts very little.

The four, in one paragraph each

• Microsoft 365 Copilot. Lives in Outlook, Teams, Word, Excel and PowerPoint. Strongest where the team’s work and data are already in Microsoft 365; weakest when the material is outside it. Needs some tenant hygiene first, because it will surface anything a person can already open. Microsoft’s privacy documentation covers training, permissions and the EU Data Boundary.
• Gemini for Google Workspace. The equivalent inside Gmail, Docs, Sheets, Slides and Meet. Same logic: strongest where the work is already in Workspace. Google’s privacy hub states that prompts are customer data, are not used to train models outside your domain without permission, and respect existing file permissions.
• ChatGPT Business and Enterprise. The most widely recognised standalone assistant, with shared workspaces, custom GPTs for repeatable tasks and strong handling of uploaded files. OpenAI’s enterprise privacy page states that business data is not used for training by default and that the organisation owns inputs and outputs. The consumer tier has different terms, which is why the account rule in your policy matters.
• Claude for Work. Strong on long documents, careful writing and following detailed instructions, with Projects for shared context across a team. Anthropic’s privacy centre states that inputs and outputs from commercial products are not used to train models by default, with an opt-in exception for feedback that an organisation owner can switch off.

Notice what is not in that list: a verdict on which is smartest. The models leapfrog each other every few months, and any ranking written in September is wrong by Christmas. The integration, contract and administration questions do not move nearly as fast, and they are the ones that decide whether a tool is still in use a year on.

Running two is fine. Running four is not.

Plenty of organisations end up with the suite assistant for everyday work and one standalone tool for a team with heavier document needs. That is a sensible shape, provided the one-page policy says which is for what, and the prompt library records which tool each entry was tested in. Four tools, chosen team by team on personal preference, is where the data control and the training budget both leak away. Pick, standardise, revisit in a year.

We are choosing an AI assistant for a team of [number] in [function] at a [type of organisation]. Our documents mostly live in [Microsoft 365 / Google Workspace / other]. The team’s week is mostly [email and meetings / long documents / data / client deliverables]. Constraints we know about: [contract clauses, regulator guidance, existing licences, IT preferences].

Write a one-page recommendation for our leadership team with these headings: Default choice and why; When we would add a second tool; Contract and data questions to resolve first (as a checklist); Two-week trial plan with the three numbers we will count; Decision date. Be direct. Do not compare model benchmarks. Where a point depends on something I have not told you, write [CHECK].