Start with where your work already lives
The strongest predictor of success is not benchmark scores. It is how close the assistant sits to the documents, email and meetings your people use every day. A tool that needs files copied into it will be used less than one that can already see them.
If your business runs on Microsoft 365, Copilot has a structural advantage. Copilot Chat is included with eligible Microsoft 365 licences and runs inside the Microsoft 365 service boundary with enterprise data protection7. The paid Microsoft 365 Copilot licence adds grounding across your mailboxes, files and meetings; without it, Copilot Chat cannot reach your organisation's shared data through Microsoft Graph7,8.
That advantage is narrower than it once was. Claude's Microsoft 365 connector can search and analyse SharePoint, OneDrive, Outlook and Teams content on Team and Enterprise plans, once a Microsoft Entra Global Administrator grants tenant consent4. Its enterprise search only returns results a person can already open in the source system15. ChatGPT Business also connects to SharePoint and other work tools, with read-only search that respects existing permissions10.
Where Claude is the stronger choice
Claude earns its place on reading and writing that needs care. Teams that work with long contracts, tenders, policy documents, technical specifications or board papers tend to prefer it, because it holds a large amount of material in one conversation and is good at quoting the passage an answer came from. On paid plans, including Team, the context window in chat is at least 200,000 tokens, and up to 1M tokens on current models, which is enough for several long reports at once18.
Projects let a team keep shared reference material and standing instructions in one place, so nobody re-explains the house style or the client background in every chat1. Artifacts turn a reply into a draft document, a table or a small working tool you can iterate on. For technical staff, Claude Code and Cowork are included in Team, so developers and analysts can hand over multi-step work from the same subscription1.
The trade-off is integration depth inside Office itself. Claude can read your Microsoft 365 content through the connector, and write tools such as sending email are available only when an administrator turns them on4. If what you want most is a button inside Word, Excel and Outlook that every staff member sees without training, Copilot is closer to that.
Where ChatGPT and Copilot are stronger
ChatGPT has the widest name recognition, so many staff arrive already familiar with it, and that lowers the training load. Its Business plan excludes workspace data from model training by default10. If your team built custom GPTs, plan the move now: OpenAI is retiring custom GPTs on 11 December 2026, moving their instructions into skills inside plugins, and custom actions do not transfer automatically11.
Copilot's strength is that it is already where Microsoft-centric staff spend the day. Microsoft 365 Copilot works inside Word, Excel, PowerPoint, Outlook and Teams, and it uses the permissions your tenant already has, which is both a strength and a risk if your SharePoint permissions are loose8. For a business whose meetings, email and files all sit in one tenant, that reach is hard to beat.
Data terms and data residency compared
On training, the three business offerings now look alike. Anthropic's Commercial Terms say it may not train models on customer content, and its plan page lists no model training on your content by default for Team and Enterprise2,3. OpenAI states that ChatGPT Business data is excluded from training by default10. Microsoft's enterprise data protection applies its existing privacy and compliance commitments to Copilot Chat prompts and responses7. The bigger privacy risk in all three cases is staff using free personal accounts instead.
Data location is where they differ. Microsoft has a New Zealand datacentre region, and its Advanced Data Residency or Multi-Geo add-ons can keep stored Microsoft 365 Copilot interaction content in New Zealand6. OpenAI offers storage at rest in a list of regions that includes Australia for eligible ChatGPT Enterprise, Edu and API customers, mainly for new workspaces9. Anthropic's own residency settings offer global or United States processing, with US storage, and no New Zealand or Australian option for the Claude apps5.
For most private businesses, overseas processing is lawful if the contract and controls are sound, and the Privacy Commissioner expects you to have thought it through before using generative AI with personal information14. If a client contract or a government customer insists on onshore storage, that requirement may settle the choice before features do.
- Onshore storage is a hard requirement: Microsoft 365 Copilot with the Advanced Data Residency or Multi-Geo add-ons is the clearest fit6.
- Australian storage is acceptable and you are buying at enterprise scale: ask OpenAI about ChatGPT Enterprise residency9.
- US processing is acceptable and long-document work matters most: Claude Team or Enterprise1,5.
When to pick which: worked NZ examples
These examples are generic, but they reflect patterns that come up repeatedly when teams compare the tools side by side.
- A small Nelson law practice that spends most of its week reading leases, agreements and disclosure bundles. Claude Team, with a Project per practice area and web search left off for client matters.
- A Hamilton manufacturer with fifty office staff, all in Microsoft 365, whose biggest pain is meeting notes and Excel reporting. Copilot first, because it already sits in Teams and Excel.
- A Tauranga marketing agency whose staff already use ChatGPT and who mainly need campaign drafting and image work. ChatGPT Business, with clear rules on client material.
- A Wellington consultancy writing long reports for government clients. Claude for drafting and review, but check each client's AI rules first, since agencies follow the public service GenAI guidance12.
- A software team in Christchurch. Claude Team for Claude Code, alongside whatever the rest of the business uses1.
Running two tools is normal, if it is governed
Many New Zealand businesses end up with Copilot for everyday Office work and Claude or ChatGPT for specialist teams. That is fine, provided one usage policy covers every tool and someone owns each subscription. MBIE's responsible AI guidance for businesses frames this as governance and accountability: name who is responsible, know which tools are in use, and match controls to the risk13.
The National Cyber Security Centre's guide for small businesses adds the security angle: decide what information may go into any AI tool, and treat AI output as something to check rather than something to trust16. One short policy, written once and applied to all three tools, is better than three policies nobody reads.
A two-week trial that settles the argument
Debates about which assistant is smarter rarely end. A structured trial on your own work usually does. Run it with a handful of people who do the work every day, not only the enthusiasts.
- Pick three real tasks that recur every week, such as summarising a long document, drafting a client reply and turning notes into a report.
- Use business plans for every tool in the trial, so no client data goes into consumer accounts3,10.
- Give each tool the same source files and the same instructions, then score the output for accuracy, editing time and tone.
- Ask your IT provider how each tool would connect to your existing files, and who must grant consent4,8.
- Write down the data residency answer you need, and check it against each vendor's documentation5,6,9.
- Decide at the end of the second week, and set a review date three months out.