ChatGPT, Copilot or Claude for your business: how to choose

ChatGPT is the broadest of the three: writing, creating images, voice, data analysis and your own assistants (custom GPTs) in one tool. For businesses there's ChatGPT Business, and for larger organizations ChatGPT Enterprise.

ChatGPT's strength is that almost everyone already knows it. That saves on training: you don't have to explain to anyone where the button is. With custom GPTs you build an assistant without code that knows your quote template, tone of voice and product information. Through apps (until the end of 2025 these were called connectors), ChatGPT pulls information from SharePoint, Outlook, Teams and Google Drive, among others. And since July 2025 there's ChatGPT agent, a mode that visits websites, fills in forms and does research by itself (OpenAI, 2025).

Plans

• ChatGPT Business (until August 2025 this was called Team): $ 20 per user per month with annual billing or $ 25 per month, from two users.
• ChatGPT Enterprise: price on request, with extra administration features and the option to store data in Europe.

Prices excluding VAT, according to OpenAI on 26 October 2026. Check the current price with the provider.

By default, OpenAI doesn't use data from Business, Enterprise and the API to train models (OpenAI, 2026). The free version and personal plans are different: there OpenAI can use conversations for training, unless the user turns that off.

Watch out

OpenAI offers storage of your conversations in Europe for new Enterprise and Edu workspaces (OpenAI, 2025). If you work with sensitive personal data, check where the data of your plan is stored when you choose. More in using AI safely.

Copilot is strong at everything that happens in Microsoft 365: summarizing an email in Outlook, catching up on a meeting in Teams, making a first draft in Word or digging into a table in Excel. It works with your own files and respects the permissions you've already set.

Microsoft has three versions, and they often get mixed up:

• Microsoft 365 Copilot Chat comes at no extra cost with a business Microsoft 365 subscription. It's a chat with protection for your company data, but without deep access to all your files.
• Microsoft 365 Copilot Business is the version for SMEs, for organizations with a Microsoft 365 Business subscription (up to 300 users). $ 21 per user per month with annual billing, with a promotional price of $ 18 until 31 December 2026 (Microsoft, 2026).
• Microsoft 365 Copilot for larger organizations: $ 30 per user per month with annual billing, on top of a Microsoft 365 license (Microsoft, 2026).

Prices excluding VAT, checked on 26 October 2026. Check the current price with the provider.

Microsoft doesn't use your data to train the underlying models and, for European customers, keeps it within the EU Data Boundary.

Watch out

Copilot finds everything an employee has permissions for. Including that salary spreadsheet that has accidentally been open to everyone for years. So clean up your SharePoint permissions before you roll Copilot out widely.

And look at two settings. Since 2026, Copilot can process requests outside the EU at peak times (flex routing). For environments created after 25 March 2026 this is on by default; for older environments, check the setting. Your administrator can turn it off (Microsoft, 2026). In addition, you can also use models from Anthropic in Copilot, but according to Microsoft those fall outside the EU Data Boundary.

Claude, from the American company Anthropic, is strong at writing, reading and analyzing long documents and independent work such as programming. Many people find that Claude's texts sound more natural. In Projects you give Claude fixed instructions and documents, and with skills you teach it a fixed way of working.

For developers there's Claude Code, which writes, tests and adjusts code independently. That's what we build with too. And with the Microsoft 365 connector, Claude reads in SharePoint, OneDrive, Outlook and Teams, on all plans (Anthropic, 2026).

Plans

• Claude Team: $ 20 per user per month with annual billing or $ 25 per month for a standard seat, $ 100 or $ 125 for a premium seat with five times as much usage. For 2 to 150 users (Anthropic, 2026).
• Claude Enterprise: $ 20 per user per month plus usage at API rates, from 20 users and annual billing only (Anthropic, 2026).

Prices excluding VAT, checked on 26 October 2026. Check the current price with the provider.

On Team and Enterprise, Anthropic doesn't train on your data by default. On the personal plans (Free, Pro and Max) it does, since September 2025, unless you turn that off.

Watch out

Anthropic itself runs the model in the United States or globally, and stores data in the United States. Anthropic doesn't yet have a European option of its own (Anthropic, October 2026). If you want Claude with processing in the EU, you use it through AWS or Google Cloud in a European region. That's something for a built application, not for the chat app.

With all three business versions, your data isn't used to train models by default, and you can sign a data processing agreement. The biggest difference is in where your data is stored and processed. And the biggest risk isn't in the tool, but in employees who use the free version.

In 2024, the Dutch Data Protection Authority (Autoriteit Persoonsgegevens, AP) received several reports of data breaches because employees pasted personal data into an AI chatbot, such as patients' medical data at a GP practice (AP, August 2024). That often happens with a personal account, when there's no business alternative. Banning AI therefore backfires: people then carry on in secret. That's called shadow AI.

Arrange at least this:

• One business tool for everyone who works with company data, managed by your IT department.
• A data processing agreement with the provider. Under the GDPR you need one as soon as personal data goes in.
• Agreements on what can go in. Customer names, citizen service numbers (BSN), medical data and passwords don't belong in any chat.
• Knowing where your data is stored. Within the EU, in the US with the right safeguards, or in your own environment.

The risks and what to arrange for each are covered in using AI safely.

If you work entirely in Microsoft 365, start with Copilot Chat and give heavy users a Copilot license. If you work with Google Workspace or lots of separate tools, look at ChatGPT Business or Claude Team. And if you write a lot of long pieces, take a serious look at Claude.

You work in Microsoft 365 all day

Start with Copilot Chat, which already comes with your subscription. Then give a Copilot license to the people who work a lot in Outlook, Teams and Word, and after three months look at who really uses it. Clean up your SharePoint permissions beforehand. See also Copilot training.

You work with Google Workspace or lots of separate tools

ChatGPT Business or Claude Team are then the more obvious choice. ChatGPT if you want a broad tool that everyone already knows, Claude if there's a lot of writing and analysis. See ChatGPT and Claude training.

You write a lot: advisory, legal, reports

Claude is often the favorite here. It reads long files in one go and writes texts that sound less like AI. Have the people who do the work test it first.

You have developers in-house

Claude Code and OpenAI's coding tools are the serious choices there. Note: a prototype built in an afternoon isn't a secure application yet. Read from vibe coding to production.

You work with sensitive data, such as in healthcare

Then you want processing in the EU and control over what happens. A chat app is rarely enough there. You then build the application in your own environment, for example with Azure OpenAI or a model in a European cloud region, with a human who checks.

And the combination?

You see it more and more: Copilot for everyone in Microsoft 365, and ChatGPT or Claude for the group that works with it heavily every day. That's fine, as long as there's one set of ground rules.

The model at the top of a general ranking isn't automatically the best for your work. That's why we test models ourselves on ordinary office work and publish the results on our benchmark page.

Three lessons from those benchmarks:

• The most expensive model isn't always the best. For tasks you run thousands of times a month, a model that's almost as good but costs a fraction is often the better choice.
• General scores say little about automation. In our automation benchmark, the model at the top of the general rankings isn't automatically at the top. Which model wins depends on the work you give it.
• Small differences are often noise. If two models are close together, price, speed and where your data is stored tip the balance.

For your team's chat app, the model matters less than you think: the difference is in how well your people prompt. For a workflow that reads a thousand invoices every day, it matters a lot. There we choose the model per task, and switch when a better or cheaper model comes out.

Test the tools on five real tasks from your own work, check the data agreements, let a small group trial it for a month and train everyone before you roll it out widely. Then you choose on experience instead of on a demo.

1. 01Pick five real tasks. Answering a customer email, drafting a quote, summarizing a report, analyzing an Excel export. No tricks, just this week's work.
2. 02Check the agreements. Business plan, data processing agreement, no training on your data, and where the data is stored. Have your data protection officer or lawyer look at it too.
3. 03Run a one-month trial with ten people. Preferably a mix of enthusiasts and skeptics. After a month, ask: what do you use, for what, and how much time does it save?
4. 04Train before you roll out. A license without training is an expensive notepad. With a workshop on your own work, you bring the team to the same level in one half-day session.

And remember: choosing a chat tool is phase one and two of AI adoption. Prompting and assistants deliver personal gains. The big gains are in phases three and four, where workflows and agents take over the work itself. There, the model behind the scenes is interchangeable, and that's where we help in the AI Partnership.