If your company runs on Microsoft 365, start with Copilot Chat, already included in your licence, and pay for full Copilot only for people who live in Outlook, Teams and Excel. On Google Workspace, start with Gemini. ChatGPT Business or Claude earn their place when you want the strongest general assistant. A 30-day pilot decides, not a demo.
"Which AI is best?" is usually asked as if it were a benchmark contest. For a company, the model underneath matters less than three plain things: where your data already lives, which contract covers what people type into prompts, and whether anyone is still using the tool a month later.
All plan details below were read on the vendors' official pages on 26 September 2026. Names change often: in 2026 alone, Microsoft 365 Copilot became "Microsoft Copilot" and Microsoft 365 Copilot Chat became "Microsoft Copilot Chat" (Microsoft Learn).
What is the real difference between ChatGPT and Copilot for a business?
What the tool can reach. Microsoft Copilot, with a paid licence, works across your email, meetings, chats and files in Microsoft 365, using the permissions you already have. ChatGPT Business is an excellent general assistant that sees company documents only through what you upload or the connectors you enable.
Microsoft has two tiers, and this is where most confusion starts (Microsoft's overview):
- Copilot Chat is included with eligible Microsoft 365 licences, without a Copilot licence. It is primarily web-grounded: you can upload files, paste text and, in Outlook, it works across your inbox, calendar and meetings. It has no broad access to company data.
- Microsoft Copilot (the add-on licence, sold as "Copilot Business" for the Business plans) adds reasoning across Microsoft 365 through Microsoft Graph: email, meetings, chats, documents, SharePoint, plus the full experience in Word, Excel and PowerPoint and priority access.
On the other side, ChatGPT Business (formerly Team) and ChatGPT Enterprise offer a general assistant, custom GPTs and "company knowledge", which connects to sources such as SharePoint or Google Drive. Claude (Team and Enterprise plans) also integrates with Microsoft 365 and searches internal sources. Gemini is included in Google Workspace editions, inside Gmail, Docs, Drive and Meet, plus the Gemini app; which features you get depends on your edition.
Which should I choose, given my company's ecosystem?
My rule is simple: start from the ecosystem your people spend the day in, and only then compare models.
| Your situation | Start with | Add only if |
|---|---|---|
| Microsoft 365, people who live in Outlook, Teams, Excel | Copilot Chat (included) for everyone | the Microsoft Copilot licence for the group that summarises meetings, searches email threads and works in Excel daily |
| Google Workspace | Gemini, included in your edition (check which features) | a separate general assistant for people who write or analyse a lot |
| Mixed stack, or work mostly in an ERP, local files, in-house apps | ChatGPT Business or Claude Team, tested side by side on the same tasks | connectors to SharePoint/Drive, if your documents actually live there |
| People who write a lot, analyse, code, handle long documents | ChatGPT Business or Claude, regardless of ecosystem | nothing; here the general model matters more than integration |
| Exact answers from company documents, with citations, across systems outside M365 or Drive | none of the above on its own | an assistant built on your documents (see below) |
A hypothetical example: imagine a 25-person accounting firm on Microsoft 365. Everyone gets Copilot Chat on day one, because it is already paid for. The partners, who summarise meetings and dig through long email threads every day, get the full licence in the pilot. The two people who write reports and procedures test ChatGPT Business and Claude side by side on the same ten documents. Nobody buys 25 licences in month one.
Which data settings should I check first?
Which tool you pick matters less than whether people use it from a company account. All four major platforms state that, under their business terms, they do not train models on your company's data by default:
- Microsoft: Copilot and Copilot Chat are covered by the Data Protection Addendum, with Microsoft as processor; prompts, responses and data accessed through Microsoft Graph are not used to train foundation models, and the commitments include the EU Data Boundary. Two exceptions worth knowing: web queries sent to Bing are outside the EU Data Boundary, and the Anthropic models available inside Copilot are currently excluded from it (Microsoft Learn).
- OpenAI: data from ChatGPT Business, Enterprise and Edu is not used for training by default (OpenAI). Data residency in Europe is available on some plans, including Enterprise and Edu; if residency matters to you, check the eligibility list before you choose.
- Anthropic: Claude Team and Enterprise run under the Commercial Terms, with no training on your content by default; your company is the data controller (Anthropic Privacy Center).
- Google: in Workspace, your content is not human-reviewed or used to train generative models outside your domain without permission (Google Workspace).
The real risk sits elsewhere. On personal accounts, free or paid, whether your conversations train future models depends on a setting each person has to check. If you don't give people an approved tool, they will use an unapproved one, and customer data ends up in accounts the company does not control. That links straight to Article 4 of the AI Act: your AI policy should say which tool is approved and which data never goes into a prompt. I wrote up what Article 4 of the AI Act asks of SMEs.
Before the pilot, check:
- company accounts, with SSO where possible, not personal ones;
- which data may go into prompts (and which may not: health data, national ID numbers, trade secrets);
- in Microsoft 365, your SharePoint permissions: Copilot sees everything the user can see, so a folder accidentally "shared with everyone" suddenly becomes very easy to find;
- conversation retention and who holds admin access.
Is Microsoft 365 Copilot worth it for a small business?
It is worth it for people who spend the day in Outlook, Teams and Excel, with many meetings, long threads and scattered files. It is not yet worth it for someone who writes two emails a day. The full licence is per user, on top of Microsoft 365, so the right question is not "is it worth it for the company?" but "for which five people?".
The cost is not only the licence. It also depends on how tidy your SharePoint is (Copilot will find what it shouldn't), how much time you put into training, and whether people change their habits. Current prices are on Microsoft's Copilot pricing page; look at the Business option, for companies up to 300 users.
How do I pilot for 30 days, and what do I measure?
With a small group, real tasks and one main metric: time saved per task, checked for quality. Not "how much they like it".
- Pick 2–3 recurring tasks the group does weekly: a meeting summary, a reply to a request for quotation, a monthly report, a search through a contract.
- Measure the baseline in week one, without AI: how long the task takes, on a few real examples.
- Choose a group of 5–10 people who do these tasks often, and give them short training on their own tasks, not a generic course.
- For three weeks, log per task: the time with AI, and whether the output was used with light edits, heavily rewritten, or thrown away.
- At the end, decide on three numbers: average time saved per task, the share of outputs actually used, and how many people in the group still used the tool in the final week.
If you test two tools, give them the same tasks and the same documents. And if after a month half the group has stopped opening the tool, I think the model is rarely the problem; more likely the tasks you chose were not where people lose time. The wider version of this pilot logic is in the step-by-step guide to implementing AI.
When is neither ChatGPT nor Copilot enough?
When you need exact answers from your company's documents, with the source cited, across data that doesn't live in Microsoft 365 or Google Drive: contract archives, procedures, technical sheets, ERP data, scanned PDFs. General assistants search well through what you give them, but they don't guarantee they found the right document, the right version, or that they cite the exact clause. And when the job is to act across systems rather than answer, you are into AI agents.
That is where an assistant built on your own documents (RAG) begins, evaluated on real questions. I built ai-aflat.ro this way, across more than 220,000 legislative acts, updated daily, where every answer cites the law and links to the official source. For what such a project looks like technically, the Sapio team has written a guide on how to build a RAG chatbot on your company data.
So what?
Choosing between ChatGPT and Copilot is, in the end, not really about ChatGPT and Copilot. It's about what your people will do on Monday morning, with which data, from which account. I would bet that in many companies arguing over the tool, people already use three of them, from personal accounts. The question is whether the company makes the decision, or whether they already made it.
If you want your team to use the tool you picked well, the details are on the AI workshops for teams page; for the choice itself, see independent AI consulting.