ChatGPT, Copilot or Gemini: how to choose an AI tool for your company

“Which AI tool is the best?” is a question I hear on almost every first call. The honest answer is: it depends on where the company works, what data it has and what the tool is for. Models change every few months, so a ranking from an article six months ago says little. The criteria for choosing change much more slowly - and they are what this article is about.

I do not compare prices or detailed features here, because those change fastest. Always check the current terms on the vendor’s website on the day you decide.

Tasks first, tool second

Before you compare tools, write down two or three tasks they should help with and the materials they will work on. “Summarising sales calls from notes” has completely different requirements from “answering staff questions based on policies stored in SharePoint”. I write more about this order in the article on where to start with AI in your company.

Where the company already works

This criterion most often decides the choice before anyone starts comparing answers.

• Microsoft 365. Microsoft Copilot works in Word, Excel, Outlook and Teams and can draw on documents, emails and meetings. Microsoft notes that Copilot only surfaces data the user has at least view permission for. In practice this means one thing: tidy up permissions before you switch Copilot on. If an employee has access to a folder they should not see, the tool will “see” it too and may use it in an answer.
• Google Workspace. Gemini works in Gmail, Docs, Sheets and Drive. The same rule applies: the tool uses whatever the user has access to, so orderly sharing is a precondition.
• A separate tool, for example ChatGPT in a business plan. It works well when the work is mostly writing and analysing text, when the company does not use either suite, or when it needs features the suite lacks. There are other tools too, such as Claude or Le Chat - the criteria in this article apply to them in the same way.

Data: a business version is not the same as a personal one

This is the most important difference, and companies usually learn about it too late.

Vendors state that in their business versions customer content is not used to train their models:

• OpenAI - for ChatGPT Business, ChatGPT Enterprise and the API platform, business data is not used to train models by default,
• Microsoft - prompts, responses and data accessed through Microsoft Graph are not used to train the models Copilot runs on,
• Google - in Google Workspace, content is not used to train models outside the customer’s domain without permission.

Personal versions have different rules. An employee who pastes client data into a personal account moves it outside the company’s control - no matter how good the tool itself is.

Before choosing, check in the documentation and the contract:

• whether the vendor signs a data processing agreement,
• where data is processed and what the exceptions are - Microsoft, for example, describes which models fall within the EU Data Boundary and which do not,
• who in the company has admin access and whether you can switch off features you do not want to use,
• how long conversation history is kept and whether it can be deleted.

What the tool has to work with

Is it enough for an employee to paste text into a chat window? Often yes, to begin with - and it is a good way to test whether a use case makes sense without an integration project. If the tool is supposed to use a CRM, a knowledge base or a ticketing system, check whether that connection exists, who will set it up and what data will flow through it.

Cost is more than the licence

The comparison should include:

• licences for the people who will actually use the tool - not for the whole company “just in case”,
• time spent tidying up permissions and documents,
• time spent preparing instructions and training the team,
• time spent checking answers, which remains whichever tool you choose.

I describe how to count this honestly in the article on measuring the effect of an AI implementation.

Test it on your own tasks

Comparing answers to sample questions from the internet says little about your company. A short test is better:

1. Choose three to five typical tasks and prepare the same materials for them.
2. Give them to two tools and to the same people.
3. Compare the results: how much had to be corrected, how long checking took, whether the answer stayed true to the materials.
4. Write down your observations before the decision - after a week people mostly remember the last impression.

If you use real company data in the test, do it in business versions, not on personal accounts.

One tool and shared rules

A common scenario: marketing uses one tool, sales another, and someone in accounting a third on a personal account. Each has different data rules, nobody knows what goes where, and good prompts do not travel between teams. One tool with clear rules usually delivers more than three “best in class” ones.

A short cheat sheet

• You work in Microsoft 365 and documents are in SharePoint or OneDrive - start with Microsoft Copilot, but only after tidying up permissions.
• You work in Google Workspace - start with Gemini, with the same condition.
• You do not use either suite, or the work is mostly text and analysis - consider a separate tool in a business plan.
• In every case - a business version, a contract, rules for what goes into the tool and for checking answers.

In AI implementation I match the tool to the company’s tasks and data, not the tasks to whichever tool happens to be fashionable.

Sources

• Microsoft Learn: Data, Privacy, and Security for Microsoft Copilot
• OpenAI: Enterprise privacy
• Google Workspace: Generative AI in Google Workspace Privacy Hub

As of 13 September 2026.