ChatGPT, Copilot, or Claude: choosing an AI assistant for a small team

Most small businesses asking which AI assistant to buy have already half-answered the question. Someone on the team uses ChatGPT at home. The office runs on Microsoft 365, so Copilot keeps appearing in the toolbar. A developer or a writer swears by Claude. The real decision is which one the business pays for, sets up properly, and tells people to use.

All three are capable. For the drafting, summarizing, and question-answering that make up most small-business use, the gap between them is smaller than the marketing suggests, and it shifts every few months as new models ship. So instead of ranking them, this post works through the questions that actually separate them for a team of five to fifty people.

Start with where your work already lives

An assistant is most useful when it can see the material you are working on without someone copying and pasting it in.

If your business runs on Microsoft 365, Copilot’s advantage is grounding. Inside Outlook, Word, Teams, and Excel, it can work from your mail, your files, and your meeting transcripts directly. “Summarize what the client agreed to in last week’s thread” becomes a one-line request instead of a scavenger hunt. ChatGPT and Claude can reach into Microsoft 365 and Google Drive through connectors, depending on the plan, but the integration is shallower and it is one more set of permissions to manage.

That grounding cuts both ways. Copilot answers from everything the user can open, which in most tenants is far more than anyone intended. Before you turn it on, read our piece on fixing oversharing before a Copilot rollout. The cleanup is permissions work, and it pays off whether you buy Copilot or not.

If you are on Google Workspace, the equivalent candidate is Gemini, which Google builds into Gmail, Docs, and Drive on its business plans. Give it a trial on the same terms as the others, for the same reason: the assistant already sitting next to your files has a head start.

If your work lives mostly in a line-of-business system, such as a practice management tool or a trade-specific job tracker, none of the three will see into it by default. That pushes the decision back toward what the work is.

Then look at what the work is

Long documents

If your team reads and drafts contracts, proposals, policies, tenders, or reports, take one of your own long documents and ask each assistant hard, specific questions about it. Claude has a reputation for handling long material carefully and for prose that needs less editing. ChatGPT is a strong all-rounder with a wide set of built-in tools. Your own test is worth more than either reputation.

Coding

If anyone on the team writes code, even scripts and spreadsheet macros, the coding tools matter more than the chat window. OpenAI and Anthropic both ship coding tools that work inside a developer’s editor or terminal. GitHub Copilot is a separate Microsoft product from Microsoft 365 Copilot, despite the shared name. Let the person who will use it every day make this call.

Data

Spreadsheet work is where in-app grounding helps most, since Copilot in Excel works on the workbook in front of you. ChatGPT and Claude can both analyse an uploaded file by writing and running code against it, which is often easier to audit because you can see each step. Either way, check the numbers before they go into a decision. A model can misread a column header as easily as a new hire can.

For everyday email and short drafting, any of the three will do. Don’t let it drive the choice.

Use business plans, not personal accounts

This matters more than which brand you pick.

Each of these products comes in consumer and business versions, and they are not the same contract. Consumer plans differ on whether your conversations can be used to train future models, how long data is kept, and whether the business can see or control the account at all. A personal account has no admin console, no single sign-on, and no way for you to take it back when someone leaves.

The business, team, and enterprise plans from OpenAI, Anthropic, and Microsoft each state that customer data is not used to train their models by default. That is the baseline you want. Terms do change, and the details live in the data-processing terms, not on the pricing page, so check the current terms for the specific plan before you sign up.

If client confidentiality or privacy law applies to your work, staff using personal accounts on client files is a bigger risk than any difference between the tools. Our guide to writing a one-page AI policy covers how to draw that line.

Data residency for Canadian businesses

Some Canadian businesses need data to stay in Canada, or have promised clients it will. Microsoft offers Canadian datacentres for core Microsoft 365 data at rest, but where AI processing happens is a separate question, and the answer varies by feature and has changed over time. Several AI providers have added regional data-residency options, mostly on enterprise tiers and not on the plans small teams usually buy.

If residency is a hard requirement, get the provider’s current answer in writing for the exact plan you are considering. Our guide to data residency in Canada explains when it actually matters and how to verify it instead of assuming it.

Admin controls you will use in the first year

Small teams rarely need an elaborate admin setup, but three controls come up quickly.

Single sign-on lets staff log in with their existing Microsoft or Google work account, so multi-factor authentication stays consistent and there is one less password to lose. Check which tier includes it. On some products it sits higher up the price list than you would expect.

Removing a leaver should take one step. When someone leaves, their access should end the same day, and their conversation history should be kept or deleted by the business’s choice, not theirs. During the pilot, add and remove a test user to see how this actually works.

Retention settings decide how long chats and uploaded files are kept. If you have record-keeping obligations, or the opposite need to keep sensitive material short-lived, confirm what the plan lets you configure.

How the cost works

All three charge per seat per month, and annual billing is usually cheaper than monthly. Microsoft 365 Copilot is a per-user add-on to an eligible Microsoft 365 business plan, and many plans already include a no-cost Copilot Chat experience that may be enough for lighter users. ChatGPT and Claude sell team plans, often with a minimum seat count, plus higher tiers for heavy use. Prices and plan names have changed several times in the past two years, so price it on the day you buy.

The bigger waste is seats nobody opens. Look at usage around the 60-day mark and reassign or cancel the idle ones.

A pattern that works for small teams

Many small teams end up in the same place: one primary assistant for everyone, chosen mostly for where the work lives, plus a second tool for a handful of power users who do heavy writing, analysis, or coding. That is a sensible outcome, and it beats forcing everyone onto whatever one enthusiast prefers.

Get there with a pilot. Pick three to five people from different roles, including one sceptic. Give them business accounts on your leading candidate for 30 days and have them use it on real tasks from their actual week. At the end, ask for specific examples of what got faster, what got worse, and what they quietly stopped using. Then buy seats for the people whose examples were convincing, and expand from there.

Training matters more than the tool

In our experience, the teams that get real value from an assistant have one thing in common. Someone sat down with each role and showed them five to ten workflows specific to their job. The office manager learns to turn a voicemail transcript into a booked appointment. The bookkeeper learns to ask for a plain-language explanation of an odd ledger entry. The project lead drafts a change order from site notes. Generic “prompt engineering” sessions do much less.

The policy comes first, though. People need to know which data can go into the tool before you show them how to put it there.

If you want help choosing, setting up, and training a team on Copilot, ChatGPT, or Claude, that is what our AI adoption work covers. Write us a short brief about your team and the tools you use today, and we will reply in writing within one business day.