Every business leader who has looked at AI for more than five minutes runs into the same question. Microsoft is pushing Copilot. Google is pushing Gemini. Everybody talks about ChatGPT. They appear to do roughly the same job. So which one do you actually buy?
The honest answer is that the differences between the underlying models matter far less than most coverage suggests, and they change every few months anyway. What does not change is where each assistant lives and what it can see. That is the decision you are really making.
The short version
- Pick the assistant that sits inside the software your team already opens every morning.
- Microsoft 365 business? Start with Copilot. Google Workspace? Start with Gemini.
- Keep a standalone ChatGPT or Claude account regardless. It is where the harder thinking work happens.
- Before you switch anything on across a team, tidy your file permissions and write a one-page policy.
The three of them, honestly described
Microsoft Copilot is an assistant embedded in Word, Excel, Outlook, Teams and SharePoint. Its advantage is proximity. It can summarise the meeting you just left, draft a reply in the mailbox you are already looking at, and answer questions about documents sitting in your own SharePoint. Its weakness is that it is only as good as the tidiness of the estate it is reading. A well-organised Microsoft 365 tenant gets a genuinely useful colleague. A chaotic one gets confident answers built on a five-year-old draft nobody archived.
Google Gemini does the same trick inside Google Workspace: Gmail, Docs, Sheets, Meet and Drive. If your business runs on Google, this is the equivalent choice, and the reasoning is identical. It is also the strongest of the three at working with very long documents in one go, which matters if you deal in long contracts or reports.
ChatGPT is the one most people met first, and it is a different shape. It does not live inside your files. You go to it, paste in what you want to work on, and have a conversation. That sounds like a disadvantage and often is not: because it is a blank room rather than a busy office, it is better at open-ended work. Drafting a position from scratch, arguing a decision through, working out what you actually think before you write anything down.
Where each one earns its keep
| If you need to | Best fit | Why |
|---|---|---|
| Summarise a Teams or Meet call you missed | Copilot or Gemini | It has the transcript already; nothing to paste |
| Draft a reply in a long email thread | Copilot or Gemini | It can read the whole thread in place |
| Find the answer buried in your own files | Copilot or Gemini | Only these can search your document store |
| Think a hard decision through | ChatGPT or Claude | Better at back-and-forth without a document in the way |
| Write something long from nothing | ChatGPT or Claude | Stronger drafting, easier to steer over many turns |
| Analyse a spreadsheet of sales data | Any of the three | All three handle uploads; use whichever you are quickest in |
| Build a repeatable process the team follows | ChatGPT or Claude | Saved instructions and custom setups are more mature |
The question that actually decides it
Not "which model is cleverest". Ask instead: where does the work already live?
If your team spends its day in Outlook and Excel, an assistant that requires them to open a browser tab, copy something out, paste it in, wait, and paste the result back is an assistant they will use twice and forget. The friction is small but it happens forty times a day, and small friction repeated forty times a day is how good tools die quietly.
That is the whole argument for Copilot and Gemini, and it is a strong one. Not that they are better, but that they are already open.
The counter-argument is real too. The embedded assistants are tuned to be safe and brief. Ask one to write something with a point of view and you often get something polite and forgettable. That is why the recommendation below is not "pick one".
What we would actually do
For a business of five to fifty people, in this order:
- Match the office suite. Microsoft house, start with Copilot. Google house, start with Gemini. Do not fight your own stack.
- Add one standalone account of ChatGPT or Claude, at least for whoever does the most writing and thinking. It costs very little and it is where the interesting work happens.
- Give it three months and one job. Pick a single recurring task, meeting notes is the usual first choice, and do that one thing properly before adding a second.
- Write down which tool is for what. One page. Otherwise everybody defaults to whichever they saw on the news.
What we would not do is run a six-month evaluation. The tools change faster than the evaluation finishes, and the learning happens in use, not in comparison.
Two things to sort out before you roll anything out
Permissions. This one catches people. Copilot and Gemini respect whatever access rights already exist, which is exactly right, and also means they surface every filing sin you have committed since 2019. The salary spreadsheet in a shared folder was invisible because nobody browsed there. It is not invisible to an assistant that reads everything the user is allowed to read. Audit your shared drives first.
A written policy. Not a legal document. One page that says which tools are approved, what must never be pasted into them, and who to ask when someone is unsure. We have a template you can copy in Write an AI policy for your small business in one page, and the related question of what is safe to paste in the first place is covered in Is it safe to put company data into ChatGPT?
A fair test you can run this week
If you want evidence rather than opinion, give each candidate the same real job and compare. Not a puzzle, a real job, with your actual context in it.
I run a [type of business] in [town], with [number] staff. Here is a real enquiry we received: [paste it]. Draft a reply in our voice: direct, warm, no jargon, under 150 words. Then list the three things you would need to know from me to make that reply better.
The last sentence is the tell. A useful assistant asks sensible questions back. A weak one bluffs. That difference will teach you more in ten minutes than a week of reading comparison articles, this one included.
Where to start if none of this is familiar
If the phrase "office suite" is doing a lot of work in your head right now, start one step back. What is a large language model? explains what these tools are actually doing, in plain English, and why they get things confidently wrong.
And if you would rather watch someone use all three on a real business problem than read about them, that is what the free AI Breakfast Club webinar is for. It runs online every other Friday morning and there is nothing to buy.