Quick answer
Copilot, ChatGPT and Gemini are close enough in raw capability that comparing their writing is largely a waste of time. What separates them is where they sit: Copilot sees your Microsoft 365 documents and meetings, Gemini sees your Google Workspace, and ChatGPT sees nothing of yours but has the largest ecosystem and the most tooling. For a small business the decision is almost always made by which office suite you already pay for — and if that is neither, ChatGPT is the default.
Comparisons of these three usually run the same prompt through each and grade the prose. It is a satisfying format and it measures the thing that matters least, because the gap between them on ordinary writing tasks is small and shifts with every release.
The durable differences are structural: what data each can reach, where it appears in your day, and what it costs given what you already pay for. Those do not change with every model update.
| Microsoft Copilot | ChatGPT | Gemini | |
|---|---|---|---|
| Sees your business data | Microsoft 365: files, email, Teams | No, unless you paste or upload it | Google Workspace: Drive, Gmail, Docs |
| Lives in | Word, Excel, Outlook, Teams | Its own app, plus an API | Gmail, Docs, Sheets, Meet |
| Ecosystem and third-party tooling | Modest | Largest by a wide margin | Growing |
| Best single feature | Teams meeting recap | Breadth, and custom GPTs | Gmail and Docs integration |
| Weakest area | Excel on messy real spreadsheets | No native view of your work files | Less third-party tooling |
| Buy it when | You live in Microsoft 365 | You want the best general assistant | You live in Google Workspace |
My take
I would stop evaluating output quality entirely. In six months of releases the ranking on any writing benchmark will have changed twice, and it will not have changed which office suite your company uses. Pick on integration, learn one properly, and revisit in a year. The productivity difference between someone fluent in one assistant and someone sampling three is much larger than the difference between the assistants.
Teams meeting recap is the standout and the feature people miss most when it is removed. Summaries, decisions, action items, and the ability to ask what was said about something in a meeting you did not attend. Outlook thread catch-up is a close second, because email volume makes small savings compound.
Its real differentiator is answering questions grounded in your organisation’s own content — what the current policy says, which document covers a topic — without you knowing where anything lives. Nothing else on this list can do that for Microsoft estates.
Breadth. The widest range of capabilities in one place, the most third-party integrations, the most material written about how to use it well, and the largest pool of people who already know it. Custom GPTs let you package a repeated task for colleagues without writing code, which is genuinely useful and underused.
If you are building anything on top of a model, the API and tooling ecosystem is the most mature. For most small businesses with no strong office-suite tie, this is the default and a good one.
If your company runs on Google Workspace, the argument mirrors Copilot’s exactly: it is already where you work, it can see your Drive and Gmail, and the integration removes the copy-paste step that quietly limits how much anyone uses an assistant. Strong on long documents and on multimodal input.
All three offer a free tier and paid plans, and pricing changes often enough that specific figures age badly. What matters is the structure.
Watch out
Consumer and business tiers differ in how your data is handled, and the difference matters legally rather than cosmetically. If staff will paste client information, contracts or personal data into an assistant — and they will, whether or not you provide one — you need the business tier for the data processing terms. Providing a sanctioned tool is also the only realistic way to stop people using an unsanctioned one.
| If you… | Choose |
|---|---|
| Run on Microsoft 365 with lots of meetings | Copilot, for the roles that are meeting-heavy |
| Run on Google Workspace | Gemini |
| Use neither heavily | ChatGPT |
| Want to build tools or automate with an API | ChatGPT, for ecosystem maturity |
| Need it mainly for research with sources | None of these. Use an answer engine |
| Have developers | A separate coding tool. This comparison does not cover that |
This is not a capability contest and treating it as one leads people to switch repeatedly and get fluent in none. On ordinary business writing, drafting and summarising, all three are good enough that the difference will not show up in your results.
Pick the one that already sits where you work, because integration is what determines whether an assistant gets used forty times a day or four. Buy the business tier if staff will touch client data. Pilot before committing seats.
Then stop comparing and start learning it properly. Fluency with one beats sampling three, by a wide margin.
For Copilot’s packaging in detail, see Microsoft Copilot pricing explained. For how these models work and where they fail, what ChatGPT is and how it works, and for cited research specifically, what Perplexity AI is.
Microsoft’s AI assistant, embedded across Word, Excel, Outlook and Teams. Its distinguishing feature is access to your organisation’s Microsoft 365 content, so it can summarise meetings, catch you up on email threads, draft from your existing documents and answer questions about company material without you knowing where files live.
On ordinary writing and summarising they are close enough that the difference will not show in your results, and the ranking shifts with every release. What durably separates them is integration: Copilot sees Microsoft 365, Gemini sees Google Workspace, ChatGPT sees nothing of yours but has the largest ecosystem. Choose on which suite you already use.
It is the least useful comparison available. The ranking on any writing benchmark will change several times a year, while your office suite will not. Integration determines whether an assistant gets used forty times a day or four, and fluency with one tool beats sampling three by a considerable margin.
See your organisation’s own content. Summarising a Teams meeting you missed, catching up on an Outlook thread in context, answering what the current internal policy says, or finding which document covers a topic. ChatGPT has no native view of your work files, so anything of yours has to be pasted or uploaded manually.
For those without a strong Microsoft 365 or Google Workspace commitment, generally yes. Its business tiers are standalone rather than add-ons, making costs easier to reason about, it has the largest ecosystem and third-party tooling, and custom GPTs let you package repeated tasks for colleagues without writing code.
If staff will paste client information, contracts or personal data into it, yes — consumer and business tiers differ in data handling in ways that matter legally rather than cosmetically. Providing a sanctioned tool is also the only realistic way to stop people quietly using an unsanctioned one on the same material.
Remember that Copilot and Gemini business tiers are add-ons, so the real cost per head includes the underlying Microsoft 365 or Workspace licence. Adoption is uneven in all three products, so pilot with a subset of roles, measure actual usage through admin reporting, and expand from evidence rather than buying annually for everyone.
None of these three. All are weaker at cited research than a purpose-built answer engine that retrieves live pages and attaches numbered citations you can open. If your main need is factual research you can stand behind, buy a research tool alongside whichever general assistant fits your office suite.
A practical Triumphoid guide to why i do not fully trust ai faq schema suggestions,…
Architectural setup for deploying lightweight, locally hosted open models to process internal data for zero…
A practical Triumphoid guide to linkwhisper suggestions: five times i ignore them, with first-person workflow…
Copilot is six products on four billing models. Which one costs real money, the prerequisites…
A comparison of eight BPM platforms on published licensing, BPMN 2.0 support and architecture, plus…
A practical Triumphoid guide to rankmath content ai vs external ai tools: why i do…