Microsoft 365 Copilot vs ChatGPT After the Share Shift
Google’s distribution story is obvious. Microsoft’s is quieter and more expensive per seat. The ChatGPT-share headlines do not make a $30 Copilot seat a bargain.
This page is periodically reviewed to reflect current pricing and plan changes.
Fastest win
If the work starts in Word, Excel, Outlook and Teams, Copilot can be the fit. If the work is a jumble of PDFs, browsers and other vendors, ChatGPT or Claude will usually beat a per-seat Microsoft add-on. Count unused seats before you roll out.
Share charts do not know where your files are
Gemini’s rise is partly distribution: Android, Search, Workspace. Microsoft has the same kind of distribution inside enterprises that standardised on Office a decade ago. ChatGPT’s narrower consumer lead does not tell you whether Copilot will get used in Excel.
Map ten real workflows. If eight of them begin in Microsoft data, Copilot is in the conversation. If eight of them begin in a browser, a $30 add-on will become wallpaper.
Cost per accepted office artefact
The accepted outcome is a sent deck, a cleaned spreadsheet, a meeting note that action-owners recognise. Include the Copilot seat, the underlying M365 seat, and the time people still spend fixing citations and numbers.
ChatGPT Business can be cheaper per accepted artefact for mixed work even if the sticker looks similar, because you are not paying the Microsoft tax twice.
Pilot 20 people. Measure weekly genuine use. Then decide on 200.
ChatGPT Business for mixed work; Copilot when Graph is the job
ChatGPT remains the broader assistant. Copilot is a Microsoft-365-shaped product. The market-share chart does not flip that architecture.
The shelfware pattern
Wide Copilot rollouts fail in a familiar way: licences land, a few power users love Excel insights, and everyone else keeps pasting into ChatGPT because that is where their non-Microsoft files live.
That is not a training problem. It is a fit problem. Count weekly genuine use, not activation emails.
If finance will not let you buy 20 pilot seats without buying 2,000, that is a procurement process to fix — not a reason to over-buy.
Ranked recommendation
Best choice for Microsoft-native operators: Copilot for the roles that live in Graph, not for the whole company on day one.
Best choice for mixed tools: ChatGPT Business or Claude Team for the people with a named job, plus free Copilot where it already appears in the apps.
Avoid stacking Copilot and ChatGPT Business on the same 50 people “to be safe”. That is the opposite of safe for the budget.
Key Takeaways
- →Distribution explains Gemini; it does not automatically justify Copilot.
- →Start with where files and mail already live.
- →Pilot active use before a wide seat buy.
- →Do not double-seat the same humans on two general assistants.
Editorial context
Who is this for?
IT and finance teams being asked for Copilot licences because “ChatGPT is losing”.
When NOT to use this
Google Workspace companies. You are in a Gemini conversation, not this one.
Pricing insights
Copilot’s per-seat add-on is easy to underestimate because it lands on an estate you already resent paying for. Model active seats, not purchased seats.
Alternatives to consider
Claude Team for writing-heavy firms. Gemini in Workspace for Google estates. Individual $20 seats for the ten people who actually need a general assistant.
Final verdict
Buy Copilot where Microsoft data is the work. Buy ChatGPT or Claude where it is not. Do not buy both for the same people.
Frequently Asked Questions
If ChatGPT’s consumer share dropped, should we drop Microsoft 365 Copilot?
No. Copilot is a suite-native job. Buy it when work already lives in Microsoft data. ChatGPT share charts do not price that integration.
When is ChatGPT the better productivity seat?
When the work is cross-domain and does not start in Outlook, Teams or SharePoint. Map ten workflows to source systems before you add a second seat.
What makes Copilot look expensive?
Licences on people whose real work is not in Microsoft 365. Unused seats dominate the bill faster than model quality debates.