ChatGPT vs Microsoft Copilot at Work

ChatGPT vs Microsoft Copilot is a job question, not a brand question

Teams argue about ChatGPT versus Microsoft Copilot as if one winner should replace the other. That argument wastes a quarter. The products sit in different places in the work. ChatGPT is a general conversation: you bring the brief and the paste. Microsoft 365 Copilot is help inside the documents, mail, and meetings you already live in, using Graph permissions your administrator set. Comparing them as if they were two chat boxes with different logos hides the decision you actually have to make: where the file lives, who may see it, and who will sign the output.

People also mix Copilot names. GitHub Copilot helps with code. Microsoft 365 Copilot helps with office work. Windows Copilot and Bing chat are another surface. If your security review says “Copilot,” write down which one. A ban on the wrong product and a free pass on the right one is how confidential decks still leak.

What each tool is actually good at

ChatGPT is strong when the task is thinking, outlining, rewriting, role-play, or a first draft from a short pack you are allowed to provide. You control the paste. You also bear the risk of that paste. It is weak when the answer depends on a file it cannot see, or when the user treats web-flavoured fluency as a citation.

Microsoft 365 Copilot is strong when the artefact already exists in Word, Excel, Outlook, or Teams: summarise this thread, draft from this document, propose a formula next to this table. It is weak when the tenant search is messy, when permissions are wrong, or when people assume it “knows the company” rather than the files their account can read. Garbage in the share drive becomes confident garbage in the draft.

A decision rule your team can remember

If the work object already lives in Microsoft 365 and Copilot is licensed and approved for that data class, stay there. Do not export a customer workbook into a consumer ChatGPT window to “make it nicer.” If the work is a thinking pass on allowed notes, ChatGPT or an approved chat is fine. If the work is code, use the coding copilot your engineering standard named, not an office copilot by habit.

Role: You are a technology buyer helping a team pick one default tool for a task.
Task: Recommend ChatGPT, Microsoft 365 Copilot, or neither.
Context: I will paste the task, where the file lives, the data class, and who signs the output.
Constraints:
- If data is not allowed in the tool, recommend neither.
- If both could work, pick one default and say what the other is for.
- Do not recommend a second tool to “see what happens.”
Output: Recommendation, reason, and the review step a human still owns.

Write the default on one page. “Status decks: Copilot in PowerPoint. Client letters: approved chat, then Word. Source code: GitHub Copilot. Anything restricted: neither until legal says yes.” A default beats a debate in every stand-up.

Cost, training, and the hidden duplicate

Licences look cheaper than they are if people still paste the same file into the other tool. Count duplicate use as waste and as risk. Training should be the decision rule and the review checklist, not a feature tour. A twenty-minute session that shows when to stay in the document and when to open chat will save more time than a webinar on plugins.

Measure whether drafts got shorter to review, not whether more words were generated. Copilot that produces a six-page summary of a two-page memo is not a win. ChatGPT that invents a figure you then spend an hour disproving is a loss. The comparison that matters is time-to-trusted-output, not tokens.

What to do this week

List the five artefacts your team produces most. For each, write ChatGPT, Microsoft 365 Copilot, or neither, plus the data class. Share that list. Kill one duplicate path. Read the neighbouring essay on a tool stack that fits if you are still collecting icons instead of choosing defaults.