Open Windows and there’s a decent chance you’ve already seen it: a small multicoloured ribbon pinned to the taskbar, or a panel tucked into the Edge sidebar. That’s the Microsoft Copilot chatbot, and it’s quietly become one of the most widely distributed AI assistants on the planet — pre-installed on new PCs, embedded in Word, Outlook and Teams, and available as a standalone app on Android and iOS.
The trouble is that “Copilot” now refers to at least three separate products with different prices, different capabilities and different privacy terms. Most complaints about it — that it’s useless in Excel, that it can’t read your files, that it’s not worth $30 a month — come from expecting one version to behave like another. So before we get to what it’s good and bad at, it’s worth untangling the names.
Three products wearing one badge
- Copilot (free) — the consumer chatbot at copilot.microsoft.com and in the Windows/Edge app. Web-grounded answers, image generation via Designer, and a daily cap on the more capable model.
- Copilot Pro — around $20 a month for personal use. Priority access during busy periods, the full model range, more Designer credits, and Copilot inside the desktop versions of Word, Excel and Outlook.
- Microsoft 365 Copilot — the business add-on, historically sold around $30 per user per month. This is the one that can see your organisation’s email, files, Teams meetings and calendar, and respects existing permissions.
Microsoft has also been rolling out a lighter “Copilot Chat” layer bundled into many business subscriptions, which grounds answers in your work data without the premium add-on. Which means the features you can reach depend heavily on your licence — and on a blog post being recent. Check your own admin centre before budgeting.
Where the Microsoft Copilot chatbot earns its keep
Underneath the rebranding it’s a fast, well-connected assistant. Three things it does consistently well:
Answers that show their working
Because Copilot’s consumer version is wired into search, it tends to return a short answer followed by numbered citations you can actually click. For anything time-sensitive — a change in import rules, a software version dropping, a sporting fixture moving — that’s a meaningful advantage over models answering from memory. I use it as a first stop for “what happened this week” questions, then read the footnotes.
Drafting and summarising inside Office
This is the part people overlook. Asking Copilot to draft a three-paragraph project update in Word, or to summarise a 45-minute Teams call you missed into decisions and action items, is genuinely faster than writing from scratch. A prompt like “summarise this vendor contract and list every clause mentioning a termination notice period” in a Word document returns something you can work with in seconds. It won’t be final-quality. It is a solid first draft, which is most of the effort.
Small visual jobs and Windows chores
Designer handles things like “a banner image with no text in it for a newsletter header” better than most people expect. And the Windows integration — screenshots, system settings, finding a document you half-remember — sounds gimmicky until you use it three times in a day.
Where it slips, honestly
Long documents are Copilot’s weak spot. Ask it to compare two 60-page reports and it will drop sections, conflate figures, or produce a summary that’s fluent but slightly wrong. Claude’s handling of lengthy documents and careful editing is a different class for this kind of work, as this breakdown of what the Claude AI chatbot is good at and where it slips sets out.
Excel is the other one. Copilot in Excel reasons about a sample of your data, not the whole sheet, and it struggles with messy headers, merged cells and anything that isn’t a tidy table. It’s handy for formulas and clarifying a chart story. It is not a replacement for a pivot table, and treating it like one wastes an afternoon.
Citations also need scrutiny. Copilot will sometimes link a source that only loosely supports the claim. If a number matters, open the link.
Free tier or paid? A quick way to decide
If your use is occasional — checking a fact, rewriting an email, making a quick image — the free version is generous enough that you may never pay. It’s on a shortlist of the best free AI chatbots worth using precisely because the free tier isn’t crippled.
Pay for Copilot Pro if you live in Word and Outlook on a personal machine. Pay for Microsoft 365 Copilot only if your organisation has clean, well-permissioned SharePoint and nobody freaks out about an assistant reading internal mail. The value scales with how well your company’s information is already organised — which is a technology problem disguised as an AI problem.
Five habits that improve its output immediately
Most disappointing results are prompt problems. These five changes make the biggest difference:
- Name the format and length: “five bullets, under 20 words each” beats “summarise this”.
- Paste the source text instead of describing it. Copilot works with what it can see.
- Ask for the sources as links at the end so you can verify them in one pass.
- Use it for the second draft, not the first — editing something rough is faster than staring at a blank page.
- Tell it what you’ve already tried and what was wrong with it. It corrects course quickly.
How it stacks up against the alternatives
Copilot’s edge is plumbing, not raw reasoning. It’s already inside the tools you use for eight hours a day, which is why so many people end up defaulting to it. On pure conversational depth, ChatGPT still feels more capable for complex reasoning; on integration with Google Workspace and long context, Gemini has the advantage. If you’re weighing them up, this ChatGPT vs Copilot comparison and this one on whether Gemini or Copilot fits your working day are both more useful than spec sheets.
One practical note: the desktop app version is noticeably better than the browser tab you keep forgetting to close, mostly because it stays logged in, remembers context and can work with files on your machine. There’s a good argument for why the desktop app beats the tab if you’re still living in the browser.
The privacy distinction nobody reads
The free consumer Copilot is a consumer product. Treat your prompts as you’d treat a public forum post, and don’t paste client contracts, employee data or anything under NDA into it. Microsoft 365 Copilot is a different animal: it runs inside your tenant’s data boundary, only surfaces files the signed-in user already has permission to open, and produces an audit trail. Same name, opposite risk profile. If you’re rolling it out across a team, spend an hour on permissions before you spend a month on prompts.
A one-week test that tells you whether to keep it
Don’t evaluate an assistant by asking it trivia. Give it real work with a deadline. Try this across five working days: on Monday, have it summarise the longest unread document in your inbox into five bullets; on Tuesday, ask it to draft a difficult email and then critique its own draft; on Wednesday, get it to explain an unfamiliar spreadsheet formula in plain English; on Thursday, have it turn a document you wrote into a slide outline; on Friday, ask it to prepare you for a meeting using the invite and the last three relevant emails.
Score each attempt on whether you’d have had to redo it from scratch. Three or more saves and it’s worth keeping. One or two and the free tier is plenty. Zero, and the problem probably isn’t the model — it’s that your prompts are vague and your source documents are a mess. Fix those two things and the same Microsoft Copilot chatbot that felt like a novelty in week one starts quietly removing an hour of admin from your week.

