Two products come up in nearly every AI conversation with a small business. Microsoft sells Copilot, which lives inside the applications your staff already have open. OpenAI sells ChatGPT, which most of your staff have already used on a personal account. They are not the same product and they do not fail in the same places.

Prices move, so treat every figure here as a starting point and check the current list before you sign anything. Both currently sit near $30 per user per month on business plans. For a 20 person company that is about $7,200 a year if you buy it for everyone, which is usually the wrong plan.

What Copilot does well

Copilot is strongest when your work already lives in Microsoft 365. It can summarize a 40 message Outlook thread, catch you up on a Teams meeting you missed, pull the numbers out of a spreadsheet without anyone writing a formula, and find the version of a proposal someone saved in March. Nobody has to copy text into a separate window, which is the step most staff skip when they are busy.

That strength has a condition attached. Copilot can only reason over files it can reach, which means SharePoint, OneDrive, Teams and Exchange. If your documents sit on a local drive, a NAS, or a shared folder nobody migrated, Copilot is close to useless and you will be paying full price for it.

There is a second condition. Copilot honors your existing permissions, so anything already overshared becomes easy for staff to find and summarize. A folder that was technically open to everyone but buried five clicks deep is now one sentence away. Run a sharing report on SharePoint and OneDrive before you turn Copilot on, not after.

What ChatGPT does well

ChatGPT is stronger on open ended work. Drafting from nothing, rewriting a page in a different tone, reading a 60 page PDF and answering questions about it, cleaning up a messy export, working through a problem in several passes. It handles a blank page better, and it holds a longer back and forth without losing the thread.

The business plan matters here. A personal account gives you no admin console, no ownership of the work, and no guarantee about how conversations are handled. A business workspace gives you central control of accounts, keeps company conversations out of model training by default, and lets you remove a leaver from the account instead of hoping they forget their password. Seats are sold in small blocks, so you can start with two.

Its weakness is the reverse of Copilot's. It does not sit inside your mail and files, so people paste things in. That is exactly the moment client data leaves your control, which is why the usage policy matters more on this side.

The cost line most people miss

Microsoft 365 Copilot is an add-on, not a standalone product. You need a qualifying Microsoft 365 business plan underneath it, which runs roughly $8 to $27 per user per month depending on the tier. So the true cost of a Copilot seat is that base license plus the add-on, and the add-on is normally sold on an annual commitment.

If you run on Google Workspace rather than Microsoft, most of the Copilot argument disappears before you start, because the tools it plugs into are not the tools you use.

Before you compare monthly prices, write down how many seats you actually need. In most companies of 5 to 100 people, 5 to 10 staff do the writing, analysis and email volume that justifies a paid seat. Buying for everyone is how a $2,000 experiment becomes a $20,000 one with the same result.

When neither is the answer

Most Microsoft 365 business plans now include a chat-only Copilot at no extra charge. It answers general questions with commercial data protection but does not reach into your own mail and documents. For a lot of small teams that free tier covers the drafting and rewriting work, and the paid upgrade only earns its money for the handful of people who live in Outlook and Excel all day. Check what your existing plan already includes before you buy anything.

Sometimes the problem is not a chatbot problem at all. If quotes take three days to go out, and two of those days are one person waiting to approve them, no assistant fixes that. Write down the workflow and the hours it costs you per week first. If the answer is a process change or a different piece of software, buy that instead.

How to decide in two weeks

Pick by where the work already sits. Heavy Microsoft 365 use, documents properly in SharePoint, meetings in Teams: trial Copilot. Lots of writing, research, analysis and one-off tasks, or a mixed set of systems: trial ChatGPT Business.

Buy five seats, give them to the five people most likely to use the tool daily, and run it for 30 days with one measurement agreed in advance. Then decide with a number rather than an impression.

If you want help picking, the first call is free and runs about 30 minutes. We look at where your files live, what your current licenses already cover, and which five people would actually use a seat. Get in touch and pick a time.