ChatGPT for business: what it costs and when it's worth it

What ChatGPT for business actually costs

OpenAI sells ChatGPT in tiers, and the pricing moves often enough that the sensible way to think about it is by capability rather than by exact dollars. The individual paid plan has sat at around US$20 per person per month for a long time. Above that sits a much pricier individual tier for people who live in the thing all day. The two tiers that matter for a business are the per-seat workspace plan and the enterprise plan, because they're the ones that give you an admin console and a written position on your data.

The number people get wrong isn't the seat price, it's the seat count. Ten seats bought on a hunch costs more over a year than the one workflow you were hoping to fix. Buy for the people who have a named job for it, then add seats as the use spreads.

One detail worth reading twice: on the consumer tiers, your chats can be used to improve OpenAI's models unless someone turns that off in settings. On the business and enterprise workspaces it's off by default. If staff are already pasting client work into personal accounts, that difference is the whole argument for moving them onto a workspace.

The uses that reliably pay off

Across the businesses I work with, the same handful of jobs come up again and again. They share a shape: high frequency, a human still checks the output, and a bad draft costs nothing.

• First drafts of anything repetitive. Proposals, quotes, follow-up emails, job ads, product descriptions. You edit rather than start from a blank page.
• Summarising long documents. Contracts, reports, meeting transcripts, a month of customer feedback. Ask for the decisions and the risks, not a general summary.
• Turning messy notes into a usable document. Voice notes to a scope of work, a site visit to a client update, a whiteboard photo to a project brief.
• Explaining your own data. Upload a spreadsheet and ask what changed month to month. Treat the answer as a lead to verify, not a finding.
• Writing the SOP nobody has written. Have the person who does the job talk through it, then have ChatGPT turn the transcript into a procedure and edit it down.
• Thinking out loud before a decision. Not for the decision itself, for the list of things you've forgotten to consider.

There's a longer, function-by-function breakdown in the ChatGPT use cases for business guide, including the ones I'd tell you to skip.

Rolling it out without wasting three months

A rollout is five decisions, and most businesses skip four of them. Pick two or three workflows that a specific person does weekly. Give it to those people first. Write down, in one page, what may and may not be pasted into a chat: client identifiers, payroll data, anything under a confidentiality clause. Name one person who owns the thing. Then set a date thirty days out to look at real usage rather than opinions.

Training is where the money is made or lost, and a lunchtime demo isn't training. One working session per role, on that role's actual tasks, changes what people do on Monday. There's a full breakdown in AI training for employees.

Mistakes that waste the subscription

• Buying a seat for everyone in week one. Usage collapses after the novelty and you're paying for logins nobody opens.
• Using it like a search engine. One-line questions get generic answers. Give it the context, the audience and an example of good output and the quality changes completely.
• No data rules. Without a written rule, someone will paste a client contract into a personal free account. That's a policy problem, not a technology one.
• Nobody checking the output. It's confident when it's wrong. Whoever sends the email still owns the email.
• Banning it instead of governing it. Staff use it anyway, on personal accounts, with no oversight. A workspace plus a one-page rule is a better answer.
• Building something custom before proving the manual version. If the workflow doesn't work with a person driving it by hand, automating it just makes the mess faster.

Is ChatGPT even the right pick?

Usually, but not always, and plenty of businesses run more than one assistant. ChatGPT has the broadest ecosystem, the most integrations and the deepest bench of features. Claude tends to be the pick for long documents and careful writing. If your team lives in Microsoft 365 or Google Workspace, the assistant already bundled into those tools may cover half of what you were about to buy. I've written the honest version of that decision in Claude vs ChatGPT for business.

My job is to help you decide how to use AI: what to do, in what order, and what to avoid, before you spend time and money building the wrong thing. I don't sell the software I recommend and I take no commission on it. If you want something custom built after that, it's a separate engagement with its own scope, price and contract.