Somebody in your business has been using the free version for months, likes it, and has asked whether the company will pay for the proper one. It is a small decision, and it is worth ten minutes rather than a shrug, because the usual reason to pay is not the one people cite.
The short version
- The free tier is genuinely capable. Learn on it.
- Pay when you hit a limit that costs you time, not because paid sounds more serious.
- If anyone touches work data, buy the business tier rather than the personal one. That is a data decision, not a feature decision.
- Licence the people who write and think for a living first.
Three tiers, not two
Most comparisons treat this as free against paid. There are really three, and the interesting jump is the second one.
Free. Access to a capable model, usually with limits on how much you can use the best one before being moved to a smaller one. File uploads, image generation and web search are typically included with tighter caps. Historically, conversations may be used for training unless you find and change the setting.
Personal paid. Higher limits, first access to new models and features, better availability at busy times. Still an individual consumer account, held in that person's name, with consumer terms.
Business or team. Per-seat licensing under a proper business agreement. Your prompts are not used to train the models, and the provider says so contractually. Central admin, so you can add and remove people. Shared workspaces and shared saved instructions. This is the tier that has an answer when someone asks who your data processors are.
The difference that actually matters
It is not the model. On most everyday business writing, the gap between the free model and the best one is smaller than the gap between a lazy prompt and a good one. A well-briefed free tier beats a badly-briefed paid tier on any given Tuesday.
The difference that matters is the jump from consumer to business terms, and it is a governance difference rather than a capability one.
On a personal account, whether or not it is paid, the account belongs to the individual. If they leave, the history leaves with them. You have no administrative visibility. You have no meaningful data processing agreement. And unless every one of them found the training toggle, your work may be improving somebody's model.
The moment anyone is putting real work into these tools, that is the line to cross. Not because of the features. Because you can answer the question. This is set out in more detail in Is it safe to put company data into ChatGPT?
What you get for the money
| Free | Personal paid | Business / team | |
|---|---|---|---|
| Best models | Limited use | Generous | Generous |
| Usage caps | Hit regularly in daily work | Rarely hit | Rarely hit |
| Trained on your prompts | Often, unless disabled | Often, unless disabled | No, contractually |
| Data processing agreement | No | No | Yes |
| Admin control | None | None | Add, remove, audit |
| Account survives a leaver | No | No | Yes |
| Shared instructions | No | Personal only | Workspace-wide |
| Sensible for | Learning, personal use | One heavy individual user | Any team touching work data |
Signals that it is time to pay
Concrete, in rough order of how often they come up:
- Someone hits the cap mid-task, more than once a week. Being cut off halfway through a piece of work is the most expensive limit there is, because the context is lost and the job restarts.
- Work information is going in. Even anonymised internal material. Get onto business terms.
- Two or more people want the same setup. Shared instructions stop everyone reinventing the house style.
- Someone is doing this daily. At daily use, the limits and the queueing cost more than the licence.
- You need to answer a supplier questionnaire. Client due diligence forms increasingly ask which AI tools you use and on what terms. "Personal free accounts" is a bad answer to write down.
And the signals that you should not pay yet: nobody has used the free version for a fortnight; the request is "so we can say we use AI"; or one person wants it and has not yet found anything it is for. Time on the free tier is not a delay, it is the thing that tells you what to buy.
Buying it sensibly
Start with three seats, not everybody. Give them to the people who write and think for a living: whoever handles proposals, whoever answers awkward customer emails, whoever writes the marketing. Watch for six weeks, then extend based on what actually happened rather than who asked loudest.
Check what you already pay for. Businesses regularly buy standalone licences while an equivalent assistant sits unused inside the Microsoft or Google subscription they already have. Look before you spend.
Configure it on day one. Confirm training is off. Set the workspace instructions with your company description and house style. Add the people. Fifteen minutes, and it is the difference between a tool and a licence nobody opens.
Set a review date. Three months. Who used it, for what, what changed. Per-seat subscriptions are unusually good at quietly outliving their usefulness.
The way to think about the cost
Not "can we afford it" but "how many hours must it save to break even". Take the monthly per-seat price, divide by the loaded hourly cost of the person using it, and you have the number. For nearly any office role it lands under two hours a month.
Two hours a month is one meeting summarised, one proposal drafted, or one difficult email untangled. If someone is using it at all seriously, they passed that in week one.
Which is why the honest advice is rarely "do not pay". It is: do not pay yet, and when you do, buy the tier that lets you answer questions about your data. There is a one-page policy template to go alongside it in Write an AI policy for your small business in one page.
If you would rather see what a properly configured account looks like before spending anything, the free AI Breakfast Club webinar runs online every other Friday morning and there is nothing to buy at any point.