Free or paid AI for business: what changes?

A free plan gives one person a capable AI assistant at no cost, with usage limits and fewer controls. It is a good way to learn what AI is useful for. It is a poor place for anything your customers would not want shared.

Three things matter most.

The account belongs to the person, not the business

A free account is signed up with someone's own email. When they leave, their chats, their saved instructions and anything they pasted in leave with them. You can't see it, and you can't switch it off.

Your conversations may be used to improve the model

On consumer plans, whether your chats help train future models depends on the provider and on settings each person chooses. Anthropic, for example, asks consumer users of Claude to choose whether their chats may be used to improve its models, and keeps the data for five years if they say yes, against 30 days if they don't. Few people check which they picked. Assume nothing on a personal account is private to your business.

You will hit limits at the busy moment

Free plans cap how much you can use them, and the caps can change. A bookkeeper who leans on AI at the end of the tax year can find it has stopped answering exactly when it is needed most.

A business plan moves the account, the data and the controls to the business. That matters more than any extra feature.

The business owns every account

You add and remove people in one place. When a member of staff leaves, you switch off their access that day, and the work they did stays with the business.

Your data is not used for training by default

Anthropic says it does not use inputs or outputs from its commercial products to train its models by default. Google says Workspace does not use customer data to train models without the customer's permission. Microsoft says Copilot prompts, responses and the work data it reads are not used to train its foundation models. Check the provider you choose, in its own words, and check again when its terms change.

Higher limits and shared set-up

Business plans allow more use per person, and let you share instructions and documents across the team, so everyone gets the same answer to the same question. That shared set-up is where most of the value is, as our post on writing AI instructions it will follow explains.

Admin controls you will actually use

The useful ones are dull: who has a seat, which features are switched on, whether staff can connect outside apps, and a record of use. You may not look at them often, but on the day someone leaves or a laptop goes missing, they are the difference between a five-minute job and a worry.

Yes: pay-as-you-go access through the provider's API, which is what developers use to connect AI to your own systems. You pay for what you use rather than per seat. It suits a job that runs in the background, such as reading incoming invoices, more than a person chatting all day.

Your data on pay-as-you-go access

The leading providers treat API use as business use. OpenAI says data sent to its API is not used to train its models unless you opt in, and Anthropic's API falls under the same commercial terms as its business plans.

The risk is the bill, not the data

Because you pay per use, a fault that loops can run up a bill overnight. Providers let you cap it: Anthropic's spend limits set a maximum monthly cost for the account, and you can set your own lower limit in its console. Set it on day one, before anything runs.

Who sets it up

API access needs someone to build the connection and watch it. If you would rather not, Cloud Cost Controls puts spend caps and an automatic brake on cloud and AI bills, so a mistake costs pounds rather than hundreds.

Here is how the decision plays out in a typical small office. It is an illustration with example details, not a client record.

Where they started

Three of the five staff used free AI apps on their phones to draft replies to tenants and landlords. One had pasted a tenant's reference letter in to summarise it. Nobody knew which settings each person had chosen, and the owner could not see any of it.

What they changed

They bought business seats for the three people who answer tenants and landlords, on the suite they already used for email. They wrote one shared set of instructions with their tone, their fees and the things they never promise. And they agreed one rule: no tenant or landlord details in a personal AI app, ever.

What it fixed

Every reply now comes from one set of instructions, so tenants get the same answer whoever is on shift. When a member of staff left, the owner switched off their seat in a minute and their work stayed. And the agency could answer the question a landlord asked: where does my information go?

The expensive mistakes are rarely about the subscription fee. They are about data in the wrong place and nobody watching the bill.

Paying for seats nobody uses

Buying a seat for everyone on day one feels tidy, but half of them sit idle. Buy for the job, not the headcount.

Paying, then carrying on in the free app

Staff stick with the app they know. If the business plan is not the easiest thing to reach, the free one keeps getting the customer emails. Remove the habit by putting the business plan where they already work.

Assuming paid means safe

A business plan protects your data from training. It does not stop a wrong answer going to a customer, or a person pasting in more than the job needs. Checks still matter, as our post on why AI makes things up explains, and so does basic security: the NCSC's small organisations guide to cyber security covers backups and protecting your devices and accounts.

Checking the terms once and never again

Providers change plans, limits and data terms, and the change is usually announced quietly next to louder feature news. A plan that suited you last year may not today. Put a ten-minute look at your provider's own announcements in the diary once a month, as set out in how to keep up with AI without chasing launches, and you will catch the changes that matter.