Using ChatGPT doesn't mean your business is using AI

Ask a business whether it uses AI and the answer is increasingly likely to be yes. Ask what would stop working if every ChatGPT, Copilot or Gemini account disappeared tomorrow, and the answer may be: not very much.

That is not a gotcha. Staff using a large language model to research a topic, improve a draft or untangle a spreadsheet can save real time. It may be the sensible first step. But it is not the same as changing how the business works.

The phrase 'we use AI' now covers at least three very different realities: people using a tool, workflows using AI, and a business building an AI capability. Treating them as interchangeable makes adoption look more advanced than it is and makes the next step harder to see.

The numbers are bigger than the systems

Official data suggests that this gap is common. The UK Business Data Survey 2026 found that 41% of businesses handling digitised data used AI for at least one purpose. The leading uses were modest and familiar: researching information, and summarising material or drafting reports and correspondence. [1]

Only 21% of the businesses that used AI said their tools were integrated into existing business systems. Even that definition was broad enough to include Copilot embedded in Microsoft 365. Just 17% had any AI policy or guidance, and only 5% had a formal written policy. [1]

The gap in four numbers

This is a recognisable picture. AI is present at people's desks, but it is not necessarily present in the operating model.

The measurement problem is not uniquely British. The US Census Bureau widened its question in late 2025 from AI used in producing goods or services to AI used in any business function. Even under the broader wording, reported use hovered between 17% and 20% from December 2025 to May 2026. [2]

A related Census working paper found that 57% of adopting firms used AI in three or fewer business functions. Writing, document analysis and information search were the leading generative-AI tasks, and 65% of firms limited use to three or fewer tasks. The researchers also found adoption moving in both directions: employees sometimes used AI without formal firm-level adoption, while firms sometimes declared adoption without much worker use. [3]

That is why a headline adoption rate can tell us that activity is happening without telling us how much capability has actually been built.

Three things that all get called 'using AI'

The dividing line is repeatability, ownership and organisational memory. Editorial framework, not a validated maturity scale.

1. People use AI

Someone has a subscription, a handful of good prompts and a growing sense of which jobs the model can help with. The gains may be real, but they are local. The method lives in a chat history, a personal note or the user's head.

If that person is away, changes role or leaves, much of the capability leaves too. Sometimes the company's AI strategy is, in practical terms, 'Dave has some good prompts'.

2. Workflows use AI

A workflow is different. It has a recurring task, a trigger, known inputs, a sequence of steps, a defined output, checks and an owner. The AI is part of the process rather than an optional side conversation.

For example, a salesperson pasting notes into ChatGPT is individual use. A process that turns every call note into a structured CRM update, flags uncertainty for human review and records the approved result is an AI-assisted workflow.

The dividing line is not the software. It is repeatability.

3. The business builds AI capability

Capability begins to belong to the business when several workflows share standards, data rules, review methods, ownership and measures. People know what may be entered into a model, which outputs need checking, who is accountable for errors and how successful methods are improved.

This does not require a grand transformation programme. It requires the organisation to retain what it learns. A useful process should survive the departure of the person who first discovered it.

This is not an argument for a transformation programme

The sensible response is not to commission a 40-page AI strategy because somebody used ChatGPT to improve an email. Nor is it to dismiss employee experiments as trivial.

Start where use is already producing value. Pick one recurring task. Capture the inputs, the steps and the output. Define what a good result looks like, where human review is required and where the approved result should go. Then ask whether a colleague can repeat it.

The first move from tool use to workflow is usually modest:

• a defined task rather than a vague hope
• a clear owner rather than an enthusiastic volunteer
• a check for errors rather than confidence in a polished answer
• a destination for the output rather than another forgotten chat
• a way to improve the process rather than starting from scratch next time

That small amount of structure is what turns a useful conversation into something the business can own.

The five-question AI reality check

This is not a scientific maturity assessment. It is a quick way to find out which version of 'we use AI' you actually mean.

1. Can we name the recurring task that AI is helping with?
2. Can another person repeat the process without borrowing the original user's brain?
3. Do we know where the inputs come from and where the approved output goes?
4. Is there a defined check for errors, unsuitable data or poor judgement?
5. If the current AI enthusiast left tomorrow, what useful capability would remain?

Andrew's Intel editorial decision aid; not a scientific maturity assessment.

Mostly no means that people are using AI. A mixture of yes and no suggests an emerging workflow. Mostly yes means organisational capability is beginning to take shape.

A business that answers yes to all five may still be early. But it is building something that belongs to the organisation rather than merely renting productivity through individual accounts.

Use the label honestly

ChatGPT can be useful without being transformation. Individual productivity can be valuable without amounting to operating-model change. There is no shame in being at the first level. For many organisations, it is exactly where sensible adoption begins.

The problem starts only when the label becomes bigger than the capability. A company says it is using AI, assumes the work is underway, and never converts scattered learning into repeatable practice.

The next step is not necessarily another tool. It is to turn one useful conversation into a checked workflow that the business can keep.

Using ChatGPT is evidence of AI demand. It is not yet evidence that the business has built AI capability.