Your Employees Are Using AI. But Who Owns the Knowledge They’re Building?

Imagine you run a business with 30 employees.

Several of your staff are already using ChatGPT. Some have free accounts. Others pay for their own subscriptions. They use it to prepare proposals, research customers, draft reports, summarise information and solve everyday problems.

From the employee's perspective, this can be entirely positive. They're showing initiative, developing new skills and finding better ways to approach their work.

But I think there's a question that many business owners haven't yet considered.

Where is all that business knowledge going?

When an employee uses ChatGPT regularly, they may gradually develop quite sophisticated ways of working. They refine their instructions, establish useful document structures, discover better ways to analyse information and learn how to get consistent results for recurring tasks.

Some of that knowledge may exist only in their personal AI conversations, files and individual working practices.

What happens when they leave the organisation? Can somebody else reproduce the work? Does the company know what information they've been sharing with these tools?

And perhaps most importantly, is the business developing an organisational AI capability, or simply accumulating employees who happen to be good at using AI?

I believe this distinction will become increasingly important as Irish companies move beyond their first experiments with generative AI.

The AI adoption happening across Irish workplaces

The widespread use of AI tools at work is no longer speculative.

Deloitte Ireland's March 2026 GenAI Pulse reported that 65% of respondents used free external generative AI for work or paid for the tools themselves. Only 35% reported working for an employer that paid for their generative AI tools.

The study surveyed 1,000 adults aged 18–75 in Ireland.

Similarly, Trinity Business School's AI Economy in Ireland 2026 report, based on responses from 250 senior organisational decision-makers, found that 82% reported employees using free or public AI tools for work.

Interestingly, nearly 93% reported that their organisations explicitly permitted such use.

These are different surveys, measuring different things. But together they illustrate how extensively AI has entered everyday working life, often through tools that individuals can access directly.

It's worth being careful with the terminology here.

You may have heard the expression shadow AI, referring to the use of AI tools without appropriate organisational awareness or approval. That is a genuine governance concern, but using a personal or free AI account is not automatically unauthorised. Some businesses explicitly permit these tools.

The more important management question is whether their use is sufficiently understood, supported and governed.

An employee may be using AI with the best intentions, following a permitted practice and producing useful work. But that doesn't necessarily mean the organisation is capturing the resulting knowledge or establishing a repeatable business process.

The overlooked problem: individual productivity isn't organisational capability

Much of the discussion about AI in business focuses on individual productivity.

Can an employee write a report more quickly? Can they prepare a presentation, summarise a document or research a customer?

These are worthwhile questions. But they only describe part of the opportunity.

Consider what happens as someone becomes increasingly experienced with ChatGPT.

They learn which instructions work. They develop ways to structure inputs. They identify useful reference materials. They refine templates and establish a consistent approach to recurring tasks.

Over time, that experience may become an important part of how they perform their job.

Yet the business may have no documented record of the method.

Another employee facing the same task may start from scratch, develop a completely different approach or make mistakes the first employee has already learned to avoid.

The organisation can end up with several people independently solving similar problems, without combining their learning into a shared way of working.

This is particularly relevant to professional services and other document-intensive businesses, where staff repeatedly prepare proposals, reports, research, correspondence and management information.

The real opportunity is not simply to have more people using AI. It's to turn the useful things people discover into working methods the organisation can retain and improve.

That means treating practical AI knowledge as something worth documenting, testing and maintaining, much as a business would with an established process, template or standard operating procedure.

What happens when the employee leaves?

Consider a fictional professional services firm employing 30 people.

One of its senior managers has been using ChatGPT for several months to help prepare recurring client proposals.

Over time, the manager has developed a useful approach. They know which information to gather, how to structure the instructions, what the proposal should look like and which sections require particular attention.

The process works reasonably well for them.

But the instructions are scattered across conversations in their personal ChatGPT account. The company has never documented the method, agreed which information sources should be used or tested whether another employee could follow it.

Then the manager leaves.

The business still has its client files and company documents. But it may no longer have access to the working instructions or practical knowledge that made the AI-assisted process useful.

The replacement manager may have to reconstruct the method independently.

The issue isn't necessarily who legally owns the intellectual property. Employment contracts and intellectual property rights are separate matters.

The operational problem is that the company has become dependent on a working method it hasn't properly captured or made accessible to authorised colleagues.

This can happen with ordinary business software too. Employees develop useful spreadsheets, processes and shortcuts that nobody else understands.

AI introduces another place where valuable working knowledge can accumulate, particularly when an employee's conversations and working context remain in an individual account.

The answer isn't to demand access to everybody's personal conversations. Those may contain private information, material belonging to other organisations and content that shouldn't be copied into company systems.

The answer is to establish a deliberate process for identifying useful business methods and recording approved versions in company-controlled locations.

There's a second problem: what company information is being shared?

Knowledge retention is only half the issue.

The other concern is information governance.

When employees use AI tools for business work, what information are they entering?

A prompt might contain a customer's name, an extract from a commercial agreement, an internal sales report or information about employees. A document uploaded for summarisation might include personal data or confidential material.

Not every use of this information is necessarily inappropriate. The question is whether the organisation has assessed and authorised the processing.

The Irish Data Protection Commission's guidance on AI and large language models emphasises that organisations need to understand what personal information an AI system processes, how that information is used, where it goes, whether it is retained and what obligations apply.

These questions become more difficult when employees independently choose their own tools or account arrangements.

Take personal ChatGPT accounts as an example.

OpenAI provides settings through which eligible personal users can turn off the use of their conversations for model improvement. ChatGPT Business operates under different data arrangements, with business workspace content excluded from model training by default.

That is an important distinction, but it shouldn't be exaggerated.

A personal account is not automatically a data breach. Equally, paying for a business subscription does not automatically make every proposed use of personal or confidential information GDPR-compliant.

Organisations still need appropriate policies, lawful processing arrangements, access controls, vendor assessment, staff guidance and, where required, a data protection impact assessment.

The objective should be to create an environment in which employees understand what they can use, why the arrangements are appropriate and who is responsible for reviewing them.

Would moving everyone to ChatGPT Business solve the problem?

A company-managed AI workspace can be a useful part of the solution, particularly for businesses already using ChatGPT extensively.

It provides a more deliberate organisational environment for managing members, workplace access, approved tools and applicable data arrangements.

But there's an important limitation.

Moving employees into ChatGPT Business does not automatically turn their private conversations into shared company knowledge.

Individual users retain private chat histories, and other workspace members do not automatically gain access to those conversations. Personal memories are also associated with individual users rather than becoming a collective memory shared by everyone.

That distinction matters.

If a business wants to retain the useful knowledge developed through AI use, it needs to do more than provide licences.

It needs an agreed way to capture business-relevant instructions, approved information, templates, examples and workflow documentation in shared, appropriately controlled assets.

The underlying information should remain current and accessible to authorised staff. Reusable AI workflows should be tested, reviewed and updated when requirements change.

This doesn't require every employee to use an identical prompt for every task. People still need room to exercise professional judgement and adapt to the work in front of them.

It does mean that valuable organisational methods should not depend entirely on one person's private conversation history.

From personal experimentation to cumulative company knowledge

This is where I believe the bigger opportunity lies.

Imagine a business identifies a recurring task that several employees perform.

Instead of having each person independently work out how AI might help, the company agrees a common approach.

It identifies suitable information sources, establishes an approved document structure, develops a reusable set of instructions and tests the results with the people who actually perform the work.

The method is then documented and made available to authorised colleagues.

As employees use it, they can identify improvements, flag problems and suggest changes. Those improvements can be reviewed and incorporated into the approved method.

Over time, the organisation builds a better collection of working methods rather than simply accumulating more individual AI conversations.

That's what I mean by cumulative organisational AI capability.

It isn't a claim that the AI model automatically absorbs everything employees do or continuously trains itself on company activity.

The accumulation is deliberate. The organisation retains approved knowledge, templates, instructions, examples and review records. Improvements are made through a controlled process.

For some businesses, this may be as straightforward as maintaining approved templates and instructions in SharePoint or another company document system.

Others may benefit from configured AI workspaces, shared GPTs, reusable Skills, controlled knowledge sources or more specialised workflows.

The appropriate arrangement depends on the business, its existing technology, the information involved and the work employees need to perform.

What matters is that the knowledge becomes accessible, reviewable and maintainable rather than remaining scattered across individual accounts.

The AI Knowledge Continuity Test

Here's a simple exercise I would encourage business owners and managers to consider.

Think of an employee who is already making regular and productive use of ChatGPT or another AI tool.

Now imagine they're unavailable tomorrow.

Could somebody else reproduce their AI-assisted work to an acceptable standard?

Five questions can help assess the position:

1. Information: Can another authorised employee find the approved documents and information required for the task?
2. Method: Are the useful instructions, templates and working steps documented outside the employee's private AI conversations?
3. Quality: Is there a clear example of an acceptable output, including what needs to be checked?
4. Responsibility: Does somebody understand who reviews and approves the completed work?
5. Continuity: Is there an agreed process for updating the method when information, staff or business requirements change?

If several answers are unclear, the business may have an opportunity to make its AI-assisted work more consistent and resilient.

This isn't a formal AI maturity score or a compliance assessment. It's a practical way to start a useful management conversation.

The same questions apply whether the company uses ChatGPT, Microsoft Copilot, another AI platform or a combination of tools.

Start by understanding what employees are already doing

I wouldn't recommend beginning with a blanket prohibition on personal AI use.

Businesses first need to understand why employees are using these tools and which activities are genuinely valuable.

A useful starting point is to speak to staff about the tasks they're performing, the tools they rely on and the information they need.

Some applications may be suitable for continued individual use under clear rules. Others may justify approved company tools, shared workflows, staff training or more formal governance arrangements.

The business should then identify which useful methods deserve to be retained, who should approve them and how they will be maintained.

Not every prompt needs to become a standard procedure. Not every employee experiment needs to become a company-wide system.

And not every organisation needs an external provider to manage its AI activity.

Some businesses will have the internal capability to establish and maintain these arrangements themselves. Others may benefit from help with workflow development, knowledge preparation, training or governance.

The point is to make a conscious organisational decision rather than allow the arrangements to develop entirely by accident.

The next phase of business AI adoption

I think we're approaching an important change in how businesses need to think about AI.

The initial question was whether employees should use tools such as ChatGPT.

The next question is what companies should do with the experience, methods and knowledge employees are developing through that use.

Businesses that address this thoughtfully have an opportunity to build working methods that can be shared, evaluated and improved over time, while establishing more appropriate oversight of their information.

The alternative is not necessarily a dramatic security incident. It may simply be a missed opportunity: valuable knowledge remaining fragmented, duplicated and difficult to retain.

For business owners, that should be enough reason to examine how AI is currently being used.

Your employees may already be getting better at working with AI. The question is whether your organisation is getting better with them.

Do you know how AI is being used across your business?

AI Strategy & Training Ltd helps Irish organisations identify suitable AI opportunities, develop practical workflows, prepare business knowledge and establish responsible working practices.

If your employees are already using ChatGPT or other AI tools, we can help assess what's working, what needs attention and where a more structured approach might be worthwhile.

Discuss an AI workflow opportunity with AIST.