ChatGPT is free in the same way a shared drive is free. Access is easy, but trusting what you get back is the hard part. Here’s where the real cost of AI shows up for legal teams, and why governance tends to pay for itself.
I’ve noticed that legal teams tend to have two very different conversations about AI.
The first is usually about capability. What can the technology do? Which model is best? Where are the efficiency gains? How quickly can people start using it?
The second conversation happens later, usually when someone asks a simple question: why did the system produce that answer?
At that point, the discussion shifts away from technology and towards governance, consistency, and accountability. That’s why I’ve always found the argument that “ChatGPT is free” is slightly incomplete.
Technically, it’s true. Anyone can open a browser, type a question and receive a remarkably capable response within seconds. Compared with the cost and complexity traditionally associated with enterprise software, the barrier to entry feels almost non-existent.
Why “free” doesn’t mean what it used to
But legal teams have seen this pattern before.
For years, organisations relied on shared drives, email folders, and individual knowledge stores because they were easy, familiar, and inexpensive. Information was certainly available. The challenge was understanding whether it was current, consistent, complete, or connected to the decisions being made elsewhere in the organisation.
Access was easy. Trusting what you found was the hard part. And it’s how I’ve come to view AI adoption inside legal functions.
When most people think about AI costs, they naturally look at subscription fees, licensing models, and implementation budgets. Those costs are visible and easy to compare. The harder costs to quantify are often the ones that emerge much later through inconsistent decision-making, duplicated effort, unnecessary review cycles, or an inability to explain how a recommendation was reached.
In legal work, those costs accumulate quietly.
A clause is accepted that shouldn’t have been. A risk position is applied inconsistently. A recommendation looks reasonable but lacks important organisational context. None of these issues are dramatic in isolation. Yet over time they create uncertainty about whether the organisation is applying its standards consistently.
Governed AI vs. ungoverned AI
So, I’d reframe the question away from enterprise AI versus consumer AI. The line that matters runs between governed AI and ungoverned AI.
By this stage in the series, we’ve already explored why context matters, why institutional memory matters, and why legal work depends upon understanding how decisions relate to one another. Governance becomes important because it provides confidence that those things are being applied consistently.
A legal team should reasonably expect the same contractual question to result in the same reasoning regardless of whether it is asked through Teams, Slack, Copilot or another interface. The specific interaction may differ, but the underlying organisational knowledge, policies and decision-making principles should remain aligned.
That consistency becomes increasingly important as AI moves beyond helping people find information and begins influencing operational decisions.
AI will occasionally produce a poor answer. So does every other technology under the wrong conditions. What matters is understanding why an answer was produced, what information influenced it, and whether the same reasoning would be applied tomorrow.
Handled this way, governance starts to become an operational advantage rather than a compliance exercise.
Governance as an advantage, not a cost
Well-governed AI environments tend to produce more reliable outputs because they operate within clearer constraints and draw on better-structured information. Provenance is visible, and the reasoning behind a recommendation can be traced. The whole environment is easier to monitor and refine, and easier to trust. They also tend to be more efficient.
Many organisations initially approach governance as a cost centre because they focus on the effort required to establish controls. In practice, well-structured environments remove a great deal of rework. Teams spend less time validating responses, correcting inconsistencies, and resolving uncertainty because AI is operating within a clearer understanding of the organisation’s knowledge and decision-making framework.
Governance that started out as risk mitigation usually ends up paying for itself in time saved. So, I’d be cautious about evaluating AI primarily through the lens of software cost.
A relatively inexpensive tool operating without organisational context, governance or oversight can become surprisingly expensive if every answer requires manual verification. Whereas a more structured environment can be worth far more simply because people trust it enough to use it properly.
Any of these tools can generate an answer. Whether the organisation is comfortable, relying on that answer is a separate and much harder question, as capability and trust are not measured the same way.
As AI becomes increasingly embedded in legal workflows, I suspect trust will become the more important differentiator. Every organisation eventually arrives at the same point; a recommendation has been made, a decision has been taken, and someone asks why.
The ability to answer that question consistently, transparently, and confidently may prove far more valuable than any individual model feature.
This is the fifth of six articles on what it takes to get a legal team ready for AI. Click here to read the previous article, The problem is memory not access.
