The cost to claim is nothing. The cost to defend is rising.

A client reads your report and takes issue with something in it. Two years ago, they would have telephoned, or sent a short email, or most likely done nothing at all.

Now they open ChatGPT, describe the problem in one sentence, and ask it to write a letter. Ninety seconds later, they have four pages of formal language, properly structured, referring to duties they had never heard of that morning. They send it.

I published an episode of This is Surveying recently with Nick Rains, a senior contracts and wording executive at Howden Insurance Brokers Limited . He sits alongside the claims team and advises on policy wordings for architects, surveyors, engineers, and design and build contractors.

Where the claims are actually coming from

Direct negligence claims arising from a surveyor’s own use of AI are not really landing yet. That will come, and everyone is anticipating it.

What is happening now?

“What we are seeing is an increase in the number of claims that have been clearly generated using AI tools. And normally the free AI tools.” Nick

A claimant takes a grievance, puts it into a free model, receives a complete letter of complaint, and sends it off. As Nick put it, the cost to the claimant is close to nothing. After that, it costs the professional and the insurer time and money engaging lawyers to defend it.

None of this requires your firm to have used AI at all.

Most claimants do not know what they have just sent

There is a difference between a complaint and a letter of claim. A complaint is a phone call or an email saying the work looks defective. A letter of claim is a formal step that starts a clock ticking under the pre-action protocol.

People asking a chatbot how to bring a claim against a professional are being handed the formal version, and most don’t realise it. It triggers legal requirements. As Nick puts it,

“They have not really thought about it. They just want to complain. But now there is a proper process that needs to be notified to insurers.”

If the firm treats that letter as a grumble and files it, and the matter later reaches court, the firm is in a weaker position for having failed to comply with a protocol it did not know had been triggered.

The asymmetry

Lawyers spend weeks on a response. Careful, considered, expensive. The claimant receives it, puts it through ChatGPT, and replies within a few hours.

Every line of that reply still has to be analysed. Even where most of it is nonsense, one element might be correct. The cost of defending a claim has moved in one direction.

There is a second problem underneath it. Where the amount claimed is small, it may not clear the policy excess. The firm then faces its own decision about whether to fund a defence at all, or pay to make it go away.

So is AI covered by your professional indemnity policy

Nick’s answer was precise.

“It is not excluded. Which is a funny way of saying it is covered.”

A claim will not be declined at present purely because AI was involved, provided it would ordinarily be covered by the terms of the policy. If AI ran a cost calculation, a member of staff failed to check it properly, the figure went out, and a claim followed, that is a negligence claim, and it is treated as one.

The murky area is authorisation. Where a firm has a policy saying which tools may be used, and an employee works in a free consumer model instead, there is a live question about whether that constitutes use of unauthorised software, and whether it becomes a cyber matter rather than a professional indemnity one. Nick was honest that nobody knows the answer yet.

His practical conclusion was not to ban the tools. Firms that give staff no access find staff using them anyway, without controls and without visibility. A policy that sets out how and when AI may be used is a stronger position than a prohibition nobody follows.

What underwriters are now asking

The question of AI use has become a common item on proposal forms.

“Insurers are really going to want to see how the firm is using AI, what systems they have in place, what governance they have in place, how the output is checked, what is put into it, making sure it is properly supervised. And that will then inform how they write the risk, whether they are willing to write the risk in the first place. Things like pricing.”

Insurers are more likely to look favourably on a firm that can demonstrate good risk management around AI than on one that says it has not developed that yet.

What you can actually prove

The strongest thing Nick said was not about AI at all. It was about evidence.

“It is all well and good having those in your firm verbally. But if you cannot prove them, then it becomes a bit of he said. What can you actually prove.”

His point was simple. When an insurer asks at renewal whether AI use was authorised, whether output was checked, whether a named person supervised it, a policy on the shelf does not answer the question. The record does. Something on the file, written at the time, that you can point to.

A decision held in someone’s head is not evidence. A decision written down when it was made is.

The email on the file is the low-tech version of this, and it works. What it points to is a bigger shift. The record that protects a firm is the one captured at the moment the decision is made, not reconstructed months later when the file has moved on and the memory has gone. Firms that start keeping that record now will have it when they need it. The ones that wait will be trying to build it after the letter has already arrived.

What to do before your next renewal

Nick’s closing advice was about records.

“It is all well and good having those in your firm verbally. But if you cannot prove them, then it becomes a bit of he said. What can you actually prove?”

Three things follow from that.

Write down which tools are permitted and which are not, and tell your teams A short policy that people have actually read beats a long one filed on the server.
Make the human review visible on the file. Nick’s example was deliberately low-tech. An email to a colleague saying I have reviewed this and it went through the following process, sitting on the case management system, is enough to point at later.
Read your next proposal form properly before you complete it. The AI question is no longer a formality, and the answer you give is a disclosure.
The full conversation with Nick is on This Is Surveying.

Thanks for reading! Any feedback or thoughts always welcome