The AI question. The bigger risk isn't the technology. It's pretending it isn't here.

According to the 2025 Bond Solon Expert Witness Survey, the largest expert witness survey in the UK, AI use among expert witnesses doubled in twelve months. From 9% to 20%. The same survey asked whether specific guidance was needed. 89% said yes.

The technology is moving faster than the rules. The experts are saying so out loud.

The two conversations

An examiner, talking to me about how their day actually works. Document management. Note-taking. Dictation. AI tools across their clinical practice. For them, it was normal workflow, and has been for a while. Time is the most valuable thing an examiner has, and the right tools give some of it back.

A senior lawyer, in a different conversation. They love what AI can do. They're cautious about how to use it. They know about the NSW Supreme Court Practice Note SC Gen 23. They've read the horror stories, the lawyers who submitted AI-generated arguments without checking the references, the fabricated cases, the hearings where it all unravelled.

Two reactions to the same technology, from two ends of the same matter. Neither is wrong. Both are happening today.

Where the line already sits

The survey put one question to its 525 respondents worth reading carefully: would they accept an instruction where the solicitor insisted on providing a draft expert report generated by AI? 86% said no.

The number isn't really about AI. It's about expert independence. The answer would have been 86% no even if the draft had been written by the solicitor themselves. The line experts protect is that the report is the expert's own, the opinion, the structure, the words. AI didn't change the rule. The survey just made it worth asking again.

The view from the bench

In June 2025, Dame Victoria Sharp, President of the King's Bench Division, acknowledged AI was likely to have "a continuing and important role in the conduct of litigation", but warned its use required proper oversight and compliance with professional and ethical standards if public confidence in the administration of justice was to be maintained.

Her judgment referenced two cases where fake citations were submitted because of careless AI use. Then there's Kohls v Ellison, a US federal case from January 2025: an expert instructed to give evidence on AI-generated deepfakes used generative AI to help draft their report and accidentally submitted misinformation to the court. An expert on deepfakes, undone by the technology they were instructed to opine on.

The risk worth naming

There's a version of this conversation where the industry argues about whether to permit AI. That argument is over. The data settles it.

When an industry pretends a technology isn't happening, the bad uses fill the silence first. The unverified citations. The reports drafted without review. The corners cut without anyone setting a rule against cutting them. The standards come after the failures, instead of before.

What good use already looks like

From the same survey, an expert describing how they actually use AI: "Checking a statement I have made is clear and unambiguous, obviously without providing any identifying details of the case. I never use AI without fact checking information."

Three patterns sit inside that one quote. The opinion stays the expert's. The use is disclosed where it matters. The output is fact-checked before it leaves the desk. That's the standard, and it isn't new. It's the same standard that's always applied to junior drafting, paralegal work, and clinical support. The standard doesn't change because the support is software.

The question to ask your provider

Some workflows were built to be visible, designed with transparency in mind. Others weren't, they grew up around small panels and high-touch operations, where the value was the relationship and the brief, not the visibility of the process. That isn't a moral judgement. It's a structural fact about how different providers were built.

If you can't get a clear answer from your provider on where AI sits in their work, that isn't necessarily bad faith. But it is a sign the workflow wasn't designed for the questions the next few years are going to ask. Senior practitioners reading this can already think of providers they'd struggle to get a straight answer from. That's the point.

Where this is in three years

My read: within three years, an IME provider that can't tell you exactly what AI touched in a report will be unbookable for senior matters. Every provider will be using AI in some form. The differentiator won't be whether, it will be whether they can show you what AI touched and what it didn't, whether the use is disclosed, and whether a human has signed off in a way that holds under cross.

The lawyers using AI for evidence review will outpace the lawyers who don't. That gap is already opening.

If you can't get a clear answer from your IME provider on where AI sits in their workflow, the answer is the answer.

Survey data drawn from the Bond Solon Expert Witness Survey 2024 and 2025, published in association with the Law Society Gazette.