Most experts who write for the civil courts already have an AI tool somewhere in the workflow. Usually it is dictation software, a search box, or a spellchecker that has quietly become something more. The question the courts are asking is not whether experts use these tools. It is whether the person reading the report is entitled to know which tools were used, and what they were used for.
In short: the Civil Justice Council has proposed amending Practice Direction 35 so that an expert explains what use has been made of AI beyond transcription and other administrative tasks, and identifies the tools used. The proposal on AI in expert evidence is not yet in force. Nothing in it changes the responsibility that already sits on the expert, which is that every opinion, every reference and every figure in the report has to be one the expert can defend in the witness box.
Where AI Has Actually Entered Expert Work
Four uses are common, and they are not equivalent. Transcription is the oldest and the least controversial, since dictation software has been in clinical practice for decades and has simply become more accurate. Literature searching is the second, where a tool suggests papers the expert then reads. The third is summarising a bundle of medical records, which in a serious injury claim can run to several thousand pages. The fourth is drafting, where a tool produces a first version of a section that the expert then edits.
Treating those four as one activity called AI use is what has made rule-making difficult. The first two rarely touch the opinion. The third can, because a summary decides what the expert reads closely and what they skim past. The fourth can change the opinion outright, because text generated to sound like an expert report will fill any gap with plausible material unless every line is checked against a source.
What the Civil Justice Council Has Proposed
The Civil Justice Council convened a working group on the use of AI for preparing court documents. It published an interim report and consultation on 17 February 2026. The consultation closed on 14 April 2026, and an update on the responses followed on 30 June 2026. A final report is expected later in 2026.
Section 8 of the interim report deals with experts. The proposal is that Practice Direction 35 be amended so that “the expert explains what use of AI has been made other than for transcription (or other administrative uses) and that the expert identifies the AI tools used”. The obligation would sit with the statement of truth at paragraph 3.3 of the Practice Direction, which is where the expert confirms that the opinions given are their own.
Two things follow from the drafting. Transcription and administrative use fall outside it, so dictation does not need to be declared. And the proposal asks for the tool to be named, not merely for the fact of AI use to be admitted. Where the boundary between administrative and substantive use actually falls is not defined, and it is one of the things the final report will have to settle.
None of this is in force. Until the Practice Direction is amended, there is no free-standing rule requiring an expert to declare AI use in a report.
Why Expert Evidence Was Treated Differently From Advocacy
The consultation responses pointed different ways for different documents. For pleadings and advocacy documents drafted by legal representatives, there was broad agreement that no additional formal requirements relating specifically to AI are currently necessary, on the footing that existing professional duties already cover the ground. Witness statements were left open, with the working group still considering whether any form of disclosure would help or simply add friction.
Expert evidence was singled out as raising distinct issues, with the working group describing the aim as proportionate transparency. The reason is structural rather than a comment on experts. A pleading is an argument, and an argument is tested by the other side as a matter of course. An expert’s report is evidence, and where it is the only report on a point it may never be tested at all. If part of the reasoning was generated rather than reasoned, the court has been misled about what it is reading.
What the Statement of Truth Already Requires
It is worth being clear about what has not changed. Under CPR 35.3, an expert’s duty to help the court on matters within their expertise overrides any obligation to the person instructing or paying them. Practice Direction 35 sets out what the report must contain, including the literature and other material relied on in making it. The statement of truth confirms that the opinions expressed are the expert’s own.
An expert who signs that statement over material produced by a tool they have not checked already has a problem, with or without a new rule. The proposed amendment makes an existing responsibility visible on the face of the report rather than creating a new one.
The Fabricated Citation Problem
The risk driving all of this is not subtle. In June 2025 the Divisional Court gave judgment in Ayinde v London Borough of Haringey and Al-Haroun v Qatar National Bank QPSC [2025] EWHC 1383 (Admin), where case citations that did not exist had been placed before the court. Those matters concerned legal representatives rather than experts, but the mechanism is the same one. A general purpose language model asked for authority will produce something that has the shape of authority.
The medical equivalent is a reference to a study that was never published, a guideline attributed to the wrong body, or a complication rate that no source supports. In a clinical negligence claim, an invented complication rate is not a footnote error. It is the opinion.
What the Judicial Guidance Says
The Courts and Tribunals Judiciary published refreshed Artificial Intelligence Judicial Guidance in October 2025. It is written for judicial office holders rather than for experts, but three of its themes read across directly. Public AI tools can produce confident and incorrect material. Private information should not be entered into them. And personal responsibility for anything produced in your name stays with you.
The confidentiality point matters more in medicolegal work than in most other fields. Pasting a claimant’s medical records into a public chatbot puts identifiable patient data into a system the expert does not control. That is a data protection question and a professional one long before it becomes a CPR question, and it applies whichever side has instructed the expert.
Where AI Earns Its Place in a Medicolegal Report
The useful question is not whether a tool was involved but how far it sits from the opinion. The further from the reasoning, the less it matters.
| Task | Effect on the opinion | Position under the proposal as drafted |
|---|---|---|
| Dictating clinic notes or examination findings | None | Administrative, outside the disclosure |
| Spelling, grammar and formatting | None | Administrative, outside the disclosure |
| Finding literature the expert then reads in full | Indirect | Use beyond transcription, so disclosable on a plain reading |
| Summarising a records bundle | Direct, it shapes what gets read | Disclosable, and the tool named |
| Drafting reasoning or opinion text | Direct | Disclosable, and difficult to defend |
Where It Does Not Belong
Three uses are hard to defend in cross-examination. The first is generating the opinion, because the opinion is the thing the expert is instructed to provide and is the only part no one else can supply. The second is generating references, because a citation the expert has not read is a citation the expert cannot answer questions about. The third is generating a prognosis or a range of outcomes, because those depend on the individual in front of the expert rather than on a pattern drawn from text.
A simple test covers all three. If the expert could not reconstruct the passage from the records, the examination and the literature without the tool, it should not be in the report.
What Instructing Solicitors Can Reasonably Ask
The point to ask is at instruction, not at trial. Three questions cover it. Does the expert use AI tools, and for what. What checking process sits between the tool and the signed report. And how are the papers handled, given that most public tools retain what is entered into them.
A useful answer describes a workflow. A claim of no AI use at all from a practice that plainly uses dictation software is worth a second question, because it usually means the expert has not thought about the boundary. The questions matter more where a single joint expert has been instructed, since there is no second report to test the first against. They also sit alongside the wider question of how expert witnesses are regulated, which is where an expert’s professional obligations are enforced in practice.
Recording AI Use Before the Rule Arrives
Waiting for the final report is not the only option. Keeping a short internal note against each report, listing the tools used and what they were used for, costs a few minutes and answers the question years later when the case comes to trial. Experts who already include a brief factual sentence about AI use are not breaching anything, and if the amendment is made in the form proposed they will have nothing to change.
The stronger reason to do it is that a note made at the time is evidence, and a recollection given under cross-examination three years later is not.
Frequently Asked Questions
Does an expert have to disclose AI use in a report now? No. The Civil Justice Council has proposed an amendment to Practice Direction 35 that would require it, but the proposal has not been made into a rule and the final report is still awaited.
Would transcription have to be declared? Not under the proposal as drafted. It carves out transcription and other administrative uses and asks about use beyond that.
Can an AI summary of medical records be relied on? As a working aid, yes. The records are the evidence, and the expert still has to read the parts that matter. A summary that decides what to read closely is a reasonable tool. A summary that replaces reading is not.
What happens if a report contains a reference that does not exist? It undermines the report and the expert who signed it. The courts have taken a firm view of fabricated citations from legal representatives, and there is no reason to expect a softer view of an expert.
Does using AI affect an expert’s independence? Not by itself. Independence is about whose interests the opinion serves. What AI use affects is reliability, which is a different question, and it is the one the transparency proposal is aimed at.
Prof. Hindocha accepts instructions from claimant and defendant solicitors and as a single joint expert. The expert witness page sets out the scope of the work, and papers can be sent through instruct an expert.
Related reading
This article is general information about how artificial intelligence is being treated in expert evidence in England and Wales. It is not legal advice on any particular claim and it is not clinical advice for any individual.