Your Witness, Or The Chatbot's? R V FGD On AI And The Prohibition Of Witness Coaching

Key Takeaways

• Witnesses should not use AI in preparing to give evidence in proceedings, whether that evidence is to be given orally or in writing.
• Witness coaching is prohibited regardless of means and this includes using an LLM to work on responses to cross-examination questions.
• The integrity of a witness’s evidence must be preserved. AI tools suggesting “stronger” answers risk producing evidence that is not the witness's genuine recollection, undermining the court's fact-finding function and ability to dispense justice.
• Courts are becoming increasingly vigilant of AI use in legal proceedings, both in terms of preparing court documents and in proceedings more broadly. Practitioners should consider offering early warnings about AI use as a matter of course in order to protect the integrity of witness evidence.

Background

English law draws a sharp line between (permissible) witness familiarisation and (prohibited) witness coaching. Witness familiarisation is the process of preparing a witness for the experience of giving evidence in court, by explaining the general process of giving evidence and the roles of those they will encounter in the courtroom. It does not touch on the substance of a witness’s evidence or the specific topic(s) in dispute. Witness coaching, which is prohibited, involves a witness being “coached” or trained on the specific topics in dispute including the answers they might provide in cross-examination.

R v Momodou and Limani [2005] EWCA Crim 177 (“Momodou”) established that witness coaching is prohibited. Momodou is equally applicable in civil and criminal proceedings and seeks to address the harm caused by contaminating a witness’s recollection.

R v FGD

In a recent decision in the criminal court, R v FGD restated the rule that witness coaching is impermissible, including where such coaching appears in the form of a conversation with an AI Chatbot. R v FGD concerned AI-generated documents, produced at the witness's own prompting, which summarised their account and worked through anticipated cross-examination questions with suggested answers in a supportive, coach-like tone.

At first instance, the Judge held that this material amounted to coaching “whether or not done by AI or someone else” applying the Momodou principle, and stayed the proceedings as an abuse of process on the basis that a fair trial was no longer possible.

On appeal, the Court did not disturb the coaching characterisation (indeed, neither party argued the AI documents fell outside forbidden coaching) but held the first instance judge erred in finding no safeguard (such as disclosure, jury directions, or recalling the witness) could cure the prejudice. This was particularly so in circumstances where the witness had given prior evidence, at the time of the incident and years before the prohibited coaching, and there was a clear documentary record of the exchanges with the AI coach. A stay of proceedings to prevent an abuse of process ought to be exercised only in exceptional circumstances and is only appropriate when a fair trial is deemed to not be possible. As such, it is a last resort in civil and criminal proceedings. 

While the Court allowed the appeal, it gave a clear warning that coaching witnesses by any means is not permitted and witnesses should be firmly discouraged from using AI to prepare to give evidence, as doing so could risk the evidence being excluded. This is particularly pertinent in a world where AI use is becoming more prevalent and, as the Court noted, people are using AI for all types of everyday tasks (including but not limited to job applications, interviews, and research).

The Court also noted that the Civil Justice Council (“CJC”) has been consulting on AI transparency requirements in civil proceedings, including witness statement preparation. Only an interim report has been published so far, which found that many respondents were concerned AI tools could embellish or influence evidence in ways that may not be apparent; however, the CJC Working Group considered those concerns not to be unique to AI and simply reflect existing principles governing the preparation of evidence. In a June 2026 update, the CJC Working Group reported strong consensus amongst respondents that, given existing professional responsibilities (e.g., under the SRA Code of Conduct for Solicitors), no additional formal requirements on AI use in statements of case are needed, though preserving a witness's authenticity, integrity and personal recollection remains important. The CJC Working Group expects to publish its final report later this year.

What does this mean for civil litigation?

Whether in civil or criminal proceedings the integrity of a witness’s evidence must be preserved: it must remain their own account of events uncontaminated by external influence. However, this is not a new concept. It has always been an essential requirement of reliable evidence and was the driving force behind the changes that were introduced in 2021 when PD57AC on trial witness statements came into force, which expressly states that “it is improper to put pressure of any kind on a witness to give anything other than their own account, to the best of their ability and recollection.” The Practice Direction requires that statements avoid, so far as possible, any practice that might alter or influence a witness’s recollection. To that end, the Statement of Best Practice supporting PD57AC reiterates the risk of perverting human memory in a way that impacts a witness’s recollection. We expect the outcome of the CJC consultation to echo the same requirements in the context of the use of AI, where AI is precluded from drafting and improving witness statements given it would necessarily not be in the witness’s own words. The CJC Working Group has floated the possibility of recommending the introduction of a formal declaration from legal representatives that AI has not been used in generating the content of a witness statement.

This risk of contaminated evidence resulting from AI witness coaching (or any other prohibited witness coaching), is not merely one of procedural unfairness. An honest witness may unconsciously align their account with a “stronger” version of events, given the vulnerability of memory over time and its ability to be reshaped by exposure to other accounts. More concerning though is that a dishonest witness may consciously optimise their evidence. Either way there is a risk of introducing evidence that is no longer the witness's recollection. This case is not the first of 2026 to deal with the dangers of witness testimony being influenced by technology: in UAB Business Enterprise and another v Oneta Ltd and others [2026] EWHC 543 (Ch), the High Court rejected one witness’s evidence in its entirety upon finding that he had used smart glasses while giving testimony.

The Bar Society has also issued guidance on witness preparation, restating the prohibition of witness coaching and, conversely, the appropriateness of independent witness familiarisation training to ensure witnesses are familiar with the basic requirements of giving evidence so that they can best assist the court, through engaging independent counsel or a company that specialises in witness familiarisation.

Comment

AI use did not prove fatal to the proceedings here because there were matters in the fact pattern (including the fact that the witness had previously given evidence on the same matters shortly after the incident happened and years before the trial and the AI use) which allowed the prejudice to be cured, but that will not always be the case. Practitioners must be vigilant and proactive in flagging to clients the serious consequences AI use can have for witness evidence and for a claim as a whole. Failing to do so risks witnesses using AI improperly and unwittingly derailing proceedings. That warning only grows more urgent as AI becomes ever more embedded in daily life.

Witness coaching is not the only area in which AI can have serious adverse consequences in litigation. See for example Ask AI, Lose Privilege? Courts Draw Lines on AI-Generated Legal Materials, where Dechert examined how AI use led to a waiver of attorney-client privilege in US proceedings. Lawyers and clients alike should consider setting clear parameters for AI use in order to manage these risks.

The authors would like to thank Lauren Johncock for her contributions to this OnPoint.