AI and witness statements: clarifying the boundaries of responsible use

Earlier this year the Civil Justice Council (CJC) set up a working group examining the use of AI by legal representatives when preparing court documents. We reported on it here - CJC consultation on AI in litigation | Hill Dickinson

In June 2026 the CJC provided an update on its initial findings. In summary, it found that for legal professionals preparing pleadings, advocacy documents, skeleton arguments and similar documents, no additional formal requirements relating specifically to AI were required. Its view was that professional responsibility frameworks (SRA Guidance etc) were considered sufficient to govern appropriate use of AI in these areas.

There was one area however which was more troublesome: witness statements.

Current law: witness statement preparation

There is currently no specific guidance as to whether generative AI can be used in the preparation of witness statements. However, strong emphasis is placed in the rules on the evidence of the witness being as much as possible in their own words and coming from their personal knowledge. This generally appears to preclude any use of generative AI in drafting them.

The general requirement is that witness statements need to be (if practicable) in the witness’s ’own words’. This requirement is complied with by the words of a witness being transcribed by a lawyer/legal representative. But what is the position if AI is used to transcribe rather than a lawyer? PD 57AC, applicable in the Business and Property Courts (B&PCs) does contain more detailed provisions about the preparation of witness statements. In particular, trial witness statements (covered by PD57AC) prepared with the assistance of legal representatives must also carry a certificate signed by the legal professional verifying the process undertaken.

The use of AI to draft trial witness statements appears to run contrary to the idea that the statement should be in the witness's own words without leading questions having been asked. Current guidance in the White Book points out that the rules ’seem to prevent the use of generative AI, at least to the extent that it suggests any substantive content for a witness statement.’ PD57AC para 3.13 provides that a legal representative may take ’primary responsibility for drafting’ a witness statement, but that the content should be ’taken from, and should not go beyond, the content of the record or notes [obtained from the witness]’. Accordingly, where a legal representative decides to use AI to, say, improve the grammar, or to find a better way of expressing what the witness has said this would appear to go beyond what is permitted by the rules.

One obvious problem is that unless there is a requirement to retain the base information - records/notes/audio of any interview with the witness - there is scope for statements prepared using AI to be inconsistent with that base information and to be potentially misleading.

CJC’s proposal

To avoid the issue of AI producing statements which are misleading or not in the witness's own words, the CJC are proposing a rule requiring a declaration that AI has not been used for the purposes of generating the content of such a statement (including by way of altering, embellishing, strengthening, diluting or rephrasing the witness’s evidence) which would be consistent with the aims of the Practice Direction and reinforce the importance of witness statements being in the witness’s own words. There is to be further consultation on this point.

Criminal proceedings

A separate but related issue has arisen in the context of criminal proceedings. In R -v- FGD [2026] EWCA Crim 918, the Court of Appeal (Criminal Division) has found that a witness's use of AI to prepare for cross-examination at trial can amount to unlawful witness coaching. In that case, the only witness in a prosecution used AI to produce a list of questions about their evidence together with suggested answers before giving oral evidence at the trial. The judge found that the witness had sought and obtained AI assistance to identify the questions likely to be asked of her in cross-examination and how she should answer them. The court concluded that the witness had been coached as a witness by AI and stayed the proceedings as an abuse of process.

Witness training is forbidden in all jurisdictions in both criminal and civil proceedings. In this case the Court of Appeal actually overturned the imposed stay on the basis that the prejudice caused could be dealt with in other ways, but it issued a warning that coaching witnesses on their evidence by any means (including AI) is not allowed.

Although this was a criminal case, it is likely that a similar approach would be taken in civil proceedings, as witness coaching is equally forbidden. Accordingly, witnesses in civil proceedings should be strongly encouraged not to use AI tools when preparing to give evidence in all proceedings. It may appear to be an easy way of preparing to give evidence but the potential consequences are serious. A finding of witness coaching may lead the Court to exclude the witness’s evidence and it could lead to a perfectly valid claim/defence failing in its entirety due to lack of evidence.

Succinctly, Lady Justice Mayin R -v- FGD observed:

It is interesting that the CJC has not considered the issue of witness coaching via AI in its consultation and this case has provided another example of how quickly AI use is becoming embedded in legal practice and the issues that it can throw up.

For further information on the topic of AI use in witness statements, contact us today to discuss how we can help or find out more about our Commercial Dispute Resolution expertise.