Word for Word: Can You Still Use AI on a Witness Statement?

The Civil Justice Council wants every witness of fact to promise that AI hasn’t shaped their evidence. Like every other promise in a witness statement, it will be tested in court.

On 29 October, the Civil Justice Council published its final report on AI and court documents, and one recommendation reaches every disputes team. Witnesses of fact will declare that AI has not generated or influenced the substantive content of their statement. The witness signs, not the lawyer.

That declaration won’t stay on the page. In cross-examination, counsel won’t need to ask whether the statement is true. They can ask who wrote it. Our view: the teams who can show, line by line, that the words are the witness’s own will win that exchange.

So can you still use AI? Yes, but not on the evidence, and “influence” reaches further than most people expect. None of it is in force yet. The Civil Procedure Rule Committee has to act first, which gives teams time to sort out how statements are prepared before anyone has to swear to it.

The witness signs, not the lawyer

In February, the CJC’s interim report floated a declaration from the lawyer, on trial statements only. The final report moves it to the witness and widens it: any witness of fact, trial or not, represented or not (Recommendation 6; para 7.15).

It’s the right principle. The evidence is the witness’s, so the promise should be too. The difficulty is that the witness knows least about how the statement was made. Most witnesses meet their statement the way the rest of us meet a group photo: after everyone else has finished with it. If they are going to sign a declaration about the process, someone has to be able to tell them what the process was.

Which leaves a question the report doesn’t answer: what happens when a witness signs in good faith, and it later emerges that AI was used on their words before they ever saw them? The existing rules reserve contempt for false statements made without an honest belief in their truth (CPR 32.14). For trial statements in the Business and Property Courts, PD 57AC already requires a certificate of compliance from the relevant legal representative. Our reading: the witness signs, the firm carries the risk, and the case carries the cost.

“Influence” goes further than drafting

Most readers will take the declaration to mean one thing: don’t let AI write the witness’s words. Fair. But the report defines influence as AI that affects how material is selected, analysed, developed or evaluated, even if it never writes a word (para 2.6).

That reaches the ordinary work of preparing a statement: choosing the topics, sorting interview notes into themes, cutting a long account down. PD 57AC already warns against leading questions. Nobody drafting it expected the leading question to come from autocomplete.

The report doesn’t rule on each task, but it does settle the test. The CJC deliberately declines to sort AI tools into categories such as generative or administrative (Recommendation 11). What matters is what happened to the evidence, not which product was used (para 2.7). We agree, and it’s a fair rule for anyone selling software to litigators, ourselves included: no tool makes a statement acceptable on its own.

Where the line falls

How the report treats the tasks disputes teams actually do. A guide to the report, not legal advice.

‍

Godwin got there first

The courts aren’t waiting for the rules. In Godwin v Godwin [2026] EWHC 923 (Ch), the defendant and his former wife had used ChatGPT on their witness statements, for grammar and presentation, they said. The judge found there was no good reason for either of them to use ChatGPT, and approached their evidence with caution because he couldn’t be sure the words were theirs.

Nobody had to prove the AI changed anything. Not being able to rule it out was enough. For a disputes team, that’s the real cost: evidence the judge gives less weight to.

What we’d do now

• Keep the recording. If a transcript contains words nobody said, the only witness to them is the software.
• Draw the line between preparing a statement and writing it, and make sure the trainee knows where it sits.
• Be ready to explain the process to the witness, in plain English, before they sign.
• Brief witnesses early not to run their account through a chatbot. “It only fixed the grammar” now has case law attached.
• Watch the regulators. The report leaves professional guidance to the SRA, the BSB and the other regulators, alongside a new Practice Direction (Recommendations 1 and 3).

AI you can stand behind in front of a judge

TrialView is built for disputes, from first bundle to final hearing, so our test for AI is simple: would you rely on it in front of a judge? That means every answer has to point to its source.

• When TrialView’s AI answers a question about the documents in a case, it shows where in each document the answer comes from and where that document sits in the bundle.
• Witness statements can be cross-referenced to the documents they rely on, with margin references and hyperlinks to the exhibits in the bundle. Where a statement cites disclosure IDs, TrialView can link them to the documents automatically.
• Ask AI can test a witness statement against the rest of the evidence, finding documents that are inconsistent with what the witness says, each with its bundle reference.
• At trial, AI on the real-time transcript can flag where a witness’s evidence is inconsistent with their earlier testimony or the documents. It’s being rolled out now, and what it flags stays private to your team.

The question has moved from “did you use AI?” to “can you show the words are still theirs?” Our view: AI belongs on the evidence around a statement, finding it, checking it and testing it. The words stay the witness’s own.

Book a demo

Quick answers

This article summarises the Civil Justice Council’s recommendations and isn’t legal advice. Source: Civil Justice Council, Use of AI for Preparing Court Documents: Final Report, 29 October 2026. Read the full report.

‍

‍