The Dangers of Using AI to Assist with Witness Statements: Lessons from Godwin v Godwin

Introduction

In this legal article, I revisit an issue that I often discuss with lawyers in the UK: how far AI can properly assist in the production of witness statements. As one would expect, views on this issue vary across jurisdictions. For readers who are new to my legal articles, I have previously considered this issue and discussed some interesting observations from the Family Court and the Employment Tribunal here.

As I often say in public forums, the AI issues that arise in one area of law often also arise in others. This case is a further example of how the use of AI may significantly affect the court’s consideration of evidence, particularly where there is uncertainty about whether a witness statement remains in the witness’s own words.

Godwin v Godwin (HHJ Klein)

[2026] EWHC 923

This was a judgment of the High Court of Justice, Business and Property Courts in Leeds. The case concerned the disposition of a body after death. It is important context that both the Claimant and Defendant were litigants in person. The full judgment can be read through the hyperlink above.

The important passages for the purposes of this legal article, insofar as they relate to artificial intelligence, begin at paragraph 37:

“For the reasons I give in this section of the judgment, I have had to approach the evidence of the parties cautiously.”

The Judge explained the court’s previous direction relating to witness evidence:

42. By paragraphs 6 and 28 to 34 of District Judge Bond’s order, the witness statements for use at the hearing have had to comply with her order, in the following ways, amongst others:

i) they have had to be prepared in accordance with CPR Pt.32 and PD57AC;

ii) they have had to contain a complete statement of truth in proper form;

iii) they have had to include a signed certificate of compliance in the form required by paragraph 4.1 of PD57AC, which the District Judge set out in full in the order.

43. As the District Judge’s order also made clear, at paragraph 35, any non-compliant witness statements have been at risk of being excluded from the evidence or of having attributed to them reduced weight.

The Judge then turned to the evidence:

44. The Defendant and [Redacted], his former wife who gave evidence for him, explained in their witness statements that they have had “limited assistance from a “digital assistant” for grammar, spelling, and “presentation”. By “digital assistant”, they meant an AI tool; ChatGPT. They claimed that they each prepared the first draft of their respective witness statements, that they each uploaded their own draft to ChatGPT and that ChatGPT’s product did not add, remove or re-arrange any words when compared to their first drafts.

45. There was no good reason for either the Defendant or [redacted] to use ChatGPT in this case. Both struck me as sophisticated people. The Defendant is a key account manager who travels in Europe making presentations for a company in the lift industry, sometimes to large businesses. Both the Defendant and [redacted] are sufficiently technically adept to have used ChatGPT. They must have been able to use the spelling and grammar checking facility in Word (which the Defendant used to prepare his first draft) and in whatever word processing programme [redacted] used to prepare her first draft. Also, the District Judge set out in detail, in paragraph 30 of her order, how the witness statements should be structured.

46. It is not conventional, in spoken English, to refer to an AI tool, as “a digital assistant”. I did not receive a satisfactory explanation from the Defendant about why, on his case, he chose that phrase. I cannot rule out the possibility that that is a phrase ChatGPT contributed to the witness statements. Nor can I rule out the possibility that the Defendant might have been trying to obscure his use of AI in the production of his witness statement.

47. I have not been provided with the first drafts.

48. For all these reasons, I cannot be sure that the Defendant’s, or [redacted], witness statements are in their own words, reinforcing my decision to approach the Defendant’s evidence cautiously and requiring me to approach [redacted] evidence cautiously too.

CJC The Use of AI for Preparing Court Documents Consultation (17/02/2026 – 14/04/2026)

At this stage, it is worth reminding readers of the Civil Justice Council’s eight-week consultation, which closed on 14 April 2026, on the use of AI in preparing court documents. That consultation included the following discussion points:

“Witness Statements

8. The CJC makes different proposals for different kinds of witness statements, in particular drawing a distinction between trial witness statement and non-trial witness statements. Do you agree with that approach? What distinction if any would you propose?

9. In relation to non-trial witness statements, the proposal is that provided the statement bears the name (or firm name?) of the legal representative who is taking professional responsibility for its preparation, there is no need for any (further) rules relating to these documents produced with the assistance of AI. Do you agree? If not, why not?

10. An alternative would be to require a specific declaration to make clear if the legal representative has used AI in the preparation of non-trial witness statements. Do you prefer this alternative? If so, please explain why and consider which uses of AI ought to be covered.

11. In relation to witness statements covered by PD57AC and within the scope of this paper (i.e. trial witness statements prepared with the involvement of a legal representative), the proposal is that there be a rule requiring a declaration on such a statement that AI has not been used for the purposes of generating its content (including by way of altering, embellishing, strengthening, diluting or rephrasing the witness’s evidence). Do you agree? If not, why not?

12. In relation to witness statements under CPR Part 32, not covered by PD57AC but within the scope of this paper (i.e. trial witness statements prepared with the involvement of a legal representative), the proposal is that there be a rule requiring a declaration on such a statement that AI has not been used for the purposes of generating its content (including by way of altering, embellishing, strengthening or diluting or rephrasing the witness’s evidence). Do you agree? If not, why not?

13. In relation to witness statements involving translation, one issue relates to use of AI by translators. Should there be a rule making provision for the use of AI by human translators? If a translator is prepared to sign a statement of accuracy, taking responsibility for it, is there any need to enquire further? A further proposal is to permit the use of publicly available machine translation, provided the tool used is identified, and provided (if necessary) that provision is made clarifying that other parties are entitled to check the translation themselves by using such a tool. Do you agree? If not, why not? Do you favour the alternative below? If so why

14. An alternative to the previous proposal would be only to permit such use by a legal representative and to require that the legal representative involved in the preparation of the translation should identify what tool has been used. Do you favour this alternative?”
 

Comment

This is a timely judgment in light of the Civil Justice Council’s consultation. I look forward to reading the final report on that consultation once it is available.

In this case, the use of artificial intelligence led the Judge to approach the relevant evidence cautiously. The Judge took the view that there was no good reason for ChatGPT to have been used. The court was concerned that the witness evidence before it may not have been in the witnesses’ own words.

I have discussed this important feature of witness evidence at some length in a previous legal article, where I considered judicial observations from the Family Court and the Employment Tribunal. In that article, I set out some competing arguments about whether this requirement should continue in its current form or whether it may need to be revisited in light of what we now know about AI.

“Some argue there is no place for AI, since a witness statement should capture the person’s voice as if they were speaking. The requirements use mandatory language such as “must”, not the softer “should”. And as readers of my legal articles know, certain AI systems can be sycophantic, misleading, or prone to producing false information even without being asked to. If AI-generated witness statements became commonplace, those who wrote their statements genuinely in their own words might be at a disadvantage if an AI-produced statement came across as especially persuasive. How could that help a judge arrive at the truth of any issue and deliver a just outcome?

Others take the opposite view, suggesting this is an outdated concern. AI is already woven into everyday life. Even opening a word processor to draft a witness statement involves autocorrect, predictive text, synonyms, and suggested phrasing. Is that really so different from what lawyers already do when helping a client to complete a statement? On top of that, AI could play an important role in supporting people with learning difficulties or limited education to express their perspective more clearly, which could in fact assist the court. Current AI guidance does not expressly address its use in witness evidence.”

Since writing that article, and discussing this many times with others, my concerns have become increasingly focused on enforceability and the risk of satellite litigation. The reason is practical. I am becoming progressively less convinced that either current technology or human assessment can reliably identify whether AI has been used in a legal document and to what extent. Unless there is an admission, a clear audit trail, or some obvious sign of AI use, can the courts and parties realistically identify prohibited AI use?

That is not a reason to abandon the principle that witness evidence should be the witness’s own evidence. It is, however, a reason to think carefully about how any future rules are framed, how they are evidenced and how they can be applied fairly without creating disproportionate procedural disputes.

I will shortly be writing in more detail about legal education and the related problem of identifying the use of AI in essays and other assessments. I will return to these practical concerns then.