Blog
Can AI write the first draft of a witness statement?
Not the evidence itself. AI can build the chronology, gather the document references and tidy the structure, but the account must come from the witness in the witness's own words.
Alesis · · 4 min read
The honest answer is that AI can help with the scaffolding of a witness statement but not with the evidence in it. A witness statement is the witness's own account, in the witness's own words, and it carries a statement of truth signed by that person. Anything that puts words into a witness's mouth, however well drafted, is a problem whether a person wrote it or a machine did.
That still leaves real work an AI assistant can take off a fee earner's desk: assembling the chronology from the papers, finding which page supports which fact, checking the statement against the documents, and flagging where the account and the file do not match.
What a witness statement has to be
The Civil Procedure Rules and the accompanying practice directions set out how statements are to be prepared and presented, including the requirement for the statement of truth and the expectation that the statement is in the witness's own language. Courts have made clear over the years that statements drafted as advocacy, full of argument and phrases the witness would never use, are unhelpful and can damage the case.
So the test for any drafting tool is simple. Does it help the witness say what they actually remember, more clearly and in the right order? Or does it produce a polished narrative that nobody has verified against the witness's own recollection?
The second is the failure mode to guard against. A fluent draft is persuasive to read, and a busy witness signing it may nod along rather than correct it. The risk is not that the drafting is bad. The risk is that it is good enough to go unchallenged.
What AI can usefully do
Used on the matter file rather than on the witness's memory, an AI assistant can carry a fair amount of the preparation:
- Build a chronology from the correspondence, attendance notes and contemporaneous documents, with a page reference for each entry.
- Pull together every document that touches a particular meeting, payment or conversation, so the fee earner walks into the interview knowing what exists.
- List the questions the papers raise: gaps in the sequence, dates that do not line up, a letter that refers to an attachment nobody can find.
- Check a completed draft against the disclosure and identify statements that no document supports, so the fee earner can go back and ask the witness what they are relying on.
- Produce a clean structure with numbered paragraphs and consistent defined terms, once the content is settled.
None of that is evidence. All of it is preparation that would otherwise take a paralegal a day and often gets skipped.
What AI must not do
Do not ask a tool to draft a witness's account of events from the documents alone. The output will read like a witness statement, but it is an inference from paperwork dressed up as memory. A witness who signs it is attesting to something they have not actually said.
Do not use AI to smooth a witness's language into professional register. If the witness says the delivery turned up late and the site was a mess, that is what the statement should say. Tidying it into commercial prose removes the very thing that makes the statement credible.
Do not let a tool fill a gap. If the witness cannot remember whether a call happened before or after the site visit, the statement should say so. Generative tools are built to produce continuous text, and continuous text has no natural place for "I do not recall". That has to be put back in deliberately.
And do not paste a witness's personal account into any tool the firm has not checked for confidentiality and data handling. Witness statements often contain health information, family detail and third party names.
A process that works
- Take the interview yourself, or have a supervised fee earner take it. Record or note it properly.
- Use AI beforehand on the file to build the chronology and the document map, so the interview is well directed.
- Draft from the interview note, not from the documents.
- Use AI afterwards to check the draft against the papers: does each factual assertion have a page behind it, and are there documents that contradict it?
- Take the draft back to the witness. Read it with them. Change what they want changed, including the wording.
- Record on the file what the tool was used for and who checked the result.
The supervising fee earner owns the statement either way. Regulatory obligations to the client and to the court do not soften because part of the preparation was automated, and the person signing off is answerable for what goes to the other side and to the court.
Where Alesis fits
Alesis answers questions about a matter from the matter's own papers and names the page each answer came from; if the papers do not say, it says so. Documents are read page by page, so citations point at pages, and any page it could not read is flagged rather than skipped. It prepares drafts for a qualified person to review and sign off; it does not file, serve or email anything. It assists qualified professionals and does not replace them, and it does not provide legal advice.