For Legal Secretaries and PAs, AI has become a part of everyday work, drafting letters, completing forms, and preparing documents quickly and efficiently. It's a real time-saver, and used well, it can take a great deal of the routine burden out of document preparation. But it's worth knowing where its limits are: AI can produce a document that looks flawless and still get the details wrong.
Why this happens
Generative AI tools like ChatGPT, Claude and CoPilot can produce documents that look polished and professional whilst still containing errors. This happens because AI predicts what text is likely to come next, rather than retrieving information the way a database does. It gets this right most of the time, but it does not guarantee exactness, so figures, dates or reference numbers can occasionally be altered without warning, especially in longer documents where a detail is repeated more than once.
Numbers are handled differently to ordinary words, and a large figure can be broken down and reconstructed in a way that leaves more room for an error than a short, simple word would. If several similar numbers appear close together, there's also more chance of one being substituted for another. And because generative AI does not have a built-in 'copy' function, even a detail taken directly from your instructions is generated fresh each time it appears, rather than pasted across automatically. Most of the time this works perfectly well, but it means accuracy is never entirely guaranteed.
A document can also read as entirely consistent, the same figure appearing correctly every time it's mentioned, and still be wrong from the very first mention. Generative AI has no way of checking its own output against the original file. It only knows what it has already written, so an early error can be repeated confidently and consistently throughout a document. Consistency is not the same as correctness.
Where the responsibility sits
This is exactly why the responsibility for catching these errors before they cause a problem rests with the person reviewing the document, not the tool that drafted it. AI has not changed that. That responsibility, in practice, often lands with the Legal Secretary or PA doing the final check. It is a role that has not changed, even though the tools have. Reading carefully, checking figures against the file, confirming names and addresses, and cross-referencing dates against instructions is exactly the same discipline that has always defined good legal support work. AI has not removed the need for it. If anything, it has made it more important, because the errors are now wrapped in a level of polish that can make them easy to miss.
AI use in legal drafting has moved from novelty to normal. Recent industry research suggests a large and growing share of law firms now use AI tools for routine drafting and document preparation as standard practice. The Law Society has been equally clear that this comes with real risk, stressing that human oversight is not an optional extra when using AI in legal practice, and that the professional using the tool remains fully responsible for what goes out under the firm's name, regardless of how the first draft was produced. You can read their guidance on generative AI here.
If you would like to build these skills further, including how to review AI-assisted drafts and forms with confidence, ILSPA's AI in Legal Practice course covers this in depth, alongside the practical judgement Legal Secretaries and PAs need to use AI safely and effectively in day-to-day legal work.
