AI POLICY | L. E. Proofreading
Last updated: 15 May 2026
The use of AI, and specifically language models (LMs), has some superficial benefits for proofreading. LMs follow predictable patterns, so are good at identifying clear spelling errors and simple grammatical mistakes. In some cases (e.g., writing an email), this is appropriate but in more complex writing, proofreading requires more than pattern recognition.
LMs are prone to producing fluent-seeming text in which the words are spelt correctly and the grammar follows the rules, but the sentences themselves are meaningless. LMs are also limited. For example, they will want to “fix” errors that don’t exist or fail to maintain consistency in terminology across large documents. Most importantly, LMs are apt to drown out the author’s voice.
For these reasons, as a point of principle, I never use LMs to do any proofreading of a client's work. More specifically:
I do not use AI to check a client’s work. I do the proofreading and copy-editing myself because: (a) I enjoy it and (b) I think human eyes and perceptions still have the edge over AI language models.
I do not expose any client’s work to an AI. A lot of clients trust me with their original research, and I will always protect and respect the confidentiality of someone’s work and ideas.
I would not knowingly proofread a document that has been written by an AI. I might be asked to do this for example when a supervisor or a journal has rejected a piece of work because some or all of it was AI-generated and the client wants me to “fix” it.
Having said that, as part of my wider proofreading process, I might use LMs in the following situations:
for help in checking that I have understood why an author is using a particular statistical method, even if I might not understand how they’re using it;
to produce an example of how something works so that I can check that I have understood a broad concept in a piece of work I’m proofreading; or
to clarify units of measurement (e.g., which units are applied to mass or which to concentrations of the same substance. Not always easy when you’re not a subject expert!).
My strength is that I give authors the space to make decisions about content and vocabulary. I want their work to have their voice; not to sound as if it has been machine-generated.
I will review my AI policy regularly and update it as necessary.