Pros and cons of using AI for copyediting

Dr Denis Savelyev · October 2026 · 7 min read

AI writing tools can catch typos in seconds and rephrase a clumsy sentence on request. So do you still need a human editor? An honest look at what AI does well, where it falls short, and what thesis writers in particular need to watch for.

I am genuinely interested in artificial intelligence and in how it is changing higher education. I follow developments closely, I discuss them with colleagues and students, and I think pretending they do not exist is unhelpful. So this is not an anti-AI post. It is an attempt to set out, as fairly as I can, what AI tools can and cannot do when it comes to editing academic writing.

What we mean by “AI editing”

The term covers several quite different things:

• Spelling and grammar checkers built into Word or Google Docs, now increasingly AI-powered.
• Dedicated writing assistants that suggest rewordings, flag passive voice or rate readability.
• Consistency-checking software that scans a document for variant spellings, hyphenation, capitalisation and abbreviations. (Many professional editors use tools like these themselves.)
• Large language models (LLMs) such as ChatGPT, Claude, Gemini or Copilot, which can be asked to “edit”, “polish” or “improve” a passage and will return a rewritten version.

The first three are tools that help a writer or editor check things. The fourth is different in kind: it generates new text. Most of the benefits and most of the risks discussed below come from that difference.

The pros

Speed and availability

AI tools are fast, cheap or free, and available at two in the morning when your chapter is due at nine. For a quick check of an email, an abstract draft or a conference proposal, that is hard to beat.

Good at surface errors

Modern tools are very good at catching typos, doubled words, subject–verb disagreement and many punctuation slips. Running a spellcheck and a grammar check before sending your work to anyone, human editor included, is simply good practice.

Helpful for writers working in an additional language

For researchers writing in English as a second or third language, AI tools can offer alternative phrasings, explain why a construction sounds unnatural and build confidence. Used as a tutor rather than a ghostwriter, they can be genuinely educational.

The cons

It does not understand your argument

An LLM predicts plausible text; it does not know what you are trying to say. When it “improves” a sentence, it may subtly change the meaning: turning a cautious claim into a confident one, swapping a precise technical term for a more common near-synonym, or smoothing over a deliberate distinction. In the humanities and social sciences, where so much depends on careful wording and nuance, these small shifts can matter a great deal. They are also very easy to miss, because the new sentence reads fluently.

Your voice gets flattened

AI-polished prose has a recognisable sameness: tidy, confident, slightly generic. Examiners and reviewers increasingly notice it. Your writing should sound like you, a scholar with a particular perspective, not like an averaged-out version of the internet.

Invented facts and references

LLMs can and do fabricate citations, dates, quotations and page numbers that look entirely convincing. If you ask an AI tool to “tidy up” your reference list, check every entry against the original source. A human editor who finds a reference that does not match will query it; an AI tool may silently “correct” it into something that never existed.

Confidentiality and intellectual property

Unpublished research is valuable and sometimes sensitive: ethics-approved interview data, commercially confidential findings, or simply ideas you have not yet published. Before pasting it into an online AI service, read the terms. Some services may store or use what you enter, and your university, ethics committee or funding body may have rules about where research data can go.

Academic integrity rules

This is the most important point for research students. Universities now have policies on the use of generative AI, and they vary widely between institutions, faculties and even individual units. Some permit AI for spelling and grammar but not for rewriting; some require any AI use to be declared; some prohibit it for assessed work. A thesis is assessed work. Using an LLM to rewrite sections of it could, depending on your university’s policy, raise the same concerns as having someone else write it.

Note too that IPEd’s Guidelines for editing theses (2025) make clear that the student remains responsible for the integrity of their thesis, including any AI-generated content. If in doubt, ask your supervisor and check your university’s policy before you use these tools, not after.

So where does a human editor fit?

The honest answer is that AI tools and human editors do different jobs. A good human editor brings things software cannot:

• Judgement. Knowing when a “rule” should be broken, when a long sentence is doing important work, and when repetition is deliberate.
• Subject knowledge. Recognising that a term is used in a specific technical sense in your field, or that a transliteration convention matters.
• Ethical boundaries. A professional thesis editor works within your university’s rules and the IPEd Guidelines, and will tell you when a request falls outside them.
• Whole-document consistency. One style sheet across the whole manuscript, which AI tools working passage by passage cannot reliably maintain.

A sensible way to combine them

If you want to use AI tools responsibly in your writing process, here is a practical approach:

• Check your university’s (or journal’s, or publisher’s) AI policy first, and keep a note of how you used any tools.
• Use spelling, grammar and consistency checkers freely; they are low-risk and save everyone time.
• Use LLMs, if at all, to explain rather than to rewrite: ask why a sentence is unclear, then fix it yourself.
• Never accept AI changes to citations, quotations, data or technical terms without checking them against the source.
• Do not paste confidential or unpublished data into online tools without understanding where it goes.
• Bring in a human editor for the final, whole-document copyedit and proofread, and tell them what tools you have used.

My own approach

For transparency: I do not put clients’ manuscripts into generative AI tools. I may use established, non-generative checking software (for example, to flag inconsistent spellings) to support my own reading, but every editorial decision is mine, every change is visible to you in Track Changes, and your work stays confidential. If you would like me to avoid any software at all, just say so.

AI is a useful assistant. It is not yet a substitute for a careful reader who understands what you are trying to say and why it matters.

Questions about editing and AI for your thesis, article or book? Get in touch.