Can Proofreading be Automated?

Why Robots Still Can’t Pass Your Viva

It is 2026. You have ChatGPT-5. You have Grammarly Pro. You have Quillbot.

So, the question isn’t “Can proofreading be automated?” anymore. The question is: “Why are supervisors still rejecting AI-proofread theses?”

I see this every day at Universiti Malaya and USM. A student runs their chapter through an AI tool, fixes all the red lines, and hits submit. Two weeks later, the supervisor returns it with a note: “This feels robotic” or “The flow is illogical.”

Here is the hard truth: AI is excellent at Grammar (fixing commas and typos). AI is terrible at Context (understanding why you wrote that sentence).

An automated tool doesn’t know that your Chapter 2 contradicts your Chapter 4. It doesn’t know that “Proposed Framework” in a Malaysian PhD context means something very specific compared to a generic Master’s essay. It just fixes the syntax.

Worse, if you rely 100% on automation, you risk triggering the Turnitin AI Detector. A perfectly grammatically correct sentence often gets flagged as “100% AI” because it lacks the “burstiness” and messy nuance of human writing.

The Verdict for 2026: Use AI to fix your typos. Use a human to fix your argument. If you want to pass your viva, you need a human editor who understands the difference between “technically correct” and “academically sound.”

Humans currently manually proofread their work, whether academic or non-academic. The proofreading process involves checking and correcting a given text for errors: spelling/typographical, grammatical and punctuation. One must possess a sufficient amount of knowledge on the lexical and syntactical rules of a human language, in order to accurately proofread a text. Using Natural Language Processing and Computational Linguistics, which involve interactions between machines and human language, this task may be automated. This has been attempted, but with a lower accuracy than humans. Syntactic rules can be hand-crafted and applied so that automated programs can understand how to spot and adjust errors. Spelling errors can be corrected and the correct word predicted using spelling prediction programs. This feature is now commonly included in word processing software such as Microsoft Word. Machines can now perform part of speech tagging with 95% accuracy (Brill 1995). This allows for Word Sense Disambiguation. WordNet and other related knowledge bases can also help with POS tagging and predicting the correct word sense of a polysemous word, based on its context and surrounding text. Lexical and syntactic analysis of text enables the generation of parse trees, which are diagrams that show the syntactic relations among words. This can better help to understand the meaning underlying an expression. The syntactic element of linguistics is easier, and machines have performed with good accuracy for these tasks.

The Challenge of Computational Semantics

However, this is not the case for semantics. For true artificial intelligence, and the ability to fully understand the meaning underlying a text, semantics must be considered. Humans have the ability to automatically infer meaning using logic and prior real-world knowledge. This is not feasible for machines, since it would require a massive knowledge base, and would be time-intensive. Computational semantics is only slowly advancing using ontologies and KBs, machine learning and syntactic rules.

For example, in the sentence “We saw her duck.” Based on the surrounding text in the paragraph, a human is able to infer whether “duck” in the sentence refers to the animal she possesses, or whether it refers to the verb “move downwards”. However, this is complex for a machine, and requires processing the surrounding text in order to make an accurate judgement. Therefore, it is these kinds of ambiguous sentences that would create problems in the case of automated proofreading. Semantics and pragmatics require more research, and once these core problems can be addressed, programs that rely on true AI can be applied to automate the task of proofreading.

Why Human Understanding Still Leads the Way

It is very interesting to see how far technology has come since those early studies in the 1990s. We now have tools that can write whole paragraphs in seconds, but the core problem mentioned above is still very much alive. A computer can follow the “rules” of English perfectly, but it still doesn’t truly understand the story you are trying to tell or the specific point your research is trying to make. It can fix a comma, but it can’t always tell if you have used the right technical term for your specific field of study.

In my experience, this is where many students and researchers get into trouble. They trust the “green and red underlines” in their word processor too much. These tools are helpful for catching a quick typo, but they can’t understand the nuance of a complex argument. If a machine changes a word to something that is grammatically correct but factually wrong, it can damage the credibility of your entire paper. This is why having a real person read your work is still so important.

A human proofreader doesn’t just look at the words; they look at the meaning behind them. They use the “prior real-world knowledge” that machines are still trying to figure out. At PM Proofreading, my team and I focus on that deeper level of understanding. We make sure that your “semantics”, the actual meaning of your work, stays clear and accurate, so that your readers understand exactly what you intended. Technology is a great assistant, but for high-stakes academic work, the human touch is still the best safeguard.