How to recognise AI generated academic writing

AI has disadvantages for research and writing.  Even so, like lots of other people, researchers and students have adopted AI to save them time, including for research and writing. AI offers marvellous benefits for medical interventions and a huge variety of everyday uses. But the way it finds information is problematic.  Tracey Spicer has written about the many issues with AI, including its error filled so-called factfinding, in her book Man-Made.

This time we are looking at some very obvious issues when AI generates academic writing. University teaching staff have been struggling to find ways to persuade students to show responsibility when they use AI in their assignments. Innovative ideas include designing assignments where students investigate the ways they can use AI and critically assess what is helpful or unhelpful about this. Some university administrations have developed a policy so that students acknowledge their use of AI. At some universities this means that students must complete a formal declaration state for each assignment stating whether they used AI and if so, how they used it.

Since 2023, I have seen many examples of AI use in writing at both undergraduate and postgraduate levels. Whenever some writing tends not to make much sense but is written in perfectly correct grammar, it could have been AI generated. Here are some very common examples:

Certain words have emerged as favourites

“Role”, “crucial”, “critical”, “vital”, “need” and “emphasise” are very frequent. For example:

It is crucial (or critical) to investigate…

The need for … is emphasised.

Sometimes these occur in silly combinations which say nothing but might seem important:

This research emphasises the need for further investigation into … which is crucial.

Sentences which conclude paragraphs are vague

Concluding sentences in paragraphs tend to extend the significance of the information provided in the paragraph by attempting to sum it up and stress its importance more generally. These generalizations lack precision. This is an example:

A leader’s role in an organization is crucial.

This is an obvious point. In any organization a leader tends to have some kind of influence. That is what it means to be a leader. Even if the leader is ineffective, this affects the organization.  As a summary of a paragraph, this sentence adds nothing.

Ideas are repeated

There is frequent repetition of the same point in different words, like this:

Autonomy-oriented helping behaviours can significantly affect the performance of team members (Zhu, Zhang, & Guo, 2021). In a team environment, a leader’s self-directed helping behaviour significantly affects team members’ performance (Zhu, Zhang, & Guo, 2021).

Here the second sentence includes the words “a leader’s self-directed behaviour” which is similar in meaning to “autonomy-oriented helping behaviours”. It is somewhat confusing as we can’t immediately understand if the “self” here refers to the leader who helps the team members or the team members who are being helped. (Actually the idea is that a leader gives team members help but the team members have some autonomy in the process). The rest of the two sentences are basically the same. So the second sentence is a poor paraphrase of the first sentence.

There can also be repetition in the same sentence. This means the sentence can take the form of a circular argument, where the last part states the same thing as the first part, like this:

The combined approach of psychology and geography adopted by He, Blye, and Halpenny (2023) in their study of environmental issues in nature-based tourism settings emphasises the importance of considering psychological and geographical aspects in the search for solutions.

You can see here that the first part links to the second part with the words “emphasises the importance of”. This is a typical AI generated way to connect ideas. This sentence raises two questions for me. Aspects of what? Solutions for what? As well as being repetitious the sentence is very vague. However, the language used makes it seem that it might make sense if you assume that artificial intelligence used for academic writing is sufficiently intelligent to do the job.

These are three ways to easily identify AI generated writing. These examples show that AI clearly is not capable of generating successful academic writing. If you have found any other examples that seem typical of AI generated academic writing, please email me so we can share these later.