AI can support scoping review writing without taking over the researcher's thinking, arguments or writing style. A useful approach is to write the research manually, then use AI selectively for factual checking, feedback, identifying weaknesses and improving clarity while retaining control over the final decisions. 

How can you use AI ethically when writing a scoping review?

Writing a scoping review with AI support requires you to decide what AI should and should not do. The writing of the paper can remain manual while AI is used for specific tasks that support the research process.

For this review, I did not use AI much for planning the document because I already knew how I wanted the paper to look. The writing itself was fully manual, but I relied on AI heavily for specific purposes where I needed support.

This was particularly important because it is very easy to get AI to write everything for you, especially when English is not your first language or academic writing is not your strongest area. However, there is value in developing your own arguments and writing voice.

I experienced a sense of accomplishment when I completed the first full draft and read through the arguments I had developed. I do not think AI would have produced the same arguments or made the same critical observations from the data, particularly in the discussion section, because those arguments extended from my own understanding and knowledge of the topic.

How can AI help when writing the introduction?

The introduction was more difficult to write because this was a review and I was not coming into it with extensive knowledge of the topic. I needed to decide how much detail to provide and how technical the writing needed to be.

I knew that I wanted to introduce the importance of AI in research and provide background information about how AI can produce themes. This would give the reader enough context to understand the later discussion of how AI was being used in practice. The background would then lead to the rationale for the review.

Most of the writing involved simply getting the words onto the page. In the first draft, I was not concerned about how the writing presented itself, grammatical mistakes, sentence structures or even citations. I needed to write without constantly editing and agonising over individual sentences.

AI was more useful when I needed to write about how different AI models work and how they can generate themes. There were many terminologies that I wanted to confirm so that I did not misrepresent them or use the wrong terms in the wrong context.

I prompted AI for summaries of areas I was uncertain about and then checked my paragraphs for factual errors. Accuracy was the priority.

How can AI be used for the methods and results sections?

The methods and results sections were approached more systematically. In a scoping review, there is considerable structure around how these sections should look, so much of the process involved filling out the relevant sections.

The main difficulty was locating information across Rayyan, ASReview, Elicit AI, Excel spreadsheets, Zotero and other sources.

The methods section was largely an explanation and justification of each research step. It was one of the easiest sections to write because it was mainly descriptive. The key focus was transparency and ensuring that everything was properly supported, particularly because the review used AI and this remains relatively new in the field.

The results section was also relatively straightforward. In a scoping review, the results involve conveying what was contained in the Excel spreadsheets in an understandable way. The challenge was organising the information.

AI was not particularly helpful here because of the mistakes already identified during data extraction. There was a risk that AI would misunderstand the papers again, meaning that the work involved in tidying and checking the data could be wasted.

Why should you avoid letting AI plan your discussion?

As I wrote the results section, I was already planning my discussion mentally. This is one of the things that happens in the background when you conduct your own research.

Using AI to write the discussion can take that process away from you. It can become a slippery slope where the next section becomes increasingly difficult to write because you have not developed the thinking yourself, making it more tempting to use AI again.

One of the main reasons I did not let AI tell me how to plan my discussion was that the discussion contained the core contribution of the paper. I needed original arguments based on the findings of the review that I could develop and synthesise as the researcher.

My discussion was therefore constructed around three interrelated arguments about the findings. Because this involved generating creative arguments, there were several ways they could go wrong, including fallacies, poor logic and under-substantiated points.

This was where AI became particularly useful. I uploaded my three arguments separately and asked AI to suggest improvements and identify weaknesses. I also asked whether some of my discussion was too critical or analytical for what a scoping review could conclude, and asked for further reading to help me improve.

How can AI help you edit an academic paper?

AI was used most heavily during the editing stage. My problem as a writer is that after looking at my writing for a long time, I can stop seeing the problems with it. At this point, another academic would normally provide a useful perspective, but I was not ready for the paper to be reviewed by someone else.

I first uploaded the whole draft to ChatGPT and asked for general feedback. It provided positive feedback as well as areas for improvement.

I then worked through the paper section by section, starting with the abstract. I asked how well it reflected the paper and where the writing could be clearer. Some observations were useful, although some were inappropriate. For example, AI suggested pushing the critique further in the abstract, which went beyond what a scoping review should do.

I also asked AI to identify verbosity and then prompted it for more specific areas to address. After redrafting, I asked for further feedback on the updated document.

I repeated this process for each section. Rather than automatically accepting the feedback, I questioned its judgement and only adopted suggestions when I was convinced they were appropriate.

How should you challenge AI feedback on academic writing?

One useful difference between AI feedback and feedback from another academic is that I felt more comfortable rejecting and challenging AI's suggestions.

When another person reviews your work, you may be reluctant to challenge their feedback because you also want to demonstrate that you value their input. With AI, I could question its judgement directly.

Some of my prompts also explained how I wanted AI to read the writing. I briefed it on my intentions for the section and what I wanted the section to convey, then asked it to assess how well I had achieved that.

I approached the editing process one section at a time. This allowed each part of the paper to receive attention without asking AI to jump around the document and potentially confuse both itself and me.

The purpose was therefore not to ask AI to rewrite the paper. It was to use AI as a critical reader whose feedback I could assess, challenge and selectively accept.

What can you learn from using AI to write a scoping review?

Writing is enjoyable, and AI can easily take that away from you. It can be tempting to let AI take over when you are staring at a blank page and struggling to start.

There is a genuine achievement in developing an original argument based on your own findings. The process of writing allows you to develop your thinking, make observations and produce something that reflects your own understanding.

The way I incorporated AI into this writing process allowed me to harness its capabilities while retaining my writing style and the intrinsic rewards of writing.

AI was useful for checking factual accuracy, identifying weaknesses in arguments, providing feedback and suggesting areas for further reading. I kept the actual writing and core arguments under my control.

This approach allowed me to use AI as part of the research process without outsourcing the thinking that made the research my own.