AI can simplify and speed up several stages of a literature review without replacing the researcher’s own analysis. Used carefully, it can support topic exploration, search terms, database selection, paper screening and data extraction while keeping the underlying literature review process intact.

I break this down with UK marker examples in the video.

How Can AI Help You Start a Literature Review?

When you are new to a research area, it can be difficult to know where to begin reading or where the gaps in the literature might be. If your preliminary research is not sufficiently developed, you may find yourself having to reposition the whole literature review, which can be a significant drain on the limited time you have, especially as a Masters student.

Generative AI can give you that initial boost. You can use AI to help identify areas of interest and areas where there is a lack of research.

This only helps identify literature gaps that have been stated in the articles and requires more in-depth analysis. When conducting a literature review, you are looking for what is known in the research. Information about what is not yet known can help identify what your empirical research could be about or what you propose as the research gap.

Once you have identified what is not yet known, you have identified the boundaries of your literature review. You can then look at everything adjacent to this boundary within your review. At Masters and PhD level, you are expected to understand the research frontier, so your literature review should focus on this narrow section.

You can also use generative AI to probe further into this area. Scite Assistant can give you some papers to start with and help kick-start your preliminary research.

Your next step is to identify your review question using SPIDER, PICO, PCC and similar frameworks, and identify your review objectives. This step is unique to your analysis of the preliminary literature searched. If you rely solely on AI for this step, you can get very typical and repetitive review questions. Combining AI with your own analysis can produce more refreshing research questions and areas of focus.

How Can AI Help With Literature Search Terms and Databases?

Once you have your review question and objectives, you are ready to start your literature review search.

Two problems commonly arise. You need to identify your keywords and make sure you have covered them all. You also need to identify appropriate databases.

For key terms, you can put your research question into an AI tool and ask it to suggest keywords. If you are a health researcher using PubMed or MEDLINE, you can also ask generative AI to suggest MeSH terms. This can save time tracing the MeSH terms you need.

However, make sure you check the MeSH terms.

Databases can also be identified using generative AI. Context is very important here because AI may not grasp the context of your search well and may suggest the wrong databases. Provide more specificity and double-check the relevance of the databases suggested.

Once you have your key terms, you need to combine them using Boolean operators. ChatGPT can do this quite well, to a certain degree, but you must check everything.

For example, an AI-generated search string may contain terms that are not actually MeSH terms. Some wildcards may also be placed in the wrong location or poorly optimised.

To refine an AI-generated search string:

1. Break it up into the respective key terms so that it is easier to read.

2. Remove invalid MeSH terms.

3. Fix the terms one by one.

For students new to conducting literature reviews, AI can be a useful way to get started with putting your search string and databases together. However, you need to check it properly. As a marker, I sometimes copy and paste search terms into databases to check a student’s work when marking dissertations. Errors can become obvious when you run the searches because the database itself can tell you if the terms are invalid.

If you are an experienced researcher, it may be faster to do your search string manually.

How Can AI Help With Selecting Papers?

Once you have completed your search, you need to start filtering papers.

The issues you will face here are mostly about organisation and priority. If you only have to filter 30 papers, this might not be a problem. However, it is quite likely that you will have to filter papers in the hundreds.

As you review papers, you need to keep a record of relevant papers and perhaps even irrelevant ones, especially when they may become relevant when you write your background or discussion sections.

Priority is another issue. If you review papers in the order they appear in the database search, you might only start finding relevant papers after rejecting 100 papers because of their titles.

If a search produces 6,000 papers and only 5% are relevant, it may not be feasible to filter more than 5,900 papers within the time available for your dissertation or PhD. This is time consuming, so you want to get to the relevant papers as quickly as possible.

This is where ASReview comes in. ASReview is an open-source AI programme that helps researchers sort relevant and irrelevant papers using machine learning while still giving you full control over the decision-making.

ASReview presents the papers you have selected one by one. You click relevant or not relevant. As you do this, the programme starts to learn which papers are more likely to be relevant and prioritises them in the order in which they are presented. Irrelevant papers are pushed down the order, which also means you are less likely to miss relevant papers within the stack.

To use ASReview:

1. Download the CSV or RIS files of your search results.

2. Import this file into ASReview.

3. Warm up the programme by filtering at least one relevant and one irrelevant paper.

4. Start reviewing the whole stack of papers with the AI assistant.

5. Return to the process at a later date if needed.

6. Go into the history to see which papers were deemed relevant and irrelevant.

7. Export the files and download the required documents separately.

ASReview has existed since 2021, and published journal articles show how other systematic reviewers have utilised this method. This can help support your justification for using AI programmes in a systematic review.

How Can AI Help With Reading and Data Extraction?

Prior to this step, you would already have read your selected papers.

Data extraction and critical appraisal can be time consuming because you typically have to return to the papers multiple times to find the information you need. There is also a degree of potential human error, where you can make mistakes or miss details.

You want a way to get this information appropriately presented in some sort of table so that you can compare your studies.

You also need an AI system that enables you to pull out customised information because there will be specific elements of data that you want, such as barriers and facilitators, factors or outcome measures used.

Elicit AI can support this process of data extraction.

There are two ways to approach this. You can read the papers first, extract the data manually and then put the papers through Elicit AI to pull out the data you need so you can double-check your data extraction.

Alternatively, you can let the AI do its job first, pull the data you need and then go back to read and confirm the information that has been extracted.

Either way, you need to double-check the work of the AI and make sure that you are familiar with the papers yourself. You need to conduct your own analysis later, and you cannot analyse the papers if you do not know enough about them.

To use Elicit AI:

1. Boot up Elicit AI.

2. Upload your papers.

3. Select the papers you want to extract the data from.

4. Add the type of data you want to extract.

5. Add custom columns by telling the AI what you actually want from the paper.

Sometimes, when information is presented in a diagram or table, the AI is not able to read the data properly. The AI can also misread information, so its output needs to be checked.

Where Should You Be Cautious About Using AI in a Literature Review?

The use of AI in literature reviews tends to be very focused on the search, selection and data extraction stages. There is less evidence of it being used in the analysis of reviewed papers.

There are significant academic integrity concerns around using AI in analysis and writing. As a marker of a dissertation, it can be difficult to award marks for critical analysis if the use of AI is declared in the writing or analysis.

For this reason, AI can be used to simplify and speed up particular steps of the literature review while keeping the research process itself intact. It should support the researcher rather than replace the researcher’s analysis.

If you are applying this approach to your own literature review and want a second pair of eyes, you can get support with your work here: https://www.drkenyanwong.com/start-here