In September 2026, I attended the Cheonan engineering library’s workshop, “Starting a Prior-Research Search with AI and Academic Databases”. It followed the path from refining a research question to finding papers and identifying a problem they have not yet answered. As I work towards a paper submission, it prompted me to distinguish collecting search results from understanding the prior research.
The workshop material introduced several academic databases and AI tools. This Note records how I intend to use the search method; it is not an account of testing or evaluating those services myself.
Narrow the question before searching
The search process did not begin by pasting a broad topic into a search box. Separating the research subject, problem, method and evaluation criteria makes it easier to form both a question and useful search terms. The workshop treated AI as a way to examine a question, rather than a means of deciding the answer. Its prompt example specified a role, context, subject, constraints and output format, with the researcher reviewing the questions it produced.
If the initial question is too broad, I need to decide which subject and conditions I can actually study. If it merely repeats an established method, I need to ask what I can compare or improve. A plausible AI-generated sentence is not, by itself, a settled research question.
Design search terms and choose sources
Once the question is clearer, I can expand the terms to include Korean and English expressions, abbreviations and related technologies. The slides showed how to combine terms with AND, OR and NOT, then narrow a broad result set with filters. Because a change in terms can change the result set, I want to record the query and its conditions so that I can revisit the search later.
The source should suit the purpose. RISS, for example, can help with Korean research and theses, while Web of Science and Scopus offer routes into international journals and citation information. Google Scholar and the library’s subscription information can help locate a full text. I will check the current coverage of these services and the university’s subscriptions against the library’s guidance when I search.
When a full text does not appear in the first result, the workshop suggested checking subscriptions and holdings, then interlibrary loan or document delivery. An abstract helps select a candidate, but judging its methods and results requires reading the paper itself.
Expand outwards from one relevant paper
Instead of restarting every search from scratch, I can use a closely related paper as a starting point. Its references lead back to earlier work; papers that cite it lead forwards. Related-paper links may also uncover studies whose terminology differs from my first query.
The workshop introduced Semantic Scholar, Consensus, Elicit, scite and OpenAlex as examples for different stages of exploration. They can help surface related work, compare reported findings or inspect citation contexts. Their summaries and classifications are leads for choosing what to read. I still need to check each claim, method and result in the original paper. Neither citation counts nor an AI summary alone establishes a paper’s reliability.
Judge the research gap through comparison
Later in the workshop, a prior-research matrix put each paper’s purpose, data, method, main findings and limitations side by side. This makes repeated approaches and conflicting results easier to see. It also helps reveal an untested subject or evaluation condition, a remaining methodological limit, or evidence for a further question.
A research gap is not an empty cell that a tool can identify automatically. I need to examine enough relevant work, understand why findings differ, and decide which question my own research can answer. Rather than only saving interesting papers, I intend to record which question each one addressed and what evidence supports its answer.
Where AI helps and where my responsibility remains
The slides used NotebookLM as an example for summarising and comparing uploaded papers and notes, and for outlining a draft. They also separated those tasks from checking original sources, selecting a research question, judging results, citing accurately and taking responsibility for the final text. Even when a tool points to a passage or citation, I need to verify that it matches the source.
The closing checklist covered whether a paper exists, whether a citation matches the original, how the search terms and tools were recorded, data security and copyright, and responsibility for the research. Requirements concerning AI use may differ by institution and publication venue, so I will check the applicable guidance when preparing a submission.
For my own work, I plan to write the question first, keep a record of the search and expand from a few central papers through their citation links. I can then compare those papers on common terms and decide whether the apparent gap is one I can address. Tools may shorten the search, but explaining why that gap matters remains my responsibility.