Gains:
- Ability to narrow a broad topic into a concrete researchable question by adding population, context, and focus
- Ability to find the source itself from databases and use artificial intelligence to synthesize only the actual summaries it collects
- Ability to verify every reference suggested by artificial intelligence with DOI and existence check and eliminate the fake source
Every social research begins with a question. A good research question is an answerable, limited and meaningful statement of curiosity: "Why do some young people turn away from politics?" rather than a vague request such as "What are the reasons for university students between the ages of 18-25 not to participate in local elections?" A concrete question like: After the question comes two tasks: literature review — that is, systematically collecting and reading what has been written before on your subject — and establishing a conceptual framework — that is, determining with which theory (systematic thought structure that explains social events) and concepts you will make sense of your research. In this unit, you will learn how to use AI in these three jobs; but you will learn how to stay safe without falling for the fake source.
This is the part of the entire module that requires the most attention. Because literature review is the task where AI produces the most hallucinations. If you tell the AI "name the sources on this topic", it can make up articles, authors, and years that look real but don't exist. So the golden rule: AI doesn't discover literature, it processes literature you find. Finding sources is the job of academic databases (searchable publication collections such as Google Scholar, Web of Science, TR Index, YÖK Thesis, journals' own archives); AI synthesizes the texts you find.
Step by step: from question to frame
1. Narrow down the research question. Start with a broad topic (e.g., “social media and loneliness”) and narrow it down by adding population (who), context (where, when), and focus (what exactly are you curious about). AI is helpful in generating 5-6 alternative concrete questions from a topic; It's up to you to decide which one is worth investigating.
2. Extract key concepts. Define each important concept in your question (e.g. “loneliness,” “social capital,” “digital exclusion”). AI is good at generating alternative names and search terms for a concept; This enriches your database searches.
3. Find YOU the source. Search databases with key terms and download abstracts and full texts of relevant articles. Don't tell AI to "find resources"; Gather resources yourself.
4. Synthesize with AI. Give the actual summaries you have collected to the AI and extract common findings, contradictions and gaps. This speeds up the task of comparing dozens of summaries one by one.
5. Establish the conceptual framework. Based on the findings, decide which theory you will use to read your research. AI can explain a theory in plain language; It's up to you to decide which one fits your question.
Caution: Do not use any reference produced by AI without finding and confirming its DOI (digital object identifier — the permanent ID given to each academic publication) in the database. Just because it "looks believable" doesn't mean it's real; AI produces precisely convincing fakes.
three mini cases
Case 1 — Narrowing the question. One student came up with a huge topic like "immigration and identity." He asked the AI to derive 6 concrete sub-questions; Among them, he chose the question "Native language use and the relationship of belonging among second-generation immigrant youth". The huge topic boiled down to a researchable question for a 4-month thesis.
Case 2 — Fake DOI. A researcher queried 12 references given by YZ collectively by DOI. 3 out of 12 DOIs either did not lead to any publication or led to a completely different article. The AI produced fake DOIs for real journals. Batch checking weeded out these 3 fake sources.
Case 3 — Gap discovery. One team gave 22 real article summaries to the AI and asked "which question has not been researched yet?" YZ pointed out that most studies focus on big cities, while the rural context is almost never addressed. The team made this gap the rationale for their own research question.
Four copyable templates
1) Narrowing the research question:
Location: [broad topic]. Derive me 6 concrete sociologically researchable research questions from this topic. Make sure the population, context and focus are clear in each question. Add a note next to each question saying "Is quantitative, qualitative or mixed appropriate?" Submit questions as suggestions; I will make the choice.
2) Search term generation (I will find the source):
Generate synonyms and related terms that I can use in database searches for the following concepts: [concept 1], [concept 2]. Give Turkish and English equivalents. Suggest a sample search string with boolean operators(AND/OR). Source recommendation; Just give search terms.
3) Literature synthesis (only with the texts I provided):
Below I paste summaries of 10 articles [M1]-[M10]. Based on these ONLY, a table emerges of: (1) agreed findings, (2) conflicting findings, (3) methodological approaches, (4) gaps. In each line, indicate which text it is based on with [M#]. Do not add any external resources.
4) Source verification batch check:
For each record in the reference list below, collect the fields that need validation in a table: author, year, title, journal, DOI. If you are unsure of any area, write "MUST BE VERIFIED". If you don't know a reference at all, don't make it up; Write "unknown".List: [here]
Weak prompt / Strong prompt
Weak prompt:
List the 10 most important articles on digital inequality with bibliography.
This directly invites AI to make up. The model produces 10 sources that do not exist but appear realistic; You cannot confirm any of them.
Powerful prompt:
Your role: synthesis assistant. Below I am pasting the summary of the 9 articles I have collected. Your mission is to map (1) what these 9 studies have in common, (2) where they contradict each other, (3) the methods they used, and (4) the questions that remain unanswered. Use only these texts; Link each claim to the source with [M#]. Summaries: [here]
The difference: The AI no longer "finds" resources, it transforms your actual resources into a readable synthesis map.
Literature review: which job is whose
Stage
AI can
man must do
Narrow the topic
Generating alternative questions
Selection of the question
Search term
Synonyms, translation
Search strategy
sourcing
— (risk of fabrication)
Database scan
reading summary
Quick recap
Quality assessment
synthesis
Common/conflicting findings map
Comment and choice
source verification
Edit a checklist
DOI/entity confirmation
Common mistakes
- Asking for resources from the AI. Finding literature is the database's job; AI makes it up.
- Not confirming the DOI. Even a realistic-looking DOI can be fake; question each one.
- Blind adherence to a single theory. The finding should set the frame, not your bias.
- Jumping the gap. A good research question fits a gap in the literature; Look for this consciously.
- Copying the synthesis as is. Do not present the summary of AI as your own analysis; Go back to the sources and read for yourself.
In summary
The backbone of the research; It is a clear question, an honest literature review, and an appropriate conceptual framework. AI is a powerful aid in narrowing down questions, generating search terms, synthesizing the actual summaries you collect, and building verification lists. But you find the source, you confirm each reference, you choose the frame and you make the interpretation. The biggest trap in the human field is a fake source; The rule is clear: AI does not discover the literature, it processes what you find.
Application task
Choose a research topic. Generate 6 alternative concrete questions with AI and choose one. Search a database yourself and collect at least 8 actual article abstracts. Give these summaries to AI with the "literature synthesis" template and create a map of common findings, contradictions and gaps. Finally, if there are any references that the AI suggests at the beginning of the process, verify them all with DOI and report how many are fake.
checklist
- [ ] I narrowed down the research question by population, context, and focus.
- [ ] I collected the resources myself from the database, I did not ask for it from the AI.
- [ ] I have DOI/existence checked each reference.
- [ ] I had the synthesis done only with actual summaries and with reference to the source.
- [ ] I chose the conceptual framework based on the findings.