Gains:
- Being able to distinguish where artificial intelligence saves time in social research (summary, transcript, coding, draft) and where research questions, interpretations and ethical decisions are left to humans, depending on the level of risk.
- Ability to understand the risk of hallucinations and fabricated sources in the human field and apply a discipline that independently verifies each fact, number and reference.
- Understand why it is necessary to transparently state the contribution of artificial intelligence, preserve originality and leave no trace of prompts from the very beginning
Sociology studies how people live together; It is the branch of science that systematically studies groups, institutions, inequalities, culture and social change. A social researcher's day; It consists of collecting surveys, conducting interviews, scanning the literature (all previously written scientific publications on a subject), analyzing data and reporting the findings. All of these tasks take time, and in all of them, artificial intelligence (AI)—here, large language models that understand text, summarize it, encode it, and generate drafts—can be a serious accelerator. This module will teach you how to use AI at every stage of social research; but it teaches you why you should keep the interpretation, the theoretical (theory-based) framework, ethical responsibility, and the final finding in your own hands.
Humanities and social sciences have a feature: here the "correct answer" is often not a single and definitive answer; interpretation, context and chain of evidence are decisive. So this is where AI is most dangerous. AI can hallucinate — that is, it can fabricate a non-existent source, author, date, or finding as fact, and in extremely convincing language. In social research, this means a fake reference slips into an article, a fabricated statistic slips into a report, or something is attributed to a participant that they did not say. This is the main theme of this module: AI is an assistant, you are the author and researcher.
Step by step: Embedding AI in the research process
1. Separate the task by risk level. Not every task carries the same risk. Having the AI summarize an interview transcript is low risk (since you can read and verify the transcript). Asking AI "what does the literature say about this" is high risk (as it may make up sources).
2. Give the AI the jobs it is strong at. Summarizing long texts, editing transcripts, generating code suggestions, writing draft paragraphs, formatting tables—these are the tasks that AI is fast and good at.
3. Protect human work. Establishing the research question, choosing the theoretical framework, assessing the representativeness of the sample (the group of individuals or units included in the research), interpreting the finding, and making the ethical decision—this is yours.
4. Verify each output. No fact, number or quote produced by the AI enters the report without you verifying its source.
5. Keep track. Make a note of which prompt (the instruction you gave to the AI) you used, which model you ran, and on what date. This is necessary for future reproducibility.
Tip: Think of the AI like a “very fast but sometimes unreliable intern.” You always read the intern's draft, ask for the source, and sign it. Apply the same discipline to AI.
three mini cases
Case 1 — Fabricated reference caught. A graduate student asked AI to “suggest 5 classic papers on urban loneliness.” The AI produced 5 articles with real author names that did not exist — the titles were believable. The student searched for them all in the academic database; 4 out of 5 were not there at all. Without verification, fake references would enter the thesis.
Case 2 — Coding 3x faster. One researcher had 40 interview transcripts. He gave the first 5 transcripts to YZ and asked for code suggestions; reworked the resulting code list according to his theoretical framework and processed the remaining 35 transcripts with this consistent set of codes. 3 weeks of work reduced to 1 week; decisions were always human's.
Case 3 — Sampling warning. One team had AI summarize 900 responses to an online survey and concluded that “young people trust social media.” A consultant asked: Who is the sample? Most of the responses were from students of a single university. The AI summary was technically correct, but the sample was not representative of all youth; the result was rewritten with this limit.
Four copyable templates
1) Role and boundary setting (at the beginning of each session):
Your role: social research assistant. Your job is to summarize, edit, and propose code and outlines. Source fabrication; Write "must be verified" wherever you are unsure. Just rely on the texts I gave you; do not add statistics or references from your own memory. I will make the interpretation and final decision.
2) Text summarization (low risk, safe):
Summarize the interview transcript below. Keep the participant's own words, do not add new information. Output: (1) 5 main themes, (2) 3 notable direct quotes (with line number), (3) unclear points. Text: [here]
3) Source verification reflex:
For each of the references you just suggested: give author, year, journal/publication, and DOI (digital object identifier). If you are unsure of the authenticity of a reference, write "NOT VERIFIED". I would rather you say "I don't know" than make it up.
4) Bias review:
Examine this draft: [text]. Mark each generalization in it and ask: "What sample is this claim based on? Is there any counter-evidence? Does it describe a social group as something it is not about?" Come up with a list of biases and overgeneralizations.
Weak prompt / Strong prompt
Weak prompt:
Summarize and reference literature on gender and business life.
From its old memory, the AI produces an unverifiable text filled with sources it has never read or that do not exist. This is where the risk of fabricated references is highest.
Powerful prompt:
Your role: research assistant. BASED ON the abstracts of the 6 articles I will paste below (all downloaded from the database), synthesize common findings, conflicting findings and gaps on "gender and work life". Just use these 6 texts; adding external resources. Show the text on which each claim is based with the [M1]-[M6] tag. Summaries: [here]
The difference is clear: AI is no longer making it up, it is synthesizing the actual texts you give it and attributing each claim to the source.
The role of AI: where powerful, where humans are needed
Quest
Contribution of AI
human responsibility
Risk level
Long text summary
quick draft
accuracy check
low
Transcript editing
Cleaning, formatting
preservation of meaning
low
Qualitative coding
Code suggestion
Theoretical decision, consistency
medium
Literature synthesis
Merge given texts
source verification
high
Statistical comment
account draft
Judgment of method and meaning
high
Ethical decision
—
only human
critical
Common mistakes
- Using the resource provided by AI without verifying it. The most common and most serious mistake; Fake reference destroys your academic reputation.
- Leaving the interpretation to the AI. The question "What does this mean?" is the job of the researcher, not the model.
- Making generalizations without questioning the sample. Even if the AI summary is accurate, the data may not be representative of the entire population.
- Neglecting originality. Presenting the AI-generated text as your own sentence violates academic integrity; State your contribution transparently.
- Leave no trace. If you don't note which prompt you used and when, no one will be able to reproduce your work.
In summary
AI in social research; is a powerful assistant that summarizes texts, edits transcripts, suggests code and writes drafts. But the research question, theoretical framework, sample judgment, interpretation, ethical decision and final finding belong to humans. The greatest danger to human domains is the source of hallucination and fabrication; That's why every fact, number and reference does not enter anywhere without being verified. Position the AI according to the level of risk, feed it with powerful resources, maintain authenticity and transparency, and sign off each deliverable with your own judgment.
Application task
Choose a topic from your own research field. First tell the AI “suggest 5 articles on this topic” (the weak way) and search each suggested source in the academic database; Count how many are real. Then, give the 3 real summaries you downloaded to the AI and ask for synthesis via the "strong prompt". Write a half-page note comparing the reliability of the two approaches.
checklist
- [ ] I separated tasks by risk level (low/medium/high/critical).
- [ ] I gave the AI a starting prompt that set the role and boundaries.
- [ ] I have verified each recommended source against an independent database.
- [ ] I kept the interpretation, the framework, and the ethical decision in my own hands.
- [ ] I have transparently noted the AI's contribution and left a trail of my prompts.