METHODS
The research pathway
Seven stages, with human decisions at the centre. Choose a starting point for your project.
Map the literature
Start from two or three seed papers you trust. Use an open index for broad search and a citation-graph tool to find what the seeds cite and what cites them. Keep a log of every query.
AI search is not reproducible in the way a database query is. If the review must be reportable, keep a conventional search strategy as the record and treat AI tools as supplements.
Screen systematically
Write the protocol and inclusion criteria first. Then let an active-learning tool order the records while you make every include/exclude decision yourself.
Decide a stopping rule before you start and report it. Do not let a language model make final inclusion decisions.
Work with qualitative material
Transcribe locally, then use AI to navigate and retrieve passages from documents you are allowed to share, then do the interpretive coding yourself and keep a memo trail.
Interview and fieldwork data usually fall under IRB and consent terms. Check them before uploading anything to a cloud model.
Build data and run analysis
Learn the method before you automate it. Use open APIs (e.g. OpenAlex) for data you can document, and have an agent draft code you then read line by line.
Never report a number you have not reproduced yourself. Pin package versions and keep the prompts used to generate code.
Pilot surveys and experiments
Use simulated respondents to explore wording and survey logic, then pilot with people. A simulation does not establish validity or the distribution of human responses.
Report simulated outputs as simulation results, never as observations of people. Plan and document any comparison with human data.
Write, check and get feedback
Draft in your own words. Use AI for language polish, reference validation and a simulated referee pass before you send the paper to real readers.
Check your target journal's AI-use disclosure policy. Language tools can change claims; reread every accepted edit.
Stay current without drowning
Pick one recommender you train on your interests and one or two human-curated newsletters. Review them on a fixed weekly slot.
Most AI feeds are arXiv-centric. Add journal table-of-contents alerts for your social-science field.