Where the literature lives, and how to search it
In this lesson
By the end, you’ll be able to
- Choose a database that fits the question, knowing what each one covers and misses
- Turn a question into a Boolean search with synonyms, phrases, and controlled vocabulary
- Snowball a topic backward and forward through citations to find what keywords miss
Sign in to save your progress
You can keep reading without an account, but completed lessons won’t be saved.
There is no single place where "the research" lives, and the place you search decides what you will find. Google Scholar casts the widest net: it indexes almost everything, including preprints, theses, and books, which makes it great for a first look and for finding who cited a paper, but its coverage is opaque, it cannot be searched reproducibly, and it ranks by a mix of relevance and citation count that buries newer work. PubMed does the opposite: it covers the biomedical and life-science literature only, but every record is curated, tagged with controlled vocabulary, and searchable in a way you can write down and re-run. Preprint servers (arXiv for physics and computing, bioRxiv and medRxiv for the life sciences) hold work that has not yet been peer reviewed, which is why they are fast and why you must read them with extra care. The rule of thumb: use a curated, field-specific index when you need completeness you can defend, and use Scholar to catch what it missed.
A good search is built, not typed. Start by breaking the question into its concepts, then handle each concept in two moves.
First, gather synonyms and combine them with OR so you catch every way people phrase the idea: ("teenager" OR "adolescent" OR "youth"). Second, join the different concepts with AND so every result mentions all of them: (teenager OR adolescent) AND ("screen time" OR "smartphone use") AND sleep. Quote exact phrases, and use field limits (title, abstract) and date ranges to cut noise.
The most powerful move is controlled vocabulary. PubMed tags every article with MeSH terms, a standardized subject vocabulary, so a single MeSH term finds papers no matter which synonym their authors happened to use. A search you write this way is reproducible: you can paste it into a methods section and someone else can re-run it exactly.
Exercise
You want papers on how smartphone use affects teenage sleep. Which search is most likely to return relevant results without drowning you in noise?
Three concepts: the device, the age group, and the outcome. Each concept has synonyms (OR); the concepts must all appear (AND).
Categorize
Every search move either widens the net (more results, higher recall) or narrows it (fewer, more precise results). Sort each move into the right bin.
OR means 'either term counts', so it adds results. AND, quotes, field limits, and date limits all require more, so they cut results.
No keyword search is complete, because authors do not all use your words. The fix is to follow the citation graph. Pick one strong, on-target paper and chase it in two directions. Backward: read its reference list to find the earlier work it was built on. Forward: use "cited by" (Google Scholar and most databases show it) to find newer work that built on it. This is called snowballing, and in careful comparisons it surfaces relevant studies that keyword searches miss entirely, especially in fields with messy or inconsistent terminology. A review that relies on keywords alone has a blind spot; snowballing is how you close it.
Signature interactive
Citation Trace
Start from the recent paper and follow its references back to the foundational finding (marked ). Each paper you open costs one of your budget. Reading is a resource-allocation problem, spend opens where the trail is likely to lead.
Scaling laws for efficient model training (2021)
A constructed citation graph, for practice.
Written response
Pick any question you are curious about. Write a search strategy for it: name one database you would use and why, break the question into its core concepts, and write out a Boolean search that groups synonyms with OR and joins concepts with AND. Three to five lines is plenty.
Checkpoint · item 1 of 4
You need a search you can write down and have a classmate re-run exactly, on a health topic. Which database fits best?
Which option is curated and lets you paste the exact search into a methods section?
Reflection