Writing a Literature Review Backwards: Claim First, Sources Second
Most advice says collect broadly, then look for a pattern in what you found. Reverse it: state the claim first, then search only for what would confirm or break it. This post covers citation chaining in both directions, building a search string from the claim itself, and knowing when to stop.
Sixty saved PDFs and no idea what they add up to is the most common failure mode in a literature review, and it comes from doing the steps in the wrong order. The usual advice, search broadly, save everything relevant, then find the throughline, treats source-finding as data collection first and argument-building second. Reverse the order and the search gets sharper: state the claim first, then search only for what would confirm or break it.
How to find sources for a literature review is usually taught as a browsing problem: pick good databases, pick good keywords, save what looks relevant. It works better as a targeting problem. Once a claim exists, a source only earns a place in the review by doing one of two jobs: supporting that claim or complicating it. Anything that does neither is not evidence for the argument. It is a paper that happens to share a topic.
This assumes the claim already exists in rough form, a working thesis or the alternative hypothesis half of a hypothesis pair, from earlier in the process. If that decision is still open, back up to the fuller guide to how to plan a research paper before running any of the searches below.
How to Find Sources for a Literature Review, Backwards
Start with the claim you actually need to support or complicate (the one you got from your hypothesis or thesis). 'Heavier daily social media use is associated with shorter attention spans in university students' is searchable. 'Social media and attention' is not, because it returns everything in the field rather than the handful of sources that bear directly on that specific claim.
For that claim, three kinds of source do real work, and a claim-first search targets exactly these instead of returning everything published on the general topic.
- Something that found the same relationship, which supports the claim directly
- Something that found no relationship, or the opposite one, which the claim has to address rather than ignore
- Something that explains a plausible mechanism connecting the two variables, which turns a correlation into a candidate explanation
Chaining Backward Through a Source's Reference List
Find one strong source that already bears on the claim, ideally a recent paper or a review article, and open its reference list. Every source it cites that also touches the same relationship, social media use and attention, is a candidate. This works because a well-cited recent paper has already done a version of the search, and its reference list is a filtered set rather than a raw database dump.
Repeat this once or twice more from the strongest new sources that turn up, and the same handful of foundational papers usually starts appearing across multiple reference lists. That repetition is a real signal: a source cited by everyone already working on this claim is one your own reviewer will expect to see. The limitation is age. A reference list only points backward in time, so chaining this way alone will miss anything published after your starting paper.
Humanize your own paper
Transform your AI-assisted text and make it sound human, without touching important words or citations.
Chaining Forward Through Citing Articles
Forward chaining fills in exactly what backward chaining misses. Consider that foundational paper again, but this time look at everything that cites it using a database's cited-by function, Google Scholar, Scopus, or Web of Science all offer one. You'll find some papers that retested the same relationship with a different population or method, and that's exactly the sort of thing a claim-first search is designed to dig up.
Run it on the social media and attention claim and forward chaining is often where the complicating evidence actually turns up. Imagine a 2024 replication that found no significant effect once screen time was measured objectively rather than self-reported: backward chaining from a five-year-old foundational paper would never surface it, since the newer paper did not exist yet when the older paper's own reference list was written.
Building a Search String That Targets the Claim
A claim-first search string comes directly from the claim's own parts: the variables, and the words researchers in that field actually use for them. For the social media and attention claim, that means pulling out 'social media use,' 'attention,' and 'university students' or 'undergraduates,' then combining them with AND, and expanding each with OR to catch how different authors phrase the same variable. Check the keywords list of two or three sources you already trust for this claim; researchers in the field will have already settled on the terms a database indexes, which is more reliable than guessing synonyms yourself.
| Variable | Search terms (OR) |
|---|---|
| Exposure | "social media" OR "social networking site*" |
| Outcome | attention OR "attention span" OR "sustained attention" |
| Population | "university student*" OR undergraduate* |
The next step is to combine these three rows using the logical operator AND, and we get a complete sentence that looks like this: ("social media" OR "social networking site*") AND (attention OR "attention span" OR "sustained attention") AND ("university student*" OR undergraduate*). The asterisk at the end of a word truncates it, meaning that our search will catch the different endings of that word (student and students and studentship all at once). In most academic databases, this Boolean search syntax is supported directly by the database itself. But a general search engine mostly doesn't. This is one more reason why a dedicated database search beats broad web searching for this specific task.
A free citation finder does this same targeted search automatically, taking the claim itself rather than requiring a hand-built Boolean string, and returns real, checkable sources rather than a generated reading list.
When to Stop Searching
Stop when new searches mostly return sources already found. This is not a productivity trick. Once backward chaining, forward chaining, and a properly built search string all keep surfacing the same core set of ten to twenty sources, you have mapped out the claim's evidence base. Any further searching mostly adds noise rather than new signal.
A second signal matters more than volume: every major position on the claim, support, no effect, and a plausible alternative explanation, has at least one source behind it. A review that cites twelve sources but covers all three positions is stronger than one that cites forty and covers only the position the writer already believed. That is worth remembering when a reviewer asks why the reference list looks shorter than a classmate's.
Once a source clears that bar, format it immediately rather than circling back to a stack of PDFs later. A citation generator turns the record into a properly formatted reference in whichever style the paper requires, while the source and the claim it supports are still in front of you.
What you are searching for is set by the question itself, and writing a research question is covered separately, along with the types of research questions worth choosing between.
A literature review built this way argues something instead of summarizing everything it could find on the topic, which is the actual difference between a source list and a review. Once the sources are gathered and the review is drafted, an AI humanizer built for academic writing is worth knowing about for the sentences that need a second pass.
Frequently Asked Questions
Start from the specific claim you need to support or complicate, not the general topic, then search for sources that confirm it, contradict it, or explain a mechanism behind it. Chain backward through a strong source's reference list and forward through what has cited it since. A search built around a topic returns everything published in the field; a search built around a claim returns what actually bears on your argument.
Content planner and copywriter at TextPulse. Sara runs the blog day to day, from planning and drafting through to publishing. She writes the practical guides: clear explanations of academic writing problems, aimed at the person who actually has to hand something in.
