How do I know when to stop searching for papers for a literature review?

The search never ends on its own. Every reference list contains three more papers, every one of those has its own, and the supply is effectively infinite. If your stopping condition is "when I have found everything", you have chosen a condition that cannot be met, and the anxiety that follows is a design flaw rather than a character flaw.

The workable condition is saturation, borrowed from qualitative research and applied to reading. You are saturated when additional sources keep confirming what you already have and stop adding new claims, new counterexamples, or new methods. It is visible in your notes: the last stretch of screening produced nothing you had to write down.

Bramer et al. describe a test that turns that feeling into something you can run. You never know the total number of relevant articles or which ones they are, so instead of counting, add synonyms to your existing strategy and then run a "new strategy NOT old strategy" search. Read only what the expanded version returns that the original did not. If those records are not relevant, the added terms did not improve coverage, and the original search was already reaching the topic (Bramer et al., 2018, p. 536).

Two checks make it more than a feeling. First, run a backwards and forwards citation sweep on the five or six works your argument leans on hardest. If the papers citing them and cited by them are already in your set, your coverage of that conversation is good. Second, ask a supervisor or a colleague in the field to name three papers you should have. If you already have them, that is meaningful evidence.

Bramer et al. turn the same sweep into a recall check you run once your included set is settled. Take the studies you reached through reference lists or a colleague rather than through your database strategy, and check whether the database held them all along. Any study the database contained but your search missed shows you which terms are missing, so add them and rerun the search (Bramer et al., 2018, p. 538). Pautasso gives the lighter version of the same instruction, use different keywords across different databases and follow the citations out of the papers and book chapters your argument depends on (Pautasso, 2013, p. 2).

Saturation is also local rather than global. You can be saturated on your central claim and thin on a secondary one. Check it per section rather than for the chapter as a whole, because the thin section is where an examiner will land.

One thing saturation is not is a screening cutoff. Haddaway et al. looked at seven environmental-management systematic reviews and found several had screened only the first 50 Google Scholar records and one the first 260, a habit they report as common but not evidence-based (Haddaway et al., 2015, p. 4). The counts show why a fixed number is arbitrary. For three of those topics, Web of Science returned 4,151, 47,932, and 2,550 records, while the matching Google Scholar full-text searches returned 318,000, 554,000, and 216,000 (Haddaway et al., 2015, p. 5). They also found the densest grey literature in their title searches around page 35, and recommend about the first 300 results for title searches aimed at academic literature and more than that when grey literature matters (Haddaway et al., 2015, p. 12). A cutoff is defensible only when you can say why the relevant material would appear within it.

How many papers should a literature review include?

There is no number, and any figure you are given is a discipline norm rather than a rule. A thesis chapter typically cites somewhere between sixty and two hundred works. What matters is coverage of the claims your argument depends on, not the size of the reference list.

The question behind the question is usually whether the chapter looks credible. It does when every substantive claim carries support and the important debates are represented by their actual participants. It does not when the reference list is padded with works cited once, in passing, to raise the count.

If you need a target, count claims rather than papers. List the assertions your argument requires and check that each has at least one solid source, and that any contested one has both sides represented.

What stopping rule can I defend in my methods section?

Name the databases searched, the search strings, the date the searches were run, the inclusion criteria, and the point at which new results stopped producing new claims. That is a reproducible account. "I searched until it felt complete" is not, and an examiner will notice.

Write this down while you are searching rather than reconstructing it afterwards. A short log with the date, the database, the string, and the number of hits takes seconds per entry and turns into a methods paragraph you did not have to remember.

Recording the date matters more than people expect, because it is what allows you to say honestly that the review is current as of a stated point rather than current in general. Nobody expects a thesis to include something published last week. They expect you to say where the line is.

Pautasso adds a reason to keep that date live. Reviews go out of date before they are published, because papers stay in press for months before they reach the databases, so monitor papers in press and run the search again during revision (Pautasso, 2013, pp. 3-4). The same check works in the other direction, look for older overlooked studies, the sleeping beauties, before you claim a gap that has already been filled. A record that states the last search date, the databases and platforms, the exact strings, the citation chaining you performed, the inclusion and exclusion criteria, and your reason for thinking further searching would not change the central claims is a record someone else can verify (Pautasso, 2013, p. 1).