Blog

· By Thesisroom

How to Do a Literature Search for a Thesis: Steps and a Worked Example

A step-by-step literature search for a thesis, run on PubMed with the result count after every change, plus a known-paper test, a stop rule and a printable search log.

This guide has a printable version: open the printable search log

An open illustrated book, a blank notepad with a pencil, a laptop and a white mug on a wooden table.
Photo by Wouter on Unsplash

How to do a literature search: break your question into concepts, then list synonyms and subject headings for each one. Join synonyms with OR and concepts with AND, as UCL teaches, run the string in your field’s databases, test it against papers you already know, and write down every run.

This guide runs the method on PubMed on 10 November 2026 and shows what each change does to the number of results. It also covers testing a search, deciding when to stop, and keeping a log for your methods chapter. For the whole review, see how to write a literature review for a thesis, which this guide expands at the search step. Rules and recommended databases differ by university and field, so check your own library’s guide as you go.

How to do a literature search for a thesis in seven steps

Work through the seven steps in order, and keep notes as you go. The sections after the list give the detail.

  1. Break your question into concepts. UCL’s guide to defining search terms says to pick out the main keywords and concepts. “Does sleep affect grades in university students?” has three: sleep, academic performance and university students.
  2. Run a quick scoping search. UCL describes this as a brief search to see how much research exists, and how to choose a thesis topic explains what it tells you. Research Compass takes a typed research interest and shows papers sorted into groups, key authors, the most-cited papers and the newest papers, and seeing the results needs a free account.
  3. List synonyms, spelling variants and subject headings for each concept.
  4. Combine the terms: OR inside each concept, AND between concepts, and brackets around each concept.
  5. Choose databases for your field and adapt the string to each one.
  6. Test the string against papers you already know.
  7. Record every run and decide when to stop.

How do you build the search string?

Build it one concept at a time: a group of synonyms joined by OR for each concept, then the groups joined by AND.

Synonyms, spelling variants and abbreviations

List every other way a paper might name the concept. UCL’s guide to defining search terms prompts you to check synonyms, acronyms, plurals, wider and narrower terms, and words spelled differently in UK and US English, such as centre and center. In the worked example below, none of the five known papers has “university students” anywhere in its PubMed record, and four say “college students”, so that group needs both phrases.

AND, OR and NOT

OR widens a search, AND narrows it, and NOT removes records, so UCL says to use NOT with caution. On UCL’s search techniques page, OR joins synonyms of one concept and AND joins different concepts, and a search for teenager NOT adult also drops papers that discuss both. PubMed needs operators in uppercase and reads left to right, so brackets decide the order (PubMed User Guide).

Subject headings (controlled vocabulary)

A subject heading is a standard label that a database gives each record, whatever words the authors used (UCL’s page on subject headings). PubMed’s version is MeSH, and a MeSH search also covers the narrower headings beneath the one you pick (PubMed User Guide).

Not every database has subject headings, and each thesaurus differs: Medline uses Stereotyping where PsycINFO uses Stereotyped attitudes. UCL says to combine headings with a free-text word search rather than rely on them alone.

What about truncation, wildcards and phrase marks?

Truncation and wildcards stand in for letters, so one term can catch several endings or spellings. UCL’s search techniques page says most databases use the asterisk: psycholog* finds psychology, psychologist and psychological, but psych* also finds psychiatry and psyche.

UCL’s page on wildcards says the symbols vary between databases, with the question mark, hash and dollar sign as other examples. It also says not every database supports them, and they can switch off automatic term mapping in PubMed.

Phrase marks differ too. Without quotation marks, UCL says, databases often search your words as if joined by AND, so they may not sit side by side. Some, such as the Ovid platform, treat adjacent words as a phrase by default. Read the database’s own help page before you rely on any symbol.

What does a worked example look like on PubMed?

Each change to a search string moves the count in a predictable direction. This example uses a neutral topic, sleep and academic performance in university students, run on PubMed on 10 November 2026.

We typed each string into PubMed and read the number of results through NCBI’s public search service. PubMed adds records every day, so your counts will differ, and PubMed rewrites some searches behind the scenes. Its Search Details show the translated query (PubMed User Guide).

StepWhat changedSearch typedResults on 10 November 2026
1One concept alonesleep373,082
2A second concept aloneacademic performance49,227
3Two concepts with ANDsleep AND academic performance2,097
4Three concepts with ANDsleep AND academic performance AND university students924
5Phrases in quotation markssleep AND "academic performance" AND "university students"215
6Synonyms joined by OR inside each conceptThe first string below514
7Every word tagged to title and abstractThe same words, each with [tiab]422
8Subject headings only"Sleep"[MeSH Terms] AND "Academic Performance"[MeSH Terms] AND "Students"[MeSH Terms]121
9Subject headings or words, for each conceptThe second string below759
10Step 9 plus a date limitStep 9 AND 2021:2026[dp]479
11Step 9 plus NOTStep 9 NOT (school[tiab] OR schools[tiab])528

The string for step 6:

(sleep OR "sleep quality" OR "sleep duration" OR insomnia OR "sleep deprivation" OR sleepiness) AND ("academic performance" OR "academic achievement" OR "grade point average" OR GPA OR "academic success") AND ("university students" OR "college students" OR undergraduates OR "undergraduate students")

The string for step 9:

("Sleep"[MeSH Terms] OR sleep[tiab] OR "sleep quality"[tiab] OR "sleep duration"[tiab] OR insomnia[tiab] OR "sleep deprivation"[tiab] OR sleepiness[tiab]) AND ("Academic Performance"[MeSH Terms] OR "academic performance"[tiab] OR "academic achievement"[tiab] OR "grade point average"[tiab] OR GPA[tiab] OR "academic success"[tiab]) AND ("Students"[MeSH Terms] OR "university students"[tiab] OR "college students"[tiab] OR undergraduates[tiab] OR "undergraduate students"[tiab])

Paste either string into PubMed to run it yourself.

What the table shows:

  • AND narrows. Steps 1 to 4 fall from 373,082 to 924 as each concept is added.
  • A bare phrase is not always a phrase. Step 2 returned 49,227 because PubMed also accepted records that contain “academic” and “performance” anywhere, not side by side. Quotation marks make PubMed look for the exact phrase and stop the MeSH mapping, which is why step 5 cuts 924 to 215 (PubMed User Guide).
  • OR widens inside a concept. Adding synonyms lifts 215 to 514 in step 6.
  • Field tags switch off PubMed’s helpers. A bare sleep was translated into the MeSH heading Sleep plus the words sleep, sleeping and sleeps. A tag such as [tiab] stops that translation. Step 7 also searches only titles and abstracts, so it drops to 422, and you must type plurals and spellings yourself.
  • MeSH alone is strict. Step 8 gives 121, because only records indexed with all three headings match.
  • Headings or words casts the widest net. Step 9 reaches 759 by letting either route match, more than steps 6, 7 or 8.
  • Limits and NOT cut records, and can cut the wrong ones. The date limit gives 479 and NOT school gives 528. The known-paper test shows what both removed.

The MeSH heading Students is broader than university students. Its MeSH record covers anyone enrolled in a school or formal educational program, so step 9 can pull in school-age studies. Skim the results for relevance.

Which database should you search for your field?

Search the databases your own library’s subject guide names, plus a broad index if your topic crosses fields. Recommended databases differ by university and field, so treat this table as examples and check your library guide.

FieldDatabases the guide namesUniversity guide
Health and biomedicineMEDLINE through PubMed; CINAHL for nursing and allied healthCornell University Library
PsychologyPsycINFOPenn State University Libraries
EducationERIC; ProQuest EducationPenn State University Libraries
Business and economicsEconLit; ABI/Inform; Business Source CompleteCornell University Library
SociologySociological Abstracts; Sociology Source UltimateCornell University Library
Computing and electrical engineeringACM Digital Library; IEEE XploreUniversity of Wisconsin-Madison Libraries
Topics that cross fieldsScopus; Web of ScienceUniversity of Bath Library

The Cornell and Penn State pages are written for systematic reviews, and the guides we opened give no fixed number of databases for a thesis. The University of Reading says coverage varies, so search more than one to avoid missing useful references, and tells new PhD students to check thesis databases to see whether anyone has already researched the topic. Bath asks of each database whether it covers your topic, what type of literature it holds and which journals it indexes. Cornell’s list also names regional indexes, such as IMSEAR for journals in the South-East Asia Region of the World Health Organization.

Adapt the string for each database. UCL notes that every database works a little differently, and Imperial College London suggests testing the strategy in each one to see that it translated correctly. Victoria University’s page on running your search on other databases says each database has its own syntax, operators and field codes, so rewrite subject headings and field tags such as [tiab] for each one.

Should you use Google Scholar or library databases?

Use library databases for the structured string, and Google Scholar as a supplement, for example to see who cites a paper. Four university guides explain why:

  • University of California, Santa Cruz: its Google Scholar guide lists as a drawback that you cannot search or limit by subject headings, and calls Scholar a supplement to library databases.
  • University of Bath: its guide says you do not know what Google Scholar indexes, whereas for a library database you can find out which journals it covers.
  • UCL: its page on search engines says they are not designed for structured searching, and that it is not entirely clear which sources Google Scholar includes or how it ranks them.
  • University of California, Merced: its page on when to stop suggests Google Scholar or Web of Science to see who has cited the articles you found.

How do you know if your search missed something?

Test it against a few papers you already know are relevant. If the string does not find them, read their records to see why.

Imperial College London calls this set a gold set and describes it as around 5 to 10 highly relevant papers (Imperial’s guide, written for systematic reviews). We picked five papers on the worked-example topic from the literature and ran the strings from the table against them. The table shows each paper against four of the strings, and each paper links to its DOI.

Known paperExact phrases (step 5)Synonyms with OR (step 6)MeSH headings only (step 8)Headings or words (step 9)
Trockel, Barnes and Egget (2000)MissedFoundMissedFound
Curcio, Ferrara and De Gennaro (2006)MissedMissedMissedFound
Gaultney (2010)MissedFoundMissedFound
Hershner and Chervin (2014)MissedFoundMissedFound
Okano and colleagues (2019)MissedFoundMissedFound
Found0 of 54 of 50 of 55 of 5

Each miss has a reason you can see in the record:

  • Exact phrases found none. None of the five has “university students” in its record. Four say “college students”.
  • Synonyms found four. The abstract of Curcio and colleagues refers to students from school to university and uses none of the string’s population phrases. Its MeSH heading Students is what catches it in step 9.
  • MeSH alone found none. Two of the five are PubMed records not indexed for MEDLINE, so they have no MeSH terms at all. The PubMed User Guide notes that not every citation is indexed for MEDLINE. None of the five has the Academic Performance heading.
  • Headings date from when they were introduced. The MeSH record for Academic Performance gives 2018 as the year introduced, with Educational Measurement (1972 to 2017) as previous indexing. PubMed’s help page says articles are indexed with the vocabulary in place at the time, which fits UCL’s advice to pair headings with free-text words.
  • The date limit found none. All five were published between 2000 and 2019, so the 2021 to 2026 limit removed them. Choose known papers that fit your limits, or ask whether you need the limit.
  • NOT school lost one. Curcio and colleagues mention schools in the abstract, so the NOT removed that paper. UCL’s checklist for checking a search recommends avoiding NOT.

To run the test on your own search:

  1. Choose about five papers you know are relevant, such as ones from your scoping search or your supervisor’s reading list. Use only papers that the database holds.
  2. Note each paper’s PMID, the record number PubMed shows for it (Victoria University’s guide describes this method).
  3. Search (PMID1[pmid] OR PMID2[pmid] OR PMID3[pmid]) NOT (your search string). Any paper that comes back is one your string missed.
  4. Open each missed record, find the word or heading your string lacks, add it, and run the test again.

When should you stop searching?

Stop when your known papers are all found, new searches keep returning papers you have already seen, and you have checked reference lists and citations. The University of California, Merced says that seeing the same articles over and over means you have done your due diligence. UCL adds that if you have enough high-quality material, including some key sources, it may be time to stop.

Use these five checks as your stop rule:

  • Known papers: every one is found, or you know why one is missing.
  • Repeats: new searches mostly return articles you have already seen (UC Merced).
  • Databases and terms: you searched the relevant databases with a variety of keywords and subject headings (UC Merced).
  • Citations: you read the reference lists of key articles and checked who cites them (UC Merced).
  • Enough: you have high-quality material, including key sources, to inform your work (UCL, understanding the nature of your search).

UC Merced adds that an article may still slip through, but these steps make it unlikely that an important one does. If your degree needs a systematic review, the rules are stricter: compare the options in systematic review vs literature review.

How many results are too many?

There is no right number, UCL’s checking guide says. A high number of mostly irrelevant results means the strategy needs to be more focused, and a very small number of almost all relevant results means it is probably not wide enough.

Imperial suggests ending up with around or under 1000 results in a database, to avoid a long screening process, but says the right number depends on the review. Its figure is for systematic reviews, so ask your supervisor what your department expects. Our step 9 string returned 759.

The PubMed User Guide lists three ways to cut results: use more specific terms, add terms, or use the sidebar filters for publication date, full text and article type. The date limit above removed all five known papers, so apply limits after the known-paper test. Once the list is the right size, filter it with five filters for evaluating sources.

How do you keep your search up to date?

Save the search, set an alert on it, and rerun the full string before you submit. The University of Reading lists search alerts on databases as the first way to keep a review current over a long project.

In PubMed, “Create alert” sits under the search bar and needs a My NCBI sign-in (PubMed User Guide). UCL says most databases that index journal literature offer email alerts (UCL’s page on databases). Write the date of each rerun in your log, and see how to organize sources for a PhD for where the results should go.

What should you write down about a search?

Write down the databases with platform and date range, the date you ran each search, the full string pasted exactly with the number of records, and any filters. That is the list on UCL’s page on recording your search methodology. UCL says a record lets you retrace your steps, report the search in your methods or an appendix, and let a reader repeat it.

Copy each string out of the database instead of retyping it. The log below holds these fields, plus a column for the known-paper test.

Printable search log

Fill in one row for every search you run, paste the string exactly as the database ran it, and keep the page with your thesis files.

DateDatabase and platformSearch string, pasted exactlyLimits and filtersResultsKnown papers foundNotes and changes

Before you stop searching, tick each line:

  • Every concept has synonyms, spelling variants and, where the database has them, subject headings.
  • The string has been run in each database I chose and adapted to that database’s rules.
  • My known papers are all found, or I know why one is missing.
  • I read the reference lists of key papers and looked up who cites them.
  • New searches mostly return papers I have already seen.
  • Alerts are set, and the date of my last full rerun is written down.

Frequently asked questions

How many databases should I search for a literature review?The guides we opened give no fixed number. The University of Reading says coverage varies between databases, so search more than one to avoid missing useful references. Choose them from your own library’s subject guide, and ask your supervisor what your department expects.
Is Google Scholar enough for a thesis literature search?The University of California, Santa Cruz and UCL treat it as a supplement. Santa Cruz says you cannot search or limit by subject headings in it, and the University of Bath says you cannot tell what it indexes. Use it alongside a library database.
What is the difference between a keyword and a subject heading?A keyword is a word you type, and it finds records that use that word. A subject heading is a standard label the database gives a record whatever words the authors used (UCL). Use both for each concept, because headings can be missing or new.
Should I use the NOT operator?Use it with caution. UCL warns that a search for teenager NOT adult also drops papers that discuss both groups, and its checking guide recommends avoiding NOT. In our PubMed example, NOT school removed one of five known papers.
How do I keep a literature search up to date?Set a search alert and rerun the full string before you submit. The University of Reading lists database search alerts first. In PubMed, “Create alert” sits under the search bar and needs a My NCBI sign-in. Note each rerun’s date in your log.
Do I need a systematic review protocol for a thesis?That is for your department and supervisor to decide, so check your program’s handbook. UCL says it is crucial to report a systematic review’s search strategy in full detail for each source. The steps and log in this guide still help with any literature search.

Sources

How this guide was made: Thesisroom drafted this guide with the help of AI writing tools. Every fact, figure and source in this post was checked against its original source before publishing.