How to Search PubMed the Right Way: MeSH Terms, Boolean Operators and AI

The Google Method

Dr. Hammad Ali arrived twenty minutes early, which for him was a new record, and was typing furiously when Dr. Sumaira Talib walked in from her ward round.

“Searching for research,” he said proudly, before she could ask. The browser tab read Google. He had typed: is gallbladder surgery painful? and was reading a news article from 2019. A second tab was open on YouTube: lap chole shoulder pain explained.

Sumaira looked at the screen, then at me. I put down my chai and pulled up a chair.

“Hammad,” I said, “this is what I call the Google Method. I say it with full affection: it will not get you a synopsis approved, and it will not get you published.”

Dr. Junaid Rashid dropped into his usual chair. “Fifteen years of medicine and I have never done a literature search,” he said. “I am almost proud of it.” He paused. “Almost.”

Why You Search Before You Write

Last week every learner tested a topic against FINER. Zunaira’s question was the one that stayed with us: how do you know your study is novel? The honest answer is that you cannot know until you have searched. A literature search does four jobs:

• It tells you whether your question has already been answered, and how well.
• It shows the gap your study will fill. Supervisors, ethics committees, the CPSP and UHS reviewers and journal editors all look for this.
• It gives you the studies and statistics for your introduction and, later, for your discussion.
• It shows you the designs, definitions and tools other researchers used, so you do not reinvent them.

“So it is not just reading a few papers before starting?” Sumaira asked.

“No. Reading comes after searching. The search is a method, and a method can be repeated, checked and defended.”

Where to Search

Dr. Hassan Raza raised his hand. “Sir, just tell me the databases.”

• PubMed: free, maintained by the US National Library of Medicine, with more than 37 million citations. This is your main tool and the first place your supervisor and reviewers will look (pubmed.ncbi.nlm.nih.gov).
• Google Scholar: easy and wide, including theses and conference papers, but it mixes peer-reviewed papers with everything else. Use it to find things, never as proof that a journal is credible.
• Cochrane Library: systematic reviews of interventions. If you want to know whether a treatment works, look here early (cochranelibrary.com).
• Pakistani sources: many Pakistani journals are in PubMed, but not all, and many of their records carry no MeSH terms. Search PakMediNet, Google Scholar and the journal websites directly for local studies.
• Scopus, Web of Science and Embase: subscription databases that cover journals PubMed does not. Use them if your institution has access.
• AI tools: Consensus, ChatGPT and Claude can make searching much faster, if you use them the right way. We will come to that shortly.

Hammad asked about ResearchGate. “It can help you get a full text that the author has uploaded,” I said. “It is not a search engine. Search in the databases, then find the paper.”

Zunaira and the Tens of Thousands

Two evenings earlier, I had found Dr. Zunaira Malik alone in the research room at quarter to ten.

“Sir, I have found tens of thousands of papers.”

She had typed oral lesions into PubMed. Her MPhil topic is areca nut and gutka chewing and oral submucous fibrosis (OSMF). Almost none of those tens of thousands were about her question.

“By the time you finish this chai,” I told her, “you will have a list you can actually read, and you will know why every paper is on it.”

By the next morning she had become the group’s PubMed expert. This is what she learned.

MeSH: PubMed’s Own Language

MeSH (Medical Subject Headings) is the controlled vocabulary of the National Library of Medicine. Papers indexed for MEDLINE are tagged with standard MeSH headings, so a paper that says “OSMF”, “OSF” or “submucous fibrosis” carries the same heading: Oral Submucous Fibrosis. Since 2022 this indexing has been done mostly by automated methods, with human review.

When you type ordinary words without a tag, PubMed uses Automatic Term Mapping: it tries to match your words to a MeSH heading, a journal or an author, and also searches them in all fields (1). That helps, but it does not always choose the heading you meant.

MeSH alone is not enough either. New records, and many journals including several Pakistani ones, have no MeSH indexing yet. So a good search always pairs the MeSH heading with free-text words in the title and abstract, tagged [tiab], joined by OR.

To find a heading: go to ncbi.nlm.nih.gov/mesh, type your concept, open the heading, read its definition, and use “Add to search builder”. For Zunaira that gave:

"Oral Submucous Fibrosis"[MeSH] OR "oral submucous fibrosis"[tiab] OR OSMF[tiab]

Three Words on the Whiteboard

The next morning I wrote three words on the board: AND, OR, NOT.

“Like algebra?” Hammad asked.

“Exactly like algebra. Except the answer is a set of papers.”

• OR broadens. Use it to join synonyms of the same concept. Authors describe chewing habits in many ways, so Zunaira used: "Areca"[MeSH] OR "areca nut"[tiab] OR "betel quid"[tiab] OR gutka[tiab] OR gutkha[tiab]
• AND narrows. Use it to join different concepts. Every paper must contain all of them.
• NOT excludes. Use it rarely. ... NOT "Mouth Neoplasms"[MeSH] looks tidy, but OSMF is a potentially malignant disorder and many OSMF papers also mention oral cancer. NOT would throw them all away, including the ones she needed.

Parentheses group the synonyms before you join the concepts:

("Oral Submucous Fibrosis"[MeSH] OR "oral submucous fibrosis"[tiab]) AND ("Areca"[MeSH] OR "areca nut"[tiab] OR gutka[tiab]) AND ("Pakistan"[MeSH] OR Pakistan[tiab])

Tens of thousands became a short list. “This I can read,” Zunaira said quietly. “And they are all about my topic.” The exact number changes every day as new papers are added, which is why you record it with the date.

Other Field Tags Worth Knowing

• [tiab]: title and abstract. Use it for new terms and for records without MeSH.
• [ti]: title only. Very narrow, useful when a search is still too large.
• [au]: author, for example a researcher whose work you keep citing.
• [ta]: journal, for example "Pak J Med Sci"[ta].

“COVID-19 was not in MeSH on day one,” I told them. “In early 2020, anyone searching had to use free-text words in [tiab]. That is exactly why we always join MeSH and free text with OR.”

Building the Full Strategy: Sumaira’s Search

Sumaira volunteered her CPSP dissertation question: the frequency of surgical site infection after laparoscopic appendicectomy. We built it together, step by step. Use the same steps for every search you do.

1. Break the question into concepts. Surgical site infection; appendicectomy; laparoscopy. A frequency study has no comparison group, so there is no C to search.
2. Find the MeSH heading for each concept. “Surgical Wound Infection”, “Appendectomy”, “Laparoscopy”. Then list the free-text words authors actually use, including the British spelling “appendicectomy”.
3. Join synonyms with OR and concepts with AND.
   ("Surgical Wound Infection"[MeSH] OR "surgical site infection"[tiab] OR "wound infection"[tiab]) AND ("Appendectomy"[MeSH] OR appendectomy[tiab] OR appendicectomy[tiab]) AND ("Laparoscopy"[MeSH] OR laparoscopic[tiab])
4. Apply filters from the left panel (below).
5. Screen titles and abstracts, then read in full only the papers that pass.
6. Record and save the search (below).

Junaid tried his own question next, on medication non-adherence among hypertensive patients: ("Hypertension"[MeSH] OR hypertension[tiab]) AND ("Medication Adherence"[MeSH] OR adherence[tiab] OR compliance[tiab]) AND ("Pakistan"[MeSH] OR Pakistan[tiab]). He scrolled for a long minute. “There are studies from other cities,” he said. “Different tools, different patients, different settings. Very little from a medicine OPD like ours.” He looked up. “So that is what a gap looks like.”

Filters: Useful, but Use Them With Thought

• Publication date: the last five to ten years for most clinical questions. CPSP and UHS synopses expect recent references, so your search should find recent papers first. Older landmark studies still have a place.
• Article type: Randomized Controlled Trial, Systematic Review, Meta-Analysis or Observational Study, depending on what you need.
• Species: Humans, for clinical questions.
• Language: English, unless you read another language well.

Hassan once put a French paper in his reference list. It was relevant. He could not read it. He cited it anyway. His supervisor asked him what the paper found, and that was the end of that.

A warning about filters: they depend on indexing, and records without MeSH may be dropped by some of them. If a filter removes a paper you know is relevant, check it by hand.

Let AI Build the Query, Then Check It

Hammad had been quiet for a while, which usually meant he was on his phone. “Sir, I asked ChatGPT to write my PubMed search. It took ten seconds.”

I expected to be disappointed. I was not. The string had MeSH headings, [tiab] words, OR inside each concept and AND between them.

“This is one of the best uses of AI in research,” I said. “Building a PubMed query by hand takes a beginner an hour. ChatGPT or Claude can draft it in seconds, and explain every part of it if you ask. But a draft is not a search. You still check it.”

Give the AI your full question, not a one-word topic, and tell it exactly what you want. This is the prompt Junaid wrote:

I am planning a cross-sectional study in a medicine OPD in Lahore. My question: what is the frequency of medication non-adherence among adult patients with hypertension? Write a PubMed search string. For each concept, join the MeSH heading and free-text [tiab] synonyms with OR, and join the concepts with AND. Do not use NOT. List every MeSH heading you used separately so that I can check it.

Then check before you trust:

• Check every MeSH heading in the MeSH database (ncbi.nlm.nih.gov/mesh). AI tools sometimes give a heading that looks right but does not exist, or one that has been renamed. A wrong heading quietly finds nothing.
• Check the syntax: every bracket closed, quotation marks around phrases, field tags spelled correctly.
• Run it in PubMed yourself. If there are too many or too few results, tell the AI the number and ask it to narrow or widen the search. Two or three rounds are usually enough.
• Read the first twenty titles. If several are off-topic, one of the synonyms is wrong. Ask the AI which term is pulling them in.
• Save the final string in your search log exactly as you ran it, with the date.

“And never ask it for a list of papers,” I added. “Ask it for a search string. A search string can be checked in PubMed in one minute. A reference list from a chatbot can contain papers that do not exist.”

Zunaira had tried a different tool. Consensus (consensus.app) is an AI search engine built on published research papers. You ask a question in plain English, and it answers with links to the actual studies it found. For a yes-or-no question such as Does areca nut chewing cause oral submucous fibrosis? it also shows how the studies lean.

“It is a good first look,” I said. “In five minutes you see what is known, and it leads you to papers you might have missed. But it does not replace PubMed. Its coverage is different, you cannot report it as the search strategy in your synopsis, and its summaries are only as good as the papers underneath. Open every paper it gives you, read it, and check the journal.”

So the division of work is simple: Consensus for a first look at the evidence, ChatGPT or Claude to draft and fix your PubMed query, and PubMed for the search you actually run, record and report. One rule applies to all of them: a research question is fine to share, a patient list never is.

Save It, Log It, Update It

Hassan was already opening a new tab. “Sir, is there a shortcut?”

“There is one legal shortcut. Create a free My NCBI account, run your search, and click Create alert under the search box. Name it clearly, for example SSI_LapAppendicectomy_Oct2026. PubMed will email you new papers that match. When you write your discussion months from now, your literature will already be up to date.”

Then keep a search log. Five columns in Excel are enough: database, exact search string, date searched, filters applied, number of results. A normal synopsis does not print this log, but it lets you repeat the search, answer your supervisor’s questions and update your references before submission. For a systematic review, the full strategy must be reported, and the numbers go into a PRISMA 2020 flow diagram (2). We will come to that in the systematic review session.

Getting the Full Text, Legally

Dr. Bushra Fatima had found a perfect paper for her cohort study on antenatal corticosteroids. “The journal wants forty dollars for one article, sir.”

For a trainee in Pakistan, that is not a small amount. But there are legal routes, and they work more often than people think:

• PubMed Central: look for the “Free PMC article” label in PubMed.
• HEC National Digital Library: full-text access to major publishers through HEC-recognised universities and institutions (digitallibrary.edu.pk). Ask your librarian how access works on campus.
• Research4Life (HINARI): free or low-cost access for eligible institutions. Eligibility depends on the country group and the institution, so check with your librarian.
• Unpaywall: a free browser extension that finds legal open-access versions of the paper you are viewing.
• The corresponding author: a short, polite email usually brings the PDF within days, and sometimes a collaborator.

“And the pirate sites?” Hammad asked.

“No. Not for your synopsis, not for your thesis, not ever.”

Mistakes I See Every Month

• Searching Google instead of a database. Google is for finding restaurants.
• Searching one broad word. “Diabetes” gives you a library. Concepts joined with AND give you a reading list.
• Using MeSH only, or free text only. Use both, joined with OR.
• Using NOT carelessly. It removes more than you think.
• Old references only. A reference list that stops ten years ago tells the reviewer the search was never done.
• Citing from the abstract. Read the full text of everything you cite. Abstracts sometimes say more than the results support.
• Not checking the journal. A paper found in Google Scholar can still be from a predatory journal. Bushra knows this better than anyone now.
• Not recording the search. If you cannot repeat it, you cannot defend it.
• Copying references from a chatbot. Use AI to build the search, then find every paper in PubMed yourself.

By the end of the session every learner had a saved search, an alert and a first log entry. Hammad closed his YouTube tab without being asked.

Junaid was still looking at his results. “I thought research started with data collection,” he said.

“It starts here,” I said. “Research is not rocket science. It is a skill, and skills can be taught.”

Key Takeaways

• Search before you write. The gap your synopsis claims must come from the literature, not from your impression.
• Use PubMed as your main database, and add Google Scholar and Pakistani journal websites for local studies.
• For every concept, join the MeSH heading and free-text [tiab] words with OR. Join different concepts with AND. Use NOT rarely.
• Let ChatGPT or Claude draft your PubMed query and use Consensus for a first look at the evidence, but check every MeSH heading, run the search in PubMed yourself, and never take a reference list from a chatbot.
• Apply filters with thought, then screen titles and abstracts before reading full texts.
• Save your search, set a PubMed alert, and keep a simple search log with the date and number of results.
• Get full texts legally through PubMed Central, the HEC National Digital Library, Research4Life, Unpaywall or the author.

References:

1. National Library of Medicine. PubMed User Guide. Bethesda (MD): National Center for Biotechnology Information; 2026. Available from: https://pubmed.ncbi.nlm.nih.gov/help/
2. Page MJ, McKenzie JE, Bossuyt PM, Boutron I, Hoffmann TC, Mulrow CD, et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ. 2021;372:n71.