My master's thesis hinged on a systematic literature review of peer tutoring in higher education. I had never done a systematic review. My entire search strategy, in week one, was typing 'peer tutoring university' into Google Scholar, getting 180,000 results, and feeling — I am being honest — productive. I spent the next six weeks in a state of motion that resembled progress: reading papers semi-randomly, periodically discovering an entire subfield I'd missed because I'd never searched the term it lived under.
The thing nobody had explained to me — or that someone had explained and I'd nodded through while thinking about lunch — is that a systematic review is supposed to be reproducible. Another researcher following your exact method should land on the same set of papers, give or take. That means your search terms, your databases, and your inclusion and exclusion criteria all have to be written down, justified, and defensible before you start reading anything. I learned this in week five, from a committee member's slowly rising eyebrow during what I'd assumed would be an encouraging check-in. My review had started in week one. The eyebrow was correct and I have not forgotten it.
What I'd actually built by week five was not a systematic review. It was an enthusiastic reading list assembled by following citation trails wherever they felt interesting, which is a perfectly nice way to learn a field and a completely indefensible way to claim you've surveyed it. When my committee member asked, gently, 'so how would someone replicate your search?', the honest answer was 'they couldn't, because I have no idea what I searched, and Google Scholar shows different results to different people anyway.' That sentence is where a systematic review goes to die.
The synonyms I never thought to search
When I finally rebuilt the strategy properly, I used Gemini 2.5 Pro with the Create a Reproducible Search Strategy for a Systematic Review prompt — topic of peer tutoring in higher ed, discipline of educational psychology, scope of English-language peer-reviewed work from 2010 on. I went with Gemini because the output is a set of structured tables and Boolean strings, and Gemini handles that long, organized, multi-part deliverable cleanly without losing the thread between section three and section six. The CONCEPT MAP table is where it earned its keep.
I had been searching 'peer tutoring' and entirely missing 'supplemental instruction,' 'near-peer learning,' and 'peer-assisted learning strategies' — three terms that retrieve overlapping-but-different bodies of literature, with notes in the table explaining how each differs from plain peer tutoring. Those were not edge cases. Each one pulled in papers central to my argument that I'd never seen, because I'd been searching the topic in only one of its dialects. That single table would have changed the shape of my entire review.
- Synonyms and variant terms are the hardest, most-skipped part — the concept map exhausts them for you
- Discipline-specific databases (ERIC for education) surface papers Google Scholar buries on page 40
- Inclusion/exclusion criteria, written first, stop the scope creep that eats weeks on irrelevant papers
- The PRISMA workflow is what makes your process defensible when a committee asks 'how did you find these?'
The mistake note I needed in week one
The COMMON MISTAKE section named the exact error I'd been living: treating Google Scholar as a primary database rather than a supplementary one, and not searching the field's dedicated index. It also reminded me that Google Scholar's results aren't reproducible across users and time, which is fatal for a 'systematic' review — another thing I'd have failed an audit on. Reading that in week one instead of hearing it from a frowning committee would have spared me a genuine amount of dignity.
The honest caveat the prompt always prints is the one I respect most: this is a starting strategy, and you must verify each term against the database's own thesaurus before the real run. MeSH terms in particular are finicky, and a strategy that looks right can quietly miss things until you check it against PubMed's actual controlled vocabulary. The prompt gets you to a defensible draft in ten minutes; the verification is still your job, and it should be.
My university offered a two-hour library workshop on exactly this. It was excellent and I attended it in week seven, which is roughly when a fire extinguisher arrives after the kitchen. This prompt covers the same ground in ten minutes, on demand, at 11pm when the panic actually hits. Grab the Create a Reproducible Search Strategy for a Systematic Review prompt on Prompt Dock, give it your topic and scope, and start your review with a strategy instead of a guess. Then run my Find the Real Gaps in a Literature Review prompt on whatever you pull, so you're reading for absence, not just accumulation.