A search strategy is the set of searches that produced your records: which databases, which terms, how they were combined, which limits were applied and on what date each was run. It is the part of the method that has to be reproducible, and the part most often reconstructed badly at write-up from a browser history. This page is about building one deliberately and recording it as you go, which costs nothing at the time and is close to impossible afterwards.
Build it in blocks
One block per concept in your question, each a list of synonyms and controlled vocabulary terms joined with OR, then the blocks joined with AND. Two or three blocks is usually right; four is usually one too many and quietly loses studies. Build the blocks separately, look at the count each returns on its own, and only then combine, because a block returning almost nothing is a block with a missing synonym.
The two tests that tell you it is working
First, the known-item test: pick three or four papers you already know belong in the review and check the search finds them. If it misses one, look at why, because the reason is usually a term worth adding. Second, the precision check: skim the first fifty results and see how many are plausible. Under about one in ten and the search is too broad to screen sensibly.
What has to be recorded
For every database: the exact string, any limits, the date it was run and the number of records returned. Not a description of the search, the string. Databases change, interfaces change, and a strategy described in prose cannot be rerun. Record it at the moment you run it; the funnel worksheet on this site turns those record counts into the screening hours they imply.
Beyond databases
Reference lists of the reviews you already have, forward citation searching on key papers, trial registries where they exist, and a targeted look for unpublished work all find studies databases miss. Each should be recorded as its own line with its own yield, because a reader assessing your review will want to know whether you looked outside the obvious places and what it produced.
When to ask a librarian
Earlier than most people do, and it is usually free at a university. A subject librarian will improve a search more in half an hour than a week of your own iteration, mostly through controlled vocabulary you did not know existed. If your review is going to be published, having a librarian review the strategy is increasingly expected and occasionally required.
Updating the search before you submit
Months pass between the search and the writing, and reviewers routinely ask whether the search was rerun. Plan for one update close to submission: the same strings, the same databases, a date limit from the original run, and a line in the method saying what it added. It is an afternoon if the strings were recorded properly and a rebuild if they were not, which is the most concrete argument there is for writing them down at the time.
Questions people ask about search strategy
How many databases should I search?
Enough to cover the field, which in most health topics means at least two or three plus a check of reference lists. The honest constraint is screening time: every extra database adds records to read, so the funnel worksheet is worth running before adding a fourth.
Should I use controlled vocabulary or keywords?
Both. Controlled vocabulary catches papers indexed properly and free-text keywords catch recent or badly indexed ones, and each misses what the other finds. Combining them within each block is standard.
How do I know my search is good enough?
Run the known-item test: check it finds three or four papers you already know belong. A miss points at a missing synonym, and fixing those is what turns an adequate search into a defensible one.
Do I need to record the exact search string?
Yes, with the database, the date and the number of records. A description in prose cannot be rerun, and reproducibility is the whole point of the strategy.