River
Y CombinatorBacked by Y Combinator
FREE TEMPLATE

Systematic Review Search Strategy Template

The term register, the syntax for every platform, and the dated log that turns a search into something another reviewer can run.

Free download  ·  No account needed

Search Strategy

1. [Database name]

Platform [interface, e.g. Ovid, Embase.com, EBSCOhost]
Database segment [as the platform reports it]
Date run [the calendar date]
Records retrieved [the platform's own count, before deduplication]

1     exp [Controlled vocabulary term]/
2     ([free text] or [synonym*] or "[phrase]").ti,ab,kf.
3     1 or 2
4     exp [Controlled vocabulary term]/
5     ([free text] or [synonym*]).ti,ab,kf.
6     4 or 5
7     (([concept A terms]) adj3 ([concept B terms])).ti,ab,kf.
8     "[named programme]".ti,ab,kf.
9     7 or 8
10    3 and 6 and 9
11    limit 10 to [stated limit]

Pasted out of the platform's own search history, never retyped. Where an export exists, the export is the record and this page is the transcription.

2. [Next database]

One section per source, including every register, every citation search and every hand search.

The pack is six files and one rule. Three documents hold the search: the strategy pasted per database, the reasoning behind every term and every limit, and a page on which of those facts expire and when. Three sheets hold the record: a term register scored against a seed set, a log with one row per database per run, and a yield sheet showing what each source actually bought. The rule is that nothing gets retyped. The protocol itself, registered before the first search runs, is the protocol registration pack.

Every sheet ships filled in for one illustrative review of manager mental-health training and sickness absence. Seven sources returned 4,833 records across two runs, deduplicating to 2,955 and ending in 19 included studies. CINAHL contributed 196 unique abstracts and no included studies. Citation searching contributed 7 unique records and 2 included studies. Those two rows are the argument for the yield sheet, and no reference manager can produce either of them once the records are imported.

PRISMA-S asks for each strategy copied and pasted exactly as run, the date of the last search, and the records identified from each source. A concept grid produces none of the three, because all three are properties of an event. Open the pack in River and the log gets written while the search runs. Or download the Word documents and CSV sheets and fill them in yourself, then hand the export to the screening pack, which turns those decisions into the PRISMA flow, and the extraction table once screening is done.

Every sheet in the pack

The Search Log, the Yield by Database sheet, and the Term Register.

Search Log

One row per database per run. A re-run appends a row and never edits one.

RunSourcePlatformDate runLimit appliedRetrieved
1MEDLINE ALLOvid2025-11-18yr=2000 -Current1,284
1EmbaseEmbase.com2025-11-18[2000-2026]/py1,673
1PsycINFOOvid2025-11-18yr=2000 -Current908
1CINAHL CompleteEBSCOhost2025-11-19Published 2000-2025412
1Business Source PremierEBSCOhost2025-11-19Published 2000-2025287
1ClinicalTrials.govregister2025-11-19Start 2000 onwards46
1Citation searchingWeb of Science, reference lists2026-01-22None31
2Six databases, re-runas above2026-04-14Entry date 20251101 onwards192
Total retrieved, both runs4,833

Deduplication removed 1,878, leaving 2,955 to screen. The update run used an entry-date limit rather than a publication year, because a 2024 paper indexed in 2026 is invisible to a year limit.

Yield by Database

What each source cost, and what it contributed.

SourceRetrievedUniqueFull textIncludedUnique per include
MEDLINE ALL1,3411,341639149.0
Embase1,744848365169.6
PsycINFO946449132224.5
CINAHL Complete42619650
Business Source Premier296903190.0
ClinicalTrials.gov492420
Citation searching317423.5
Total4,8332,95512619155.5

CINAHL cost 196 unique abstracts and contributed nothing. Citation searching cost 7 and contributed 2. Business Source Premier cost 90 for one study that appears in no health index, which is why the verdict column is an argument rather than an audit.

Term Register

Every whole-strategy version scored against a twelve-paper seed set.

VersionMEDLINE recordsSeeds retrievedKept
A: workplace AND managers AND training1,28412 of 12Yes
B: A AND the sickness-absence concept3916 of 12No
C: A, free text only, no controlled vocabulary1,10211 of 12No
D: A without “mental health first aid”1,2519 of 12No

Version B reads like the tighter search and loses half the literature, because in six of the twelve papers the absence outcome sits in the results and never in the abstract. Nothing about reading version B tells you that. Only the seed column does.

Term blockRecords added to MEDLINESeeds gainedVerdict
team lead*, shift lead*, foreman, forewoman611Keep
organisation*, organization*, company, firm~3,1000Drop
“psychological safety”, resilien*~8,4000Drop

What's in the pack

01

Search Strategy

Every string pasted out of the platform it ran on, with the database segment, the date and the count above it, ready to export as a supplementary file.

02

Search Log sheet

One row per database per run, carrying the pasted string, the platform, the date, the limit and the count. A re-run appends a row rather than editing one.

03

Yield by Database sheet

Retrieved, unique, full text and included per source, plus unique records per included study, so a database that cost 196 abstracts and returned nothing says so.

04

Term Register sheet

Every term and every whole-strategy version scored against your seed set, with what each block cost in records and what it bought in retrieved seeds.

05

Rationale Note

Why each source, why each dropped synonym block, and why each limit, including the limits you deliberately did not apply, which read as omissions when left unstated.

06

How This Search Is Documented

The four facts that stop being recoverable the moment records are exported, which file each one lives in, and the order the files get written.

07

Record the search as run

The standing space rule every prompt reads first. Cochrane's own instruction is that strategies should not be re-typed, because this can introduce errors.

How to use it

  1. 1

    Open in River, or download it

    Open the pack in River and the agent builds the register with you, or download the blank Word documents and CSV sheets instantly.

  2. 2

    Send the question and the seed set

    Your review question, the databases and platforms you can actually reach, and five to fifteen papers you already know belong in the review.

  3. 3

    Build and peer review one strategy

    The register gets scored against the seed set, then one strategy is written and reviewed before translation to any other platform, because an error found after translation has to be found once per database.

  4. 4

    Log every run as it happens

    Paste the string, the date and the count into the log before exporting anything, because two of those three stop existing the moment records reach a reference manager.

Frequently asked questions

Is this template free?

Yes. Download the whole pack as Word documents and CSV sheets with no signup and no credit card. Edit with AI is a separate, optional path that has the agent build the register and write the strategies with you. The template library holds the rest of the packs.

What format are the downloaded files?

Word documents (.docx) for the three documents and the rule, and CSV (.csv) for the three sheets, in one zip. They open natively in Word, Pages, Google Docs, Excel, Numbers and Sheets with nothing to convert. Inside River the same content opens as live Docs and Sheets.

What does 'Edit with AI' actually do?

It creates a free River account and installs this exact pack as a private workspace. It opens asking for your review question, your databases and your seed set, then builds the term register against those papers and writes one strategy for peer review before translating it.

Does River run the searches for me?

No. The strings are written here and executed in the platform your institution licenses, which is the only place a search history export exists. What comes back into the space is the pasted string, the date and the count, and those three are what the reporting standard asks for.

Does this deduplicate my records?

No, and the sheets assume something else does. Yield by Database records the unique count per source and the order deduplication ran in, because that order decides the column. The duplicate merge plan covers a merged library where a citation manager's own matching is the risk.

Is this only for health reviews?

No. The worked example uses MEDLINE, Embase and PsycINFO because most readers of this query work there, and the structure is identical for ERIC, Scopus, Web of Science, Business Source Premier or a legal database. Controlled vocabulary differs per database, which the register already handles as a column.

How is this different from a search strategy template?

A template gives you a concept grid, which is the search you meant to run. Reviews with a librarian as an author still report a reproducible search only 64 per cent of the time. The record is what is missing: pasted string, date, count per source. The synthesis writer takes over once screening ends.

Write down the search while it still exists

Send your review question, your databases and the papers you already know belong in the review. The register comes back scored against them.

Edit with AI