PRESS — Peer Review of Electronic Search Strategies — is a structured checklist for reviewing systematic review search strategies before they are run. It was developed by Sampson, McGowan, and colleagues, formalized through the McMaster Optimal Search Filters work, and updated in 2016 (McGowan et al., JCE). Its seven evaluation criteria cover translation of the research question, Boolean logic, subject headings, free-text terms, spelling/syntax, limits/filters, and adaptation across databases.

Is PRESS only for librarians?

No, but it works best with one. The 2016 PRESS guidance is written for librarians and information specialists, who have the syntax fluency to spot most errors. Reviewers without library backgrounds can use the checklist effectively for the higher-level criteria (translation of question, Boolean logic, subject heading vs free text balance) and should pair with a librarian for the syntactic checks. Several recent tools — including AI-assisted strategy review — automate the syntactic layer.

Where do search strategies typically fail PRESS review?

Across the published evaluation studies (Sampson et al. 2008; McGowan et al. 2016; Salvador-Oliván et al. 2019), the most common failures are: missing key concepts from the research question, over-reliance on subject headings without free-text equivalents, missing important spelling variants (UK vs US English), incorrect Boolean parentheses producing unintended logic, and inappropriate limits (date ranges, language filters) that introduce selection bias.

Does PRESS apply if my search is generated by AI?

Yes — arguably more strictly. AI-generated search strategies tend to look fluent and complete while missing important spelling variants, MeSH term updates, or domain-specific syntax. The same seven PRESS criteria apply. The 2025 Cochrane AI position treats AI-generated searches as low-risk if they are PRESS peer-reviewed before running, and high-risk if they are not.

How long does PRESS review take?

A trained librarian typically completes a structured PRESS review in 60–120 minutes per database for a moderately complex search. AI-assisted strategy review can identify syntactic errors and missing spelling variants in seconds, but does not replace the conceptual review (criteria 1–2). Plan for one full PRESS review pass before screening begins, and a second targeted pass if the strategy is significantly amended.

The PRESS Checklist: Why Search-Strategy Peer Review Catches What Reviewers Miss

The single highest-leverage moment in a systematic review is the search. Everything downstream — screening, extraction, synthesis — operates on the records the search returned. Records that are not in the search cannot enter the review.

This makes search strategy the most consequential and most error-prone step. Sampson and McGowan have run a long methodological program quantifying just how often these errors happen and what they cost. The conclusion is uniform: structured peer review of the search strategy, before it is run, is the single most cost-effective quality intervention in evidence synthesis.

That structured peer review has a name. It is PRESS — Peer Review of Electronic Search Strategies.

What PRESS is, in one paragraph

PRESS is a 7-criterion checklist for evaluating an electronic search strategy. It was first published as PRESS 2008 (Sampson et al.) and updated as PRESS 2015/2016 (McGowan et al., JCE) following a Delphi consensus process. The checklist is reviewer-facing: a librarian or information specialist runs through the seven criteria and produces a structured set of suggestions before the search is executed.

The seven criteria are:

#	Criterion	What it checks
1	Translation of the research question	Does the search reflect the PICOS/PECO concepts?
2	Boolean and proximity operators	Are AND/OR/NOT used correctly? Are parentheses unambiguous?
3	Subject headings	Are MeSH/Emtree terms appropriate, current, and exploded where needed?
4	Free-text terms	Are synonyms, abbreviations, and natural-language variants captured?
5	Spelling, syntax, and line numbers	Are spelling variants (US/UK), wildcards, and database-specific syntax correct?
6	Limits and filters	Are date, language, study-type, and other limits justified and appropriate?
7	Adaptation for multiple databases	Has the strategy been correctly translated to each database's syntax?

The reviewer rates each criterion (no revision needed / revision suggested / revision required) and provides specific suggestions where revision is needed. The author then revises and, optionally, returns for a second-pass review.

The empirical case for PRESS

Three pieces of evidence collectively make PRESS the standard.

The first is Sampson et al. (2006, 2008), who developed the original criteria and showed that structured review identified errors in roughly 90% of searches that went unreviewed. The errors were not minor — many would have materially changed the records returned.

The second is the McGowan et al. (2016) update, which Delphi-validated the checklist with information specialists and presented further evidence that PRESS-reviewed searches return 8–15% more relevant records than unreviewed equivalents. The update also clarified scoring, distinguishing "revision suggested" from "revision required" to make the checklist easier to apply consistently.

The third is the Salvador-Oliván et al. (2019) audit of 100 published systematic reviews, which found that 90.6% of search strategies had at least one error meeting PRESS criteria — most commonly missing concepts, missing free-text terms, and incorrect Boolean parentheses. Most of these reviews had been published in peer-reviewed journals; the journal peer review process did not catch the search errors.

These three findings together support a strong claim: search errors in systematic reviews are the rule, not the exception, and the journal peer-review process does not detect them. PRESS is the corrective.

What PRESS catches, by criterion

A walkthrough of the seven criteria with the failure modes each is designed to detect.

1. Translation of the research question

The most common failure is a missing concept. If your PICOS includes "older adults" as the population and the search omits a synonym set for older adults (e.g., elderly, geriatric, aged, senior), the search misses every record where authors used those terms exclusively.

A second common failure is concept overlap. A search for "AI in screening" that uses "machine learning OR artificial intelligence OR AI" without specifying the screening context returns thousands of irrelevant records about AI in adjacent domains.

PRESS criterion 1 asks: does each PICOS concept have a search block, and are blocks combined with AND?

2. Boolean and proximity operators

The classic error is parentheses. The search cancer OR tumour AND screening reads as cancer OR (tumour AND screening) to most databases — a logical error that returns the entire cancer literature plus a tiny intersection of tumour-related screening. The intended search is (cancer OR tumour) AND screening.

PRESS criterion 2 asks: are operators correctly used and are parentheses unambiguous? A reviewer reads each line and checks the implied logic.

3. Subject headings

MeSH and Emtree terms have hierarchies and update annually. A search using "Coronavirus Infections"[MeSH] without exploding to include child terms misses records indexed under more specific headings. A search using "COVID-19"[MeSH] from a 2019 strategy still works, but a search using a 2019 MeSH term retired in 2022 returns nothing.

PRESS criterion 3 asks: are subject headings current, appropriate, and exploded where needed?

4. Free-text terms

Subject headings cover indexed records. Many records (especially in newer databases or grey literature sources) are not indexed or are partially indexed. Free-text terms in title and abstract complement subject headings. A search using only "Telehealth"[MeSH] misses records using "telemedicine," "remote consultation," "virtual visit," "tele-consultation," and the dozen other variants the literature uses.

PRESS criterion 4 asks: does each subject heading have a corresponding set of free-text variants?

5. Spelling, syntax, and line numbers

UK vs US English: tumour and tumor, behaviour and behavior, paediatric and pediatric. Database-specific syntax: PubMed's [tiab] vs Embase's :ti,ab vs Cochrane's :ti,ab,kw. Wildcards: child* matches child, children, childhood; in some databases, the wildcard rules differ.

PRESS criterion 5 is where most syntactic errors are caught. It is also the criterion most amenable to automation — AI-assisted strategy review can flag missing spelling variants and incorrect syntax in seconds.

6. Limits and filters

Common errors: a date filter that excludes pre-2010 records when the protocol does not specify a date range; a language filter limiting to English when the protocol allows multilingual studies; a study-type filter that imposes a methodological cut not in the eligibility criteria.

PRESS criterion 6 asks: are all limits justified by the protocol?

7. Adaptation for multiple databases

A PubMed search uses MeSH; an Embase search uses Emtree; a CINAHL search uses CINAHL Subject Headings. Direct translation of one database's syntax to another almost always fails. The most common error is using PubMed's MeSH terms in an Embase search, where they are not recognized.

PRESS criterion 7 asks: has each database-specific version been correctly adapted?

How to operationalize PRESS without a librarian

The honest answer is: pair with a librarian if you have access to one. Most universities and academic medical centers have an evidence synthesis librarian; many will conduct PRESS review as part of their service. The marginal cost of a structured review by a trained reviewer is much smaller than the cost of a search miss.

If you do not have access to a librarian, three practical approximations.

Self-review with the checklist. Run through the seven criteria yourself, line by line. The criteria are explicit enough that even an inexperienced reviewer catches a meaningful fraction of errors — Salvador-Oliván et al. found that the median search has multiple errors, so even partial detection is valuable.

Cross-team review. Have a colleague unfamiliar with the search run the checklist. Fresh eyes catch what the search author has stopped seeing.

AI-assisted strategy review. Modern AI tools, including mapped's literature-search tooling, can run automated checks on PRESS criteria 5 (spelling, syntax) and provide structured suggestions on criteria 3 (subject headings) and 4 (free-text terms). They cannot replace human conceptual review on criteria 1 and 2, where domain knowledge is required.

The 2025 Cochrane AI position is explicit: AI-generated or AI-assisted searches should be PRESS peer-reviewed before running. AI fluency is not a substitute for methodological correctness. For the broader policy framework, see Responsible AI in Systematic Reviews.

Where PRESS sits in the protocol timeline

The scheduling matters. PRESS belongs before the search runs at scale and after the protocol is locked.

A defensible workflow:

Lock the protocol (PICOS, eligibility criteria, included databases). Register on PROSPERO or OSF.
Draft the search strategy.
Pilot the search on one database (typically PubMed) and review the first 50 records for relevance plausibility.
Submit for PRESS review. Allow 1–2 weeks for a librarian; minutes-to-hours for AI-assisted review on syntactic criteria.
Revise per PRESS feedback.
(Optional) Second-pass PRESS review on the revised strategy.
Translate to all databases. Cross-check translation per criterion 7.
Run the search at scale. Document the final strategy and the PRESS revisions in the protocol amendment trail.

Steps 4–6 are the PRESS gate. Skipping them is a common time-saver that costs much more later, when the review has to be re-run with a corrected search and the team explains the protocol amendment to a journal reviewer.

What to publish from the PRESS process

PRISMA-S (the search-reporting extension to PRISMA 2020) requests transparent reporting of the search strategy. The convention that has crystallized in 2025–2026 is to publish four artifacts.

The full final search strategy for every database, in supplementary material.
A statement that PRESS peer review was conducted, who conducted it, and the date.
The PRESS reviewer's structured feedback (in supplementary material if substantial).
Any post-PRESS amendments to the strategy, with rationale.

Together, these four make the search reproducible and the review-quality story defensible.

Common failure modes in 2026

Two patterns we see repeatedly in protocols submitted for PRESS review.

"We'll do PRESS at the end." This inverts the purpose. PRESS at the end is post-hoc audit; PRESS before the search is correction. The errors PRESS catches are designed to be caught when correction is cheap.

"We used AI, so we don't need PRESS." AI fluency masks errors more than it reveals them. AI-generated search strategies tend to look polished and complete while missing important spelling variants and using outdated subject headings. PRESS is more important for AI-generated searches, not less.

Putting it to work this week

Three concrete steps for your next review:

Add "Search strategy will be PRESS-reviewed before execution" to the protocol. Register the commitment.
Identify your reviewer in advance — a librarian if available, a colleague otherwise, AI-assisted review for the syntactic layer regardless.
Build the PRESS feedback artifact into the supplementary materials plan from day one. The artifact is the audit trail.

The seven criteria are not new. The methodological evidence behind them is two decades old. The only thing PRESS asks of a review team is the discipline to apply them before the search runs, not after.