EvidAIEvidAI
Back to Home
12 min read

Search Studio — Step by Step

Federated multi-database search with AI-translated queries, deduplication, and full PRISMA-compliant audit.

Search Studio — Step by Step

Search Studio is the second phase of every SLR (and the first real-work phase of Rapid, Scoping, and Umbrella reviews). It builds a comprehensive, reproducible literature search across the major academic databases and ships every result back into the review with full PRISMA-grade provenance.

This article walks you through Search Studio click-by-click. If you only have time for one section, read Step 1 → Step 6.

Where to find it

Inside any review, sidebar/in-page nav → Search (URL: /slr/search).

The page has four panels:

  1. Query Builder (top)
  2. Database Selector (left)
  3. Results Table (center)
  4. Search History (right drawer)

Step 1 — Open Search Studio inside your review

  1. From the dashboard or My Reviews, open the review you want to work on.
  2. From the SLR hub, click Search (or just go to /slr/search).
  3. If a protocol was loaded from the Vault, you will see a "Loaded from Vault" banner at the top — your search strategy from the protocol is already prefilled.

If this is your first time on the page for this review, the Query Builder is empty and the Database Selector has your default databases checked (you can change defaults in Settings → Profile).

Step 2 — Build the query

You have three ways to populate the Query Builder. Pick whichever matches how clearly you have framed the question.

2a. Auto-generate from your research question

  1. In the Query Builder, paste your research question in natural language. Example:

    "In adults with type 2 diabetes, does GLP-1 receptor agonist therapy reduce major adverse cardiovascular events compared with placebo or usual care?"

  2. Click Generate query.
  3. The Search Strategist agent extracts PICO elements (Population, Intervention, Comparator, Outcome) and proposes:
    • A list of concept blocks
    • Synonyms and controlled-vocabulary terms (MeSH, Emtree, etc.) per concept
    • Boolean structure (AND between concepts, OR within concepts)
  4. Review the suggestion and accept, edit, or reject each concept.

2b. Type concepts manually

  1. Click Add concept.
  2. Name it (e.g. "GLP-1 agonists").
  3. Add synonyms one per line (e.g. liraglutide, semaglutide, dulaglutide, exenatide).
  4. Repeat for each concept (Population, Intervention, Comparator, Outcome).
  5. The builder auto-joins inside each concept with OR and across concepts with AND.

2c. Reuse a saved search

  1. Click the history icon → Saved searches.
  2. Pick a previous strategy (your own or a teammate's, if shared).
  3. Edit as needed.

Concept-block tips

  • Synonyms expand sensitivity — list every term a paper might use.
  • Concepts narrow specificity — too many AND-ed concepts return nothing.
  • Use truncation — diabet* matches diabetes, diabetic, diabetics.
  • Avoid broad NOT — use it only to exclude clearly off-topic terms (e.g. NOT veterinary).

Step 3 — Pick the databases

In the Database Selector on the left, tick the databases you want to search. EvidAI translates your concept-block query into the correct syntax for each one before running it. The most common picks for clinical SLRs are:

  • PubMed / MEDLINE — 35M+ biomedical citations (translated to MeSH + free-text)
  • Embase — pharmacological and clinical (translated to Emtree)
  • Cochrane CENTRAL — RCT-rich (translated to MESH DESCRIPTOR)
  • Scopus — broad coverage (translated to TITLE-ABS-KEY)
  • Web of Science — multidisciplinary
  • CINAHL — nursing and allied health
  • PsycINFO — psychological literature
  • ClinicalTrials.gov + WHO ICTRP — ongoing and unpublished trials
  • Europe PMC + OpenAlex + Semantic Scholar — open metadata for sensitivity

Open the Preview panel to see the exact translated query for each ticked database before you run.

Step 4 — Apply filters (optional)

Below the database selector you can apply standard limits:

  • Date range (publication year)
  • Study design (RCT, observational, etc.)
  • Language
  • Publication type (excludes letters, editorials)
  • Geography

Filters apply to every selected database. If a database does not support a given filter natively, EvidAI applies it post-fetch and labels which results were filtered client-side so the search remains transparent.

Step 5 — Run the search

  1. Click Run search in the Query Builder.
  2. A progress strip appears showing each database in flight (queued → running → completed → counts retrieved).
  3. As databases complete, results stream into the Results Table. Each row shows:
    • Title, authors, year, journal
    • Source database (PubMed, Embase, …)
    • Snippet of abstract
    • DOI / PMID badges
  4. Deduplication runs automatically across databases — exact-DOI matches collapse first; near-duplicate detection flags fuzzier matches for your confirmation in a Possible duplicates tab.

When the run finishes, the page header shows totals: N raw hits across M databases, K duplicates removed, L unique records. These three numbers feed straight into your PRISMA flow diagram.

Step 6 — Review and save the results

You can act on the results in several ways:

  • Save the run. Click Save search. The strategy + results snapshot is stored on the review and will appear on the PRISMA flow diagram.
  • Send to screening. Click Send to screening to move every kept record into the Screening phase. Excluded duplicates are not sent.
  • Export the results. Use the export menu to download as RIS, BibTeX, EndNote XML, CSV, or PubMed XML. Useful for peer review of the search strategy or for an external screening tool.
  • Re-run later. Saved searches can be re-executed any time. Use this for living reviews — schedule a re-run weekly/monthly and EvidAI surfaces only the new records since the last run.

Importing existing search results

If you ran the search elsewhere (or you are migrating an in-flight review), you can import results instead of running a federated search.

  1. From the Search Studio header, click Import.
  2. Pick a file: RIS, BibTeX, EndNote XML, CSV, PubMed XML.
  3. The importer parses, deduplicates, and adds the records to the review with the same provenance fields a federated run would produce.

You can also mix-and-match: import a PubMed RIS export, then run a complementary search on Embase from inside Search Studio. Deduplication runs across the union.

Quality assurance

Peer-review your strategy

Before screening, hand the search to a methodologist:

  • Use Export → Search strategy (Markdown) to generate a human-readable record of every concept, synonym, filter, and database with its translated query.
  • Have your methodologist review and email back comments.
  • If you change the strategy after peer review, save a new version (don't overwrite); the version history will appear in the audit trail.

PRISMA reporting

Search Studio writes a structured search-strategy artifact that the Report Prep phase consumes automatically. Your PRISMA flow diagram, search-methods table, and database-coverage report are auto-generated from the saved searches — no manual transcription needed.

Living reviews

If you switch the review to living mode, every saved search becomes a scheduled re-run. New records since the last run surface in the dashboard with a new badge so you can incrementally screen them without re-doing the whole run.

Common pitfalls

  • Too few synonyms — under-specified concepts lose relevant studies. Aim for at least 3–6 synonyms per concept.
  • Too many concepts — more than 4 concept blocks rarely improves precision and often kills sensitivity.
  • Forgetting trial registries — for clinical SLRs, always include ClinicalTrials.gov and WHO ICTRP to capture unpublished evidence.
  • Date filters too tight — limit by publication date only when you have a real methodological reason (e.g. intervention not licensed before 2010).
  • Skipping the preview — always open the per-database Preview and confirm at least the translated PubMed and Embase queries look like what you intended.

Next phase

When you are happy with the results, click Send to screening to move into the Screening phase — see the AI-Powered Screening article in this category.

Did this article help?
Still stuck?