Every run keeps the exact query sent to each database, the filters, the date and the result count, by version. Every imported file keeps its source database, record count and, if you add it, the query you ran there. These records make the review reproducible and feed the search-strategy appendix that PRISMA and PRISMA-S ask for.
The two paths
When you enter this stage, revisia asks one question: have you already completed your search? It then shows you two cards.
- Path 1 · Build it with me. Select “Create my search”. You start from your question and review revisia's proposal. revisia builds the strategy and searches PubMed, Europe PMC, OpenAlex, Semantic Scholar and ClinicalTrials.gov. You do not need to know what a MeSH heading is.
- Path 2 · Import your search. Select “Import my search”. Export results from PubMed, Web of Science, Scopus, Embase, Cochrane or another database and import them. revisia removes duplicates and prepares the records for screening. This path suits a strategy you already have, often one reviewed by an information specialist.
Both paths produce the same thing: a deduplicated corpus and the Criteria stage. You can switch paths until you approve this stage.
Path 1 · Build it with me
The starting query
On this path revisia reads your question, extracts its concepts and writes the query. It usually takes under a minute. When it finishes, the “Your search” box shows the query in PubMed syntax, with each concept in its colour. You can edit it or paste your own (MeSH, [tiab], AND, OR). Nothing runs automatically: the search only starts when you launch it.
Further down, “Searching for” shows which concepts enter the query.
Population and intervention enter the query. Comparison and outcome start as “out” on purpose: many studies do not name them in the title or abstract, and requiring them would lose those studies. If the result set is too broad, each has an “Also search by this concept” button.
Search terms
Open “Search terms”: one column per concept. Ticked names are searched; under “Suggestions” other names wait unticked. The MeSH label marks the official headings PubMed uses to classify articles; everything else is searched in title and abstract. Tick or untick, and the query above rewrites itself.
- The proposal arrives reviewed. revisia switches off names that bring nothing in PubMed. If almost every name of a concept shares a word, it also searches that word alone, so you do not lose articles that combine it differently. It only does so when PubMed confirms the search does not explode.
- Missing a name? Type it in its column and press the plus. Or select “Suggest with AI” for new, unticked synonyms you can choose from.
- Want to rename a concept or add one? “Edit in detail” opens the step-by-step editor.
This controls breadth. More ticked names broaden the search; fewer narrow it. Adding another concept with AND narrows it sharply, because every result must mention it.
Article estimate and “By term”
- Choose databases. “Article estimate” has a card per database, each with its “Select” button. PubMed and Europe PMC are on by default. You can add OpenAlex, Semantic Scholar and ClinicalTrials.gov, the trial registry guidelines ask for so you do not miss unpublished trials. All five are on the free plan; there, OpenAlex allows three searches per review.
- Select “Run estimate”. Each card says how many articles that database would return with your filters, without importing anything. Its question mark explains what the number measures. With PubMed selected, Europe PMC and OpenAlex only import what is not already in PubMed, and the question mark gives both figures. “See 5 examples” shows a sample.
- Look at “By term”. Under the box, each name has two numbers in the database you choose. “To import” counts only the new articles the names above do not bring, so the rows add up to your search total in that database. With PubMed selected, fewer will arrive from Europe PMC and OpenAlex, because they only import what PubMed lacks. “In PubMed” is what that name brings on its own. A warning triangle marks a name with over 50,000 articles and 10 times its concept's median: it is usually too broad. A zero under “To import” means it adds nothing to the names above.
- Adjust with numbers. Names that add nothing to your search in any selected database are removed automatically, with a notice and “Undo”. Remove any other with its bin. Selecting the name shows how many articles arrive only through it: what you would lose by removing it.
![“By term” in PubMed: how much each population and intervention name brings, with a warning triangle on “diabetes[tiab]”](/guia/por-termino-en.webp)
Path 1 · Check and run
Filters and syntax for each database
- Review the filters. The active filters sit next to the box as chips. By default there is one, “English”, which you remove with its cross. The funnel icon opens “Filters”: language, “From (year)”, “To (year)”, “Max results” per database and study type. Every filter can drop good studies: set only those your protocol justifies.
- Read the study-type note. Study type does not filter the same way in every database. As soon as you pick one, a note below the filters says how each selected database applies it. In PubMed, “Randomized controlled trial (RCT)” uses Cochrane's broad filter, which also brings reviews that discuss randomisation. Europe PMC only lets through articles tagged with that type, and OpenAlex approximates it with words in title and abstract. Filtering by type misses articles the database has not tagged yet.
- Look at “Syntax by source”. The code icon opens one tab per selected database, with the exact query it receives. A fixed program does the translation, not the AI, so it always comes out the same. A note says what was approximated: for example, OpenAlex has no MeSH and searches those concepts in title and abstract.
- Act on syntax notices. As you type, revisia checks the query. An unclosed parenthesis or a wildcard PubMed ignores is underlined, and “Review” proposes the fix. If a MeSH heading does not exist in the catalogue, revisia warns you too: it retrieves nothing and, inside an AND block, reduces the search to zero.
Check articles in your search
This is the most reassuring test: seeing whether your query finds the studies you know must appear.
- Find the study. Paste its DOI or PMID, or type the title or an author, and select “Search”. The search tool's funnel chooses what you search by (“Title”, “Title or abstract”, “Author” or “DOI / PMID / NCT”) and in which databases.
- Read the pills. Each result has one pill per selected database: green where your query brings it, struck through where it does not, grey where it cannot be checked. ClinicalTrials.gov has no authors, DOI or PMID: it searches by title, abstract or NCT.
- Mark it as a favourite. With the star, revisia rechecks the study every time you change the query and, if a change drops it, tells you which term excludes it. The star fills at once and a notice confirms it. You can have up to eight.
- See them under “Your favourites”. At the top of the card, “Your favourites” lists every study you have marked, each with its verdict, and the ✕ removes it. Clicking its star again also removes it. A ClinicalTrials.gov record is saved too, but without a DOI or PMID we cannot check it and it shows as “Not identified”.

Passing this test is a minimum, not a ceiling: it says nothing about the studies you do not know yet.
Run the search
- Select “Run search”. Before anything is spent, a window shows how many results each database brings, duplicates still included. Confirm with “Run this search in N databases”.
- Follow the progress. The search runs on the server and you see finished queries and retrieved records live. You can close the window: the search continues.
- Read the summary. Per database: how many it found, how many it imported and how many duplicates it removed. A study found in two databases is kept once.
Every run is saved under “Saved strategies”, by version: the exact query for each database, filters, date and result count. From there you download your search appendix exactly as it ran. If you also use Web of Science, Embase or Cochrane, the optional next stage, Add more studies, gives you the query already translated and accepts those exports.
Path 2 · Import your search
This path has no AI proposal and no query to review. Stage 2 becomes the import screen, and the Add more studies stage disappears from the path because it is no longer needed.
- Export from each database. Open “How to export from each database”: PubMed, “Send to → Citation manager (.nbib)”; Web of Science, “Export → Plain text file → Full record”; Scopus, RIS or CSV with abstract; Embase and Cochrane, RIS; Zotero and EndNote, the collection as RIS.
- Drop all files together. Several at once, from several databases: RIS, BibTeX, CSV, PubMed .nbib, Web of Science .txt and WHO ICTRP XML, up to 5 MB each. revisia reads each file, counts its records and recognises its database: recognised files are marked “Detected”, and several files from the same database count as one search.
- Choose the database when needed. A file that does not say where it comes from stays under “No database”, and the import button waits. Pick the database from the list or, for a list from your supervisor or another review's bibliography, “Other methods”. That database appears later in the PRISMA diagram, which is why revisia does not guess when it is unsure.
- Record your search in each database. Each database has its own “Your search” fold: exact query, platform, date, limits and result count. It is optional, but PRISMA-S asks for it and it goes into the appendix. If you enter the database's result count, revisia compares it with the files: “✓ Matches”, or “N missing” if a batch was not exported.
- Select “Import N files”. Files run one after another, each with its database. revisia compares each reference with what is already in the project, by identifier (DOI, PMID) and by title, and merges duplicates.
- Read the summary. How many records you received, how many were duplicates, how many stayed out because of your plan's cap and how many were added to the project. If a batch is missing or you update the search months later, the drop area is still there: only new records enter.
- Approve the stage. Confirm you have the right to use those documents and select “Approve & continue”. You go straight to criteria.
Cross-database duplicates are removed automatically and every merge enters the audit log. The plan's record cap is cumulative: records removed from a project still count. If you bring more than your plan allows, revisia warns you before importing, not after.
Switch paths
The top of the stage shows the selected path and either “Switch to importing my search” or “Switch to building it with me”. Switching deletes nothing. Completed searches and imported files stay in the corpus. Only the screen changes, and whether the Add more studies stage appears. You can switch until you approve this stage. Once screening has started you cannot, because changing the corpus under judgements already made makes no sense.
If you get stuck
- I cannot see the two cards. They only appear before the project has a path. Afterwards, the stage opens on your path; use the link at the top to switch.
- The estimate is zero or very low. It is almost always the language filter or the year range. Remove the “English” chip or try “All languages”. Check “By term” too: a name with zero adds nothing, and a MeSH heading that does not exist reduces the search to zero.
- I wrote the query in Spanish. revisia translates it for PubMed, and “Syntax by source” shows the exact text each database receives.
- There are too many results. Look in “By term” for names with the warning triangle and remove them, or add the outcome with “Also search by this concept”.
- A study I know does not appear. Mark it with the star under “Check articles in your search”: revisia tells you which term drops it.
- A Web of Science file fails. Export “Plain text file → Full record”. The old UTF-16 tab-delimited format is unsupported; export plain text or RIS.
- I am missing Latin American or Iberian literature. Search SciELO, LILACS or Dialnet on their sites and import RIS, BibTeX or CSV through path 2 or Add more studies.
- I reached the search limit. The maximum is 50 searches in 24 hours, and each database counts as one: running in four databases uses four. Saved searches are unaffected.