This is the first stage that cannot start until you verify your address. If screening asks you to do so, use the button in the welcome email or request another email from the app. Your existing work stays in place.
Run screening
- Select “Launch full screening”. A window shows how many records it covers. Screening runs on our servers: you can close the tab and come back later.
- Mind your plan. On the free plan, the AI screens up to 100 records per review; the rest are marked for you to review.
- Look at the result. When it finishes, the card shows how many articles were screened and how many remain in “Review”. Project status, on the right, now has figures: identified records, matches across databases, screened, included, excluded, manual overrides and AI records used from your plan.
How decisions are calculated
For each study, the AI answers criterion by criterion: “yes”, “no” or “uncertain”, and copies the sentence it relies on. The AI does not make the decision: code calculates it from those answers.
- A “no” for a required criterion, backed by a sentence that is in the text: the study is excluded. If the sentence is not there, or the AI says “no” because the abstract lacks the information, it goes to “Review”.
- Every criterion is “yes”, each with its sentence in the text: the study is included.
- There is uncertainty and no “no”: it goes to Review. Nothing is excluded on its score alone.
- A study retracted by the journal is excluded automatically, because a withdrawn study is not valid evidence.
Each study receives s = (number of “yes” + 0.5 · number of “uncertain”) / number of required abstract criteria. A score of 1.0 means everything is confirmed; 0.5 represents complete uncertainty. The same answers always give the same decision. It is not a probability.
Adjust the “Review” band
Open “Decision threshold”. It is the real score distribution: green for what enters, amber for what goes to human review and red for what is discarded. Dark red is what a required criterion excluded, and that does not move.

- Look at the amber strip. It is the “Review” band: studies where the AI is not sure. By default only studies that meet every required criterion enter. You decide the uncertain ones.
- Move the limit. It runs from 0.5 to 1.0: at or above it a study is included; below it, “Review”. Nothing is excluded for uncertainty: excluding needs a “no” with its sentence in the text. The chart recolours instantly and tells you how many articles would remain for manual review.
- Select “Apply band & recompute”. Decisions are recalculated without calling the AI again, and the change stays in the history.
A “no” with proof for a required criterion always excludes. The band only decides when a study is included; everything else goes to “Review”.
Resolve the “Review” queue
Below the chart is the list of decisions, with its filters. For the uncertain ones, select “Review the N pending”. The queue shows one study at a time: title, abstract and, below, the verdict for each criterion with its quote. The quote is highlighted in the abstract with its criterion number, so you can check where it comes from. Studies closest to inclusion come first.
- Include or Exclude: to exclude, choose the criterion it fails. That reason goes to the audit log and to the PRISMA diagram.
- Next: leaves the case undecided for later.
- Shortcuts: I to include, E to exclude and S to move on. If you make a mistake, use “Undo”.
When you finish, select “Approve & continue”. If studies remain in “Review”, revisia warns you before closing the stage.
Double screening and overrides
Double screening and the invited reviewer are on paid plans.
- Double screening against the AI. Before launching screening, turn on the “Individual blind review of AI results” row. There are three steps: the AI screens first; then you screen blind, without seeing its decisions; finally you resolve disagreements with a written reason. Disagreements stay in the “Conflicts” filter.
- Invited reviewer. Under “Invited reviewer” you invite someone as a second reviewer: they get an email with the link, which you can also copy and share yourself. They screen blind with their own account and see only the criteria and abstracts. You can invite them at any time, as soon as there are papers in the screening stage, even before you start or halfway through. Each of you screens the same papers at your own pace, without waiting for the other, and papers added later reach you both. Decisions are compared once you have both finished: they submit their review and you have included or excluded every paper.
- Agreement. At the end you see the percentage agreement and Cohen's kappa, which measures agreement beyond chance. Kappa appears from 20 compared articles.
- Override a decision. Change a study's outcome with “Change decision”. Who did it, when, the previous decision and the reason are all recorded.
- Exclusions table. Excluded studies are grouped by criterion in the next stage. View the exclusions table.
Troubleshooting
- Too many studies are in “Review”. Check the criteria: a vague one leaves many “uncertain”. The band does not exclude for uncertainty; excluding needs a “no” with its sentence in the text. The queue shows the clearest studies first.
- A study changes between “Include” and “Review” on repeat runs. The model is not perfectly deterministic. Any variation ends in human review, never silent exclusion.
- Screening ended with an error. Run it again. Processed records do not consume allowance twice.
- The AI quota ran out. Records the AI does not screen are marked for human verification. Manual screening and the “Review” queue remain available.
- I changed a criterion and see a warning band. Some verdicts come from the previous criteria. Read changing criteria after screening.
Complete screening record
Under Export, choose Complete screening record (Excel), then select Español or English. This download is also available in the exclusions table.
Its eight sheets cover searches, identified records, decisions, full-text assessment, history, criteria and PRISMA counts. Titles, queries and notes keep their original language. Missing historical information is marked Not recorded.
Downloads become available once AI screening has finished and no articles remain unscreened. Decisions may still need manual review. During independent review, complete submission and resolve disagreements before exporting. You can download while manual review continues. The workbook captures a snapshot of saved data with its date and time. Later results will appear in the next download.