Manual assessment and figure export are free. “Generate AI draft” and “Dual-reviewer RoB” are paid. On the free plan, you judge every domain yourself, which good practice requires in any case. Like every AI feature, the draft requires a verified email.
Risk of bias is the reviewer's judgement. What the AI proposes is an unconfirmed draft, and every card says so. Your decision prevails.
The tool
At the top you see the count: how many studies are in, how many have full text and how many you have assessed. Each card has its tool, and revisia suggests it from the study design: RoB 2 for randomised trials, ROBINS-I for non-randomised studies of interventions, Newcastle-Ottawa for cohorts, its adaptation for cross-sectional studies and QUADAS-2 for diagnostic accuracy.
If it does not fit, select the tool and choose another. There are ten: besides those five, ROBINS-E, JBI, COSMIN, and ROBIS and AMSTAR-2 for reviews of reviews. Each has its own domains and judgements.
Assess each study
- Ask for a draft if it helps. “Generate AI draft” reads each study's full text (the PDF you uploaded or the PubMed Central text) and proposes a judgement for every domain: low, some concerns or high. Each judgement comes with the sentence from the article that supports it. A study without full text gets no draft: you assess that one by hand.
- Check the quotes. Each domain appears as a coloured row, with its judgement and quote. “See in the PDF” opens the original article with the sentence highlighted in yellow, and the arrows move between quotes. If a quote is not in the text, you will see the warning “quote not found in the text”.
- Confirm or correct. Select “Change assessment”. Each domain shows its guiding question and quote, and you choose the judgement. You can also assess entirely by hand, with no draft.
- Select “Save assessment”. The card changes to “Assessed by hand” and revisia calculates the overall judgement.
Make an independent assessment
Before assessing anything by hand, turn on “Dual-reviewer RoB”. You then assess first, blind, without seeing the draft. When you save, revisia shows your judgement and the AI's side by side, flags conflicts and measures agreement with Cohen's kappa.
Figure and table
- The traffic-light figure crosses studies and domains, with a symbol inside each dot so it also reads in black and white. “Expand” opens it full size, and “Download” gives PNG, SVG, PDF or TIFF at 600 dots per inch, the format many journals ask for.
- The table says the same, grouped by tool, with a dot per domain and the overall judgement at the end.
When you finish, select “Approve & continue”. If any study has not been assessed, revisia warns you first: it would appear without risk of bias in the figure and the manuscript.
Troubleshooting
- A study gets no draft. The draft needs full text. Upload it in Full text or assess the study manually.
- It will not generate drafts. At least one included study must have full text.
- I disagree with the draft. Correct the domain. This is a normal part of the review, and your decision is marked as assessed by hand.
- I already assessed by hand and cannot turn on the dual reviewer. It must be enabled before any assessment, so your judgement is truly blind.