Abstract Screening with AI

What Abstract Screening Does

Abstract screening is where Scholara reads every title and abstract returned by your search and decides — based on your confirmed PICO and protocol — whether each study is potentially relevant. This is the most time-intensive phase of a traditional systematic review (typically 80–95% of citations are excluded here); the AI does the bulk reading so you can focus on borderline cases and quality control.

How It Starts

Once your search completes, the assistant transitions into the Abstract phase automatically and starts screening. You don't navigate to a separate page — a screening artifact opens in the side drawer and the live results stream in as the AI works through the citation list.

If you'd rather not start screening immediately, you can pause the run and resume it later from the same drawer.

The Screening Drawer

The screening artifact has two tabs across the top — Abstract screening (active during this phase) and Full-text screening (locked until abstract screening completes). Below the tabs you'll see a progress strip while screening is running, with a pause/resume button, animated dots, a progress bar, and the percentage complete. The strip disappears once the run finishes.

The main panel is a sortable, filterable table of every screened article showing the title, the AI's decision (Include, Exclude, or Uncertain), and the reasoning behind it. Click any row to open the full citation.

How Decisions Are Made

For each abstract, the AI:

  • Reads the title and abstract
  • Compares the study against your confirmed PICO and protocol fields
  • Checks each inclusion and exclusion criterion individually
  • Evaluates whether the study design matches what you specified
  • Returns a decision — Include, Exclude, or Uncertain — with a written rationale that names the specific criteria it relied on

Uncertain decisions are reserved for genuinely ambiguous cases — usually where the abstract doesn't contain enough information to make a confident call. These should be your first priority when reviewing.

Reviewing and Overriding Decisions

There are two ways to override an AI decision:

  • From the screening drawer — click the decision pill (Include / Exclude / Uncertain) in the table to flip it. The change is sent to the assistant on your next chat message, so the AI is aware of your override when it answers questions or runs full-text screening.
  • From the chat — ask the assistant directly. For example, "include the Chen 2019 trial — the abstract doesn't say so but the registered protocol confirms it's an RCT." The assistant will apply the override and update the table.

Use the chat for any override that needs explanation. Ask the assistant about borderline cases ("why was the Smith 2021 study marked uncertain?") and it will summarise the reasoning and the missing information. If you want to apply a rule across many studies ("exclude all studies with fewer than 50 participants"), just say so — the assistant will update the affected decisions in one pass.

Filtering and Sorting

Use the toolbar above the table to narrow the view:

  • Filter pills — All, Include, or Exclude. Tap a pill to show only studies the AI marked one way; tap All to clear the filter. (The Flagged pill applies to full-text screening, where it filters to studies whose PDFs couldn't be retrieved.)
  • Sort menu — Title A–Z or Z–A, by Decision, or by screening timestamp (newest or oldest first).
  • Export — once screening is complete you can export the full table as CSV or Excel from the toolbar.

Best Practices

Be Liberal at the Abstract Stage

When in doubt, include. Abstracts often lack the detail needed to make a confident exclusion, and excluding here is irreversible. You'll get full text in the next phase to make a more informed call. Including marginal studies costs reading time at full-text but doesn't compromise validity.

Always Review Uncertain Decisions

Uncertain is the AI saying "I can't tell from this abstract." These need a human call. Ask the assistant to walk through them ("summarise the uncertain decisions") and work through them in batches.

Spot-Check Inclusions and Exclusions

Skim a sample of decisions in both directions. Inclusions matter most because excluded studies are dropped — but a systematic exclusion error (e.g. the AI consistently excluding a study type you actually want) is easier to catch with a small spot-check of exclusions.

Track Exclusion Reasons

The AI's reasoning is recorded automatically and is used to populate your PRISMA flow diagram later. If you override a decision and want to record a different reason, mention it in the override message — for example, "exclude Smith 2020 — duplicate of an already-included study."

What Comes Next

When abstract screening finishes, the Full-text screening tab unlocks and the assistant transitions into the Fulltext phase. Included and uncertain studies move forward; excluded studies stay in the table for reference. From there you'll be able to upload PDFs (or have Scholara fetch open-access versions) and run full-text screening against the same criteria.

AI screening can make incorrect decisions, including on apparently clear cases. Review every Uncertain decision and spot-check both inclusions and exclusions against your protocol; human reviewers remain responsible for the final selection.