Why Your Systematic Review Keeps Stalling at the Screening Stage

Why does a systematic review stall at screening, week after week, when every abstract is being logged and tagged? The team is not idle. The stall sits somewhere else: inclusion criteria drift by small degrees, and two reviewers quietly disagree. WisPaper, an AI academic agent, treats screening as a thing to interrogate rather than a queue to push through.

Week Three and the Same Forty Abstracts

By week three you have read the same forty abstracts three times, because each pass applies slightly different rules. A paper excluded on Monday looks includable by Thursday once you notice the sample size or the comparator. This is not laziness. It is what happens when criteria live only in a protocol paragraph and in two reviewers’ heads.

When Inclusion Criteria Start to Drift

Drift starts small. Someone tightens the date range after a conversation, someone else loosens the outcome definition to save a study they like. Nothing gets written down, so nothing is testable. The screening tool shows rows and checkboxes but no record of why an include became an exclude. Two or three weeks in, you are no longer screening the review’s question; you are screening whatever the team currently believes.

The Borderline Paper Two Reviewers Disagree On

Then comes the moment two reviewers mark the same borderline paper differently and your inter-rater score dips. One says the intervention matches, the other says the population is too specific. You schedule a reconciliation meeting, which turns into re-litigating criteria you thought were settled. A tiebreaker opinion will not fix it. You need a way to look at the paper through the same definition of “included,” applied the same way every time.

Screening as a Conversation, Not a Spreadsheet

This is where screening changes from a spreadsheet exercise into a conversation with your own criteria. WisPaper lets you put a review question in front of an agent that searches the literature against stated include and exclude conditions and shows its reasoning, so borderline calls get made against the written rule instead of memory. You can keep the surviving papers in one place, and you can ask whether a citation actually matches the format you are required to submit.

From 4,000 Abstracts to a Defensible Shortlist

The answer to the stalling question is that screening breaks when criteria drift and nobody can point to why a decision was made. Fix the drift, and 4,000 abstracts become a shortlist you can defend line by line, including the disagreements. The AI academic agent approach gives the review a paper trail: search reasoning and verified citations traceable to the same question. The shortlist is not the finished review, but it is the part most teams never quite reach.