PRISMA flow diagram with AI-assisted screening

A PRISMA flow diagram shows how records move through identification, screening, eligibility, and inclusion. When AI-assisted screening is used, the diagram still needs the same basic logic: how many records were found, removed, screened, excluded, assessed, and included.

AI does not remove the need for a clear flow. It adds questions about prioritization, stopping rules, automation, and human review.

This guide explains how to prepare a PRISMA-style flow record when AI helps with literature review screening.

A PRISMA flow diagram is a visual record of how sources move through a review. It helps readers understand where records came from and why records left the review.

It usually tracks:

• Records identified.
• Duplicate records removed.
• Records screened.
• Records excluded.
• Full-text reports assessed.
• Reports excluded with reasons.
• Studies included in the review.

The diagram is not just a formality. It is a transparency tool.

For review-method context, see AI in systematic reviews.

AI-assisted screening can change the order of records, suggest relevance, group records, or help prioritize what reviewers inspect first. Those actions may not fit neatly into traditional manual-screening notes unless the team records them.

The team should know:

• Did AI rank records?
• Did AI suggest exclusion?
• Did humans make final decisions?
• Were all records screened?
• If not, what stopping rule was used?
• Was an audit sample checked?
• Were AI-assisted steps reported?

The flow diagram counts records, but the methods text explains how AI affected movement through the flow.

Collect counts from the beginning so the final diagram does not require reconstruction from memory.

Track:

• Database records identified.
• Records from citation chasing.
• Records from AI-assisted discovery.
• Records from grey literature sources.
• Duplicate records removed.
• Records imported into screening.
• Records screened by title and abstract.
• Records excluded at title and abstract.
• Full texts sought.
• Full texts not retrieved.
• Full texts assessed.
• Full texts excluded with reasons.
• Studies included.

If AI search helped identify candidate records, label that source route clearly.

For search logging, see literature search log template for AI-assisted reviews.

AI-discovered records should be shown as a source route, not hidden inside another count. If an AI tool helped find papers, record which papers came from that step.

Possible labels include:

• Records identified through AI-assisted academic search.
• Records identified through citation network search.
• Records identified through seed-paper expansion.
• Records identified through manual search.
• Records identified through database search.

The exact label depends on the workflow. The goal is to show how each record entered the review.

For discovery methods, see citation network search vs keyword search.

Duplicates should be removed before screening counts are finalized. AI-assisted workflows can create duplicate risk because the same paper may appear through databases, citation maps, AI search, and manual discovery.

Track:

• Total records before deduplication.
• Duplicate records removed.
• Deduplication method.
• Records remaining after deduplication.
• Any duplicate decisions that required manual review.

Deduplication is not glamorous, but it protects the screening count. If the same paper appears multiple times, the flow diagram should not count it as multiple screened studies.

If AI ranked records, record that ranking affected screening order. Then make clear who made the inclusion and exclusion decisions.

Record:

• Tool or method used for prioritization.
• Whether all records were screened.
• Whether reviewers saw AI labels or scores.
• Whether AI decisions were advisory.
• Whether conflicts were resolved by humans.
• Any audit sample used.

Do not imply that all records were screened manually in a neutral order if AI prioritization changed the workflow.

For active learning details, see what is active learning screening in systematic reviews.

The flow diagram may show how many records were screened, but the stopping rule usually belongs in the methods text. The two should match.

Report:

• The stopping rule.
• When it was applied.
• How many records remained unscreened, if any.
• Whether an audit sample was screened.
• What happened if relevant records were found late.
• Who approved the stopping decision.

If every record was screened, say that. If not, explain why the unscreened records were left out and how risk was checked.

For stopping decisions, see when to stop screening in an AI-assisted review.

Full-text exclusion reasons should be specific enough to understand why a source left the review. AI can help draft labels, but the team should decide the final reason.

Common reasons include:

• Wrong population.
• Wrong intervention or exposure.
• Wrong outcome.
• Wrong study design.
• Not peer reviewed, if required.
• Full text unavailable.
• Duplicate publication.
• Retracted source.
• Not enough relevant data.

Use consistent categories. If categories change halfway through, update earlier records or explain the change.

The methods text should describe AI use in plain, inspectable terms. The flow diagram alone cannot carry all AI-related information.

Include:

• What AI-assisted tool or process was used.
• Which review stage it supported.
• Whether humans made final decisions.
• Whether all records were screened.
• How stopping was handled.
• How source verification was performed.
• Known limitations.

Avoid vague statements such as "AI was used to screen literature." That does not tell readers enough.

For policy wording, see AI literature review policy template for research teams.

Turn the advice into a repeatable workflow by defining the decision you need to make, the evidence required for that decision, and the record that will prove how the decision was made. In screening, the problem is rarely one missing tool. The problem is usually that search, reading, checking, and writing happen in separate places without a shared rule.

Use a short operating routine:

• Name the review question or subquestion.
• Define the source set you are working from.
• Decide what counts as enough evidence for the next step.
• Apply the same criteria to every paper in that step.
• Mark uncertain cases instead of forcing a clean answer.
• Keep source locations for claims that may enter the final review.
• Review the workflow after each major search, screening, or writing session.

This routine keeps the work moving without making the review careless. It also gives supervisors, collaborators, and future you a way to understand why the source set changed.

Record the pieces that would be hard to reconstruct later. You do not need a diary of every click, but you do need enough detail to explain the path from question to source to claim.

For this topic, the most useful record usually includes criteria, reviewer decisions, exclusion reasons, audit checks, and source flow. Add the date, tool or source used, reviewer status, and next action. If AI assisted the step, write down what it helped with and what a human checked.

The record should distinguish discovery from evidence. A tool may help find a paper, but the paper itself must support the claim. A summary may help triage a source, but the original source should support any statement that appears in the literature review.

Before writing, check whether the workflow has produced usable evidence or only useful notes. Notes help you think. Evidence supports a sentence.

Ask:

• Which claim will this source support?
• Is the claim narrower than the evidence?
• Have methods, sample, outcome, or concept details been checked?
• Are limitations visible?
• Are conflicting papers handled rather than ignored?
• Is the citation real, current, and relevant?
• Can another reader understand how this source entered the review?

If the answer is unclear, keep the point in notes rather than moving it into the draft. This is the small pause that prevents AI-assisted research from becoming polished but weak writing.

WisPaper can help researchers discover, triage, and save papers before final review reporting. Deep Search, Scholar Agent, and Inspiration Discovery support natural-language academic search, while paper cards show source labels, summaries, authors, publication details, and preview images.

This can help researchers identify candidate papers and decide what belongs in a screening workflow. Papers can be saved or uploaded into My Library, and Library QA can answer questions based on that known paper set.

WisPaper does not replace PRISMA reporting or make final inclusion decisions. It can support the paper-discovery and source-organization work that feeds into a documented review process.

Try WisPaper