This is a workflow for a bounded collection you provide. It does not establish that you searched every relevant database or completed a systematic review. Your assignment, review method, and institutional AI rules determine what work is required.
OpenAI's research-evidence workflow calls for resolving duplicates, missing files, and inaccessible references before synthesis. That first inventory prevents a fluent paragraph from hiding an incomplete or double-counted evidence base.
State the selection rule in words you can apply
For this original fictional exercise, the review question concerns help instructions for public-library item reservations. Include material that examines users completing that task and reports what happened during the attempt. Exclude studies of staff cataloging tools. An abstract that mentions “library services” without identifying the task is insufficient for a final decision.
The records below are complete invented screening notes. They are not real publications, research findings, or observed ChatGPT output.
| File ID | Available information | Proposed treatment |
|---|---|---|
| F1 | Full text: Study S1 examines public users attempting item reservations and reports observed attempts. | Candidate for inclusion; verify the task and results in the full text. |
| F2 | Earlier preprint of S1; same participants and study dates as F1. | Link to S1 as another report; do not count a second independent study. Compare versions before choosing the working report. |
| F3 | Full text examines staff entering catalog records. | Exclude under the stated task criterion; retain the reason. |
| F4 | Abstract describes “library-service usability”; reservation task unspecified. | Await full text; no final exclusion or inclusion. |
| F5 | Full text: Study S2 examines public users attempting item reservations and reports observed attempts. | Candidate for inclusion; inspect the relevant sections. |
There are five files, two candidate independent studies, one excluded study, and one unresolved candidate. That is the state of this exercise's collection. It is not a final evidence count: F4 could change it, and F1 and F5 still need the indicated reading checks.
Keep the reason and the evidence level together
An exclusion needs the criterion it fails and the passage supporting that decision. “Not relevant” is hard to audit. “Staff cataloging task; review covers public item reservations” explains why F3 falls outside this question.
Do not turn “not stated in the abstract” into “not studied.” For F4, record which text you actually have and what remains unknown. Obtain the full paper through your permitted library access. If it remains unavailable, keep its status distinct from an evidence-based exclusion and report how your review handles unavailable material.
For F1 and F2, keep both file identities attached to S1. A preprint and journal article can differ in corrections, methods detail, or analysis. Check the relationship and version differences rather than discarding one solely because its title looks similar.
Copy a screening prompt before asking for prose
Help screen this supplied collection for my literature review.
Review question and assignment/method requirements: [PASTE]
Explicit inclusion/exclusion criteria: [PASTE]
File inventory with stable IDs, references, versions and available
text level (title / abstract / full text): [PASTE]
Relevant source passages with locations: [PASTE]
Return proposed include, exclude or awaiting-information decisions.
For each, cite the exact criterion, source passage and location,
and what text you could actually inspect. Do not infer absence from
an abstract's silence. Keep missing full texts unresolved.
Identify possible multiple reports of one study; preserve file IDs
and explain the evidence for linking them. Do not invent references,
findings, publication status, independence or search completeness.
Do not write synthesis prose until I review the selection log.
Read each proposed decision against the original paper. Resolve disagreements about the criteria before changing a group of decisions. If you broaden the question, revisit excluded records affected by that change and retain the earlier decision history.
Write from the checked collection
After screening, extract the included studies' methods, relevant results, and limitations with source locations. The literature-review matrix then helps compare findings across shared themes. A screening log answers which material belongs; a matrix answers how the included material relates.
Check every generated paragraph against the sources, including claims about the number of studies and where evidence is missing. The citation-checking guide covers checking the supporting passage behind a sentence.
For this fixture, leave F4 open until its text is available, and keep F1 and F2 linked as reports of S1. Carry the selection log into your normal research records before writing what the collection establishes.