I am preparing a systematic literature review on [REVIEW QUESTION] for [INTENDED USE]. Help me build and carry out a transparent review workflow, but do not claim to have searched databases, read full texts, or found studies unless I provide the records. My eligibility criteria are: [INCLUSION AND EXCLUSION CRITERIA]. The databases, date range, languages, and practical limits are: [SEARCH SCOPE]. I can provide search results, abstracts, full-text notes, and citation data as [AVAILABLE MATERIALS]. Produce the following: 1. Restate the question in a reviewable PICO, PICo, or other suitable framework and identify the primary and secondary outcomes. 2. Write database-ready Boolean search strings for each database I name, with controlled vocabulary placeholders where appropriate, plus a concept table of synonyms, spelling variants, and exclusions. 3. Give a PRISMA-aligned protocol outline covering sources, deduplication, title/abstract screening, full-text exclusion reasons, data extraction, risk-of-bias assessment, and planned synthesis. 4. Create a screening form with operational definitions that two reviewers could apply consistently. 5. Create a data-extraction table with fields for study design, population, intervention/exposure, comparator, outcomes, follow-up, effect estimates, confounding control, funding, and limitations. 6. When I paste records or notes, classify each as include, exclude, or unclear; cite the exact supplied text supporting the decision. Build an evidence table and a cautious narrative synthesis organized by outcome and study design. Do not turn association into causation. Ask up to 3 clarifying questions only if a required input is missing. Before answering, check that every proposed criterion maps to my question and flag claims that require a database search, full text, duplicate review, or subject-methodology expert judgment.
Fill in
| Placeholder | What to enter | Example |
|---|---|---|
| [REVIEW QUESTION] | Enter the focused question the review must answer. | Among adults with type 2 diabetes, do app-based self-management programs improve HbA1c compared with usual care? |
| [INTENDED USE] | Enter the course, thesis, report, or decision the review will support. | Master's dissertation background and methods chapter |
| [INCLUSION AND EXCLUSION CRITERIA] | Enter eligible populations, designs, settings, outcomes, and exclusions. | Randomized and quasi-experimental adult studies; English; report HbA1c; exclude gestational diabetes and protocol-only papers. |
| [SEARCH SCOPE] | Enter databases, date limits, languages, and any access or geography limits. | MEDLINE, Embase, CINAHL, and Cochrane Library; January 2015 to December 2025; English. |
| [AVAILABLE MATERIALS] | Enter the records, abstracts, notes, or citation exports you can provide. | I have 312 deduplicated titles and abstracts in a spreadsheet and will paste full-text notes for shortlisted papers. |
How to use
- Paste the prompt and replace every bracketed field before asking for search strings.
- Run the proposed strings in the named databases yourself and save the exact search date and result count.
- Paste records in manageable batches and require a quoted reason for every screening decision.
- Follow up with: “Using only the included-study table, draft a results synthesis that separates high- and low-risk-of-bias evidence.”
Variations
Protocol outline
Use this before searching, when you need a registered or supervisor-ready protocol.
I need a systematic-review protocol for [REVIEW QUESTION]. My planned population, intervention/exposure, comparator, outcomes, and study designs are [ELIGIBILITY DETAILS]. Draft a PRISMA-P-style protocol outline with objectives, sources, search dates, duplicate screening, extraction fields, risk-of-bias tool, and synthesis plan. Mark every decision I must make before registration; do not invent a database search or expected findings.
Abstract screening
Use this after exporting search results and before retrieving full texts.
Screen the following [RECORDS] against these eligibility criteria: [CRITERIA]. Return a table with citation, include/exclude/unclear, one precise reason, and the exact abstract wording supporting the decision. Treat missing information as unclear, not excluded. Do not infer study design, population, or outcomes beyond the supplied record. Flag records needing full-text review.
Evidence synthesis
Use this after extraction, when you need a disciplined narrative rather than a pooled estimate.
Using only this extracted-study table [EXTRACTED DATA], write a [WORD COUNT]-word narrative synthesis for [REVIEW QUESTION]. Group findings by outcome and study design, report direction and uncertainty of effects, identify important heterogeneity, and distinguish adjusted from unadjusted findings. Do not calculate a meta-analysis or causal conclusion unless the supplied data supports it. End with evidence gaps and limitations.
Tips
- Freeze eligibility criteria before screening; changing them after seeing results should be documented as an amendment.
- Search syntax is database-specific, so translate field tags and controlled vocabulary rather than pasting one Boolean string everywhere.
- Keep an exclusion-reason column for full texts because PRISMA flow reporting requires mutually usable reasons.
- AI can organize supplied studies, but search completeness, risk-of-bias judgments, and final inclusion decisions need accountable human review.
FAQ
Can AI perform a systematic review by itself?
No. It can help design searches, screen supplied records, and organize extraction, but it cannot reliably verify database coverage or replace independent reviewers.
Can I use AI-written search strings in my methods section?
You can use them after testing and documenting the final strings, platforms, dates, and result counts. A librarian or information specialist is valuable for complex reviews.
Should I paste full articles into ChatGPT?
Use only material you are permitted to share, and keep source citations with each extracted claim. Check every extraction against the original paper.