Research · Academic research
The Best Prompt for a Literature Review
The best prompt for a literature review keeps the model inside the material you give it. This one has the model inventory your sources, tabulate what each one found, group them into themes and only then write, citing nothing that is not in your paste. You get a traceable draft, a list of gaps in your source set, and a short list of claims to check against the original papers.
Source-Bound Literature Review Prompt
Turns the papers, abstracts and notes you paste in into an evidence table, themes and a synthesized draft, with every citation traceable to a source you supplied.
- Works with
- ChatGPT, Claude, Gemini
- Difficulty
- Advanced
- Complexity
- Multi-part reply
- Variables
- 6
- Version
- 1.0
Customize this prompt
Fill in what you know. Anything left blank stays as a [bracketed placeholder]. Nothing you type leaves your browser.
6 variables · fields marked * matter most
# Role You are a research assistant drafting a literature review. You work only from the material I supply. You do not cite, quote or describe any publication from memory. # Review brief - Research question: [Research question] - Field: [Field or discipline] - What the review is for: [What the review is for] - Citation style: [Citation style] - Target length of the written review: [Target length] # Source material <sources> [Your sources] </sources> # Method Work through these steps in order, showing the output of each. Step 1 - Source inventory. Give every source an ID (S1, S2, ...). Record author, year, title and publication exactly as they appear in the material, and note what I gave you for it: full text, abstract only, or my notes. If a detail is absent, write [MISSING] instead of completing it. Step 2 - Evidence table. One row per source: aim or question, design and sample, main finding, limitations, and how directly it bears on my research question (direct, partial, background). Step 3 - Themes. Group the sources into three to five themes defined by ideas, not by author or date. For each theme state where sources agree, where they conflict, and the likeliest reason for the conflict (different definitions, methods, populations or time periods). Step 4 - Draft. Write the review theme by theme in connected prose. Compare and weigh sources against each other; do not summarize them one after another. Give more weight to sources I supplied in full than to abstracts or notes, and say so where it matters. Step 5 - Gaps. List what this set of sources leaves unanswered about my question. Describe these as gaps in the supplied set, not gaps in the field. Suggest search terms for filling them, not specific titles. # Rules - Every citation must point to a source in the inventory. Never add a paper, author, year, journal, DOI or page number that is not in the supplied material. - Where a point needs support that no supplied source gives, write [NO SOURCE SUPPLIED]. - Use direct quotations only when the exact words appear in the material. Otherwise paraphrase. - Keep a source's findings separate from its authors' interpretation or speculation. - For abstract-only sources, do not describe methods or results beyond what the abstract states. - Do not strengthen hedged claims. "Associated with" stays "associated with". # Output format Five headed sections matching the steps, with tables for steps 1 and 2. In the draft, put the source ID in brackets after each formatted citation, for example (citation) [S3], so I can trace every claim. Finish with "Claims to verify": the five to ten statements in the draft I should check against the original papers first.
Use it when you are
- Drafting the background chapter of a thesis or dissertation from papers you have already read
- Writing the related-work section of a journal article or conference paper
- Preparing the evidence background for a grant application
- Turning a folder of reading notes into a structured first draft before a supervisor meeting
Why this prompt works
Citations can only come from the inventory
Language models can produce references that look right and do not exist. Here the model must build a numbered inventory before writing, and every citation in the draft carries a source ID that points back to it. A reference with no ID, or an ID that does not match, is easy to spot, which makes fabricated citations a visible error instead of a hidden one.
An evidence table comes before any prose
Reviewers usually extract aim, design, finding and limitations from each paper before they write. Asking for that table first gives the model a structured set of facts to draw on and gives you a quick way to check that it read each source correctly before you invest time in the draft.
Themes are defined by ideas, and conflicts need a reason
A weak review walks through papers one by one. Requiring three to five idea-based themes, with agreement, disagreement and a likely cause for each disagreement, pushes the draft toward synthesis: comparing definitions, methods and populations across sources instead of restating abstracts in sequence.
Abstracts and notes are treated as weaker evidence
You rarely have full text for everything. The prompt records what was supplied for each source and forbids describing methods or results beyond an abstract. That stops the model from filling in study details it could only be guessing at, and it tells you which sources deserve a full read.
Gaps are scoped to what you supplied
A model that has seen eight papers cannot know what the field has not studied. Labeling gaps as gaps in your set, and asking for search terms instead of titles, keeps the output honest and gives you a concrete next search to run in a database.
How to use it
- Collect your sources in one document. For each, paste the citation details as printed on the paper, then the full text, the abstract or your own notes.
- State the research question as a question, not a topic. The relevance column and the themes are built around it.
- Run the prompt and check the source inventory first. If an author, year or title is wrong there, fix your paste and rerun before reading further.
- Compare a few evidence-table rows against the papers themselves, especially sample sizes and main findings.
- Work through the "Claims to verify" list against the originals, then rewrite the draft in your own voice.
- Run the suggested search terms in a scholarly database to find sources for anything marked [NO SOURCE SUPPLIED].
AI model compatibility
| Model | Status | Notes |
|---|---|---|
| ChatGPT | Expected to work | OpenAI's guidance recommends delimiters to separate instructions from reference text. Keep the <sources> block intact, and if the review is cut short, ask it to continue from the last completed step. |
| Claude | Expected to work | Anthropic's guidance recommends wrapping supplied documents in XML-style tags. The prompt already uses <sources>; for many papers, wrap each one in its own numbered tag inside it. |
| Gemini | Expected to work | Keep the numbered steps in order. If you attach PDFs instead of pasting text, tell the model which file corresponds to which source so the inventory stays accurate. |
“Expected to work” means the prompt uses plain, model-independent structure and has been reviewed for that — not that it has been run on that model. A model is marked Tested only when a test record is published below.
Worked example
Scenario: A master's student drafting a thesis background section from nine papers they have read on retrieval practice.
# Role You are a research assistant drafting a literature review. You work only from the material I supply. You do not cite, quote or describe any publication from memory. # Review brief - Research question: Does retrieval practice improve long-term retention in undergraduate science courses compared with rereading? - Field: Educational psychology - What the review is for: Thesis or dissertation chapter - Citation style: APA - Target length of the written review: 1,200 words # Source material <sources> [Nine entries pasted here. Six with citation details and full text, three with citation details and abstract only, each followed by the student's own reading notes.] </sources> # Method Work through these steps in order, showing the output of each. Step 1 - Source inventory. Give every source an ID (S1, S2, ...). Record author, year, title and publication exactly as they appear in the material, and note what I gave you for it: full text, abstract only, or my notes. If a detail is absent, write [MISSING] instead of completing it. Step 2 - Evidence table. One row per source: aim or question, design and sample, main finding, limitations, and how directly it bears on my research question (direct, partial, background). Step 3 - Themes. Group the sources into three to five themes defined by ideas, not by author or date. For each theme state where sources agree, where they conflict, and the likeliest reason for the conflict (different definitions, methods, populations or time periods). Step 4 - Draft. Write the review theme by theme in connected prose. Compare and weigh sources against each other; do not summarize them one after another. Give more weight to sources I supplied in full than to abstracts or notes, and say so where it matters. Step 5 - Gaps. List what this set of sources leaves unanswered about my question. Describe these as gaps in the supplied set, not gaps in the field. Suggest search terms for filling them, not specific titles. # Rules - Every citation must point to a source in the inventory. Never add a paper, author, year, journal, DOI or page number that is not in the supplied material. - Where a point needs support that no supplied source gives, write [NO SOURCE SUPPLIED]. - Use direct quotations only when the exact words appear in the material. Otherwise paraphrase. - Keep a source's findings separate from its authors' interpretation or speculation. - For abstract-only sources, do not describe methods or results beyond what the abstract states. - Do not strengthen hedged claims. "Associated with" stays "associated with". # Output format Five headed sections matching the steps, with tables for steps 1 and 2. In the draft, put the source ID in brackets after each formatted citation, for example (citation) [S3], so I can trace every claim. Finish with "Claims to verify": the five to ten statements in the draft I should check against the original papers first.
What a good response should contain
- An inventory of nine sources, with the three abstract-only entries labeled as such and any absent details shown as [MISSING]
- An evidence table that separates classroom studies from laboratory studies and notes the retention interval each one measured
- Themes built around ideas such as feedback, test format and delay, each stating where the supplied sources disagree and why
- A draft in which every citation carries a source ID, with [NO SOURCE SUPPLIED] wherever a point lacks support
- A gaps section framed as limits of the nine sources, with database search terms instead of suggested paper titles
This is an illustrative scenario showing how to fill in the prompt and what to look for. It is not a recorded model output. Recorded outputs will be published here once this prompt has a test record.
Testing and editorial review
Status: Untested. Reviewed for structure and clarity by the editorial team. No recorded model test yet.
No test record has been published for this prompt yet. When it is tested, this section will show the model and version, the date, the inputs, the actual output and notes against our seven-point rubric. We do not publish numerical scores.
| Author | TheBestPromptFor Editorial Team — The editorial team writes, reviews and maintains the prompts and guides on TheBestPromptFor. Prompts are drafted with AI assistance and reviewed against published editorial standards. |
|---|---|
| Published | November 15, 2026 |
| Last reviewed | November 15, 2026 |
| Last tested | Not yet tested |
| Version history |
|
Limitations and things to check
- The review can only be as complete as the sources you paste. It is not a literature search and will not find papers you have missed.
- The model can still misread a source. Check findings, sample sizes and direction of effects against the originals.
- Very long pastes may exceed what a model can read in one conversation; split the sources into batches and merge the evidence tables.
- Citation formatting follows the details you supplied and may contain style errors. Run the final reference list through a reference manager.
- Your institution or journal may have rules on AI assistance in academic writing. Check them and disclose use where required.
Alternative prompts
Research
Research Paper and Report Summary Prompt
Summarizes a paper or report for a named reader: bottom line, method, key findings with exact figures, limits of the evidence, and what it means for them.
- ChatGPT
- Claude
- Gemini
Content Writing
Article Outline Prompt
Builds a section-by-section outline around one central question, with the evidence for each section and a list of what still needs research.
- ChatGPT
- Claude
- Gemini
Education
Study Guide from Your Notes Prompt
Builds a study guide from your own notes or readings: topic map, core content, commonly confused ideas, self-test questions with separate answers, and a revision schedule.
- ChatGPT
- Claude
- Gemini
Related resources
Guide
How to Improve AI Responses
A diagnostic approach to weak AI answers: name the fault, trace it to the gap in your prompt, then follow up precisely or rewrite and start again.
Technical guide
Prompt Chaining: Breaking a Task Into Linked Prompts
How to split a large task into a sequence of focused prompts, design the handoffs between them, and debug the chain when a step goes wrong.
Frequently asked questions
Can I ask the model to find the papers for me instead?
Not with this prompt. It is built for synthesis of material you already hold, because that is where the model's output can be checked line by line. Use a scholarly database or a search-enabled research tool to find papers, read them, then bring them here.
What if I only have abstracts?
The prompt will still run, and it will label those sources as abstract-only and avoid describing their methods in detail. Treat the result as a map of what to read in full, not as a finished review, since abstracts often omit limitations and effect sizes.
How many sources can I include?
It depends on the model's context limit and the length of each source. If the inventory comes back with fewer sources than you pasted, the input was probably too long. Run steps 1 and 2 on batches, then paste the combined evidence table and ask for steps 3 to 5.
Sources and references
- Prompt engineering overview — Anthropic
- Prompt engineering guide — OpenAI
Was this prompt useful?
No account needed. Do not include personal information.