AI Prompts for Academic Writing: A Mega Guide for Researchers

AI prompts for academic writing work best when you give the model your own material and one clear job. For example, ask it to "turn these notes into a methods section" or to "list what a reviewer would question in this paragraph". This is a mega guide of AI prompts for researchers and academics to use during their research and writing, from the research question to the reviewer response. It also covers the rules journals and funders now set for AI use.

Each prompt is ready to copy. Replace the parts in [square brackets] with your own details, and paste your real text, data and sources where the prompt asks for them. The prompts work in ChatGPT, Claude, Gemini and other chat assistants, and you can write them in any language.

AI is now part of most researchers' work. In Wiley's 2025 ExplanAItions study, 84% of researchers said they use AI tools in their work, up from 57% a year earlier. Of those surveyed, 62% use AI for research or publishing tasks. The same study found that 87% have concerns about the models, including inaccurate answers, hallucinations and privacy. The prompts below are written with those concerns in mind: you supply the facts, and the model organizes, checks and edits.

What is an AI research prompt?

An AI research prompt is the instruction you give a chat assistant for one research task, together with the material it needs to do it. A good research prompt names the task, gives your field and audience, includes your own text or data, and says what the answer should look like.

Most strong prompts have the same six parts. You don't need all six every time, but each one you add makes the answer more useful.

A short prompt leaves most of these parts out, so the model fills the gaps with guesses. Here's the same request written two ways.

How to write better prompts for research

The companies that build these models publish their own prompting guides, and their advice is very similar: be specific, give context, show examples and say what format you want. Anthropic's prompting best practices, OpenAI's prompt engineering guide and Google's prompt design strategies all cover the techniques in this table.

The order of your prompt matters with long texts. Anthropic says that putting long documents above the question can improve answers by up to 30% on complex tasks with several documents. So paste the paper first and ask your question at the end.

A reusable template for any research prompt

Use this template when none of the prompts below fits your task. It covers all six parts of a strong prompt.

You are helping me with [task] for a [paper type] in [field].
Audience: [journal, examiners or readers].
Here is my material:
[paste your text, notes, data or sources]
Your task: [one clear job].
Rules: Use only the material above. Do not add facts, numbers or citations.
If something is missing, ask me instead of guessing.
Format: [length, structure, tone].

Let the model ask you questions first

When a task is large, such as planning a chapter, the model does better if it knows your situation. This prompt makes it ask before it writes.

Before you start, ask me up to 5 questions about my research, field, audience and target journal that would help you do this task well.
Wait for my answers before you write anything.
Task: [describe the task]

Which AI is best for research writing?

For research writing in October 2026, Claude Opus 5.5 is a strong default, and GPT-6 Astra and Gemini 3 are good alternatives. All three follow the prompts in this guide, and all three can work with long documents. The bigger difference is in the work around the prompt: how you upload files, keep projects and share chats.

General chat assistants are what most researchers use. Wiley's study found that 80% of researchers have used a general-purpose AI tool, against 25% who have tried a tool made for research. Our ChatGPT vs Claude for research comparison covers the models in detail, and our guide to Claude for academic research writing explains its settings.

Before you paste unpublished work, check the tool's privacy settings. Elsevier asks authors to check the terms of any AI tool so that their unpublished manuscripts stay confidential and aren't used to train the tool.

Prompts for finding a research question

These prompts help you test and sharpen a research question. They work best when you paste what you already know about the literature, so the model builds on your reading.

Turn a topic into research questions

My research area is [topic] in [field]. Here is what I know about the current literature: [paste 5 to 10 bullet points from papers you have read].
Suggest 5 research questions that address a gap in this literature.
For each question, say which gap it addresses and what kind of study could answer it (for example a survey, an experiment or a qualitative study).
Do not cite any papers.

The bullet points keep the questions tied to real gaps. Without them, the model suggests questions that may already be answered. Our free research question generator does the same job in a form.

Check a question against the FINER criteria

Here is my research question: [question].
Assess it against the FINER criteria: feasible, interesting, novel, ethical and relevant. For each criterion, give a rating (strong, adequate or weak) and one sentence of reasoning. Then suggest one revised version of the question that fixes the weakest point.
My constraints: [time, budget, access to participants, ethics].

FINER is a common checklist for research questions. Your constraints matter most here, because feasibility depends on your time, budget and access.

Structure a clinical question with PICO

Rewrite my research question in PICO format (Population, Intervention, Comparison, Outcome). If my question is missing one of the four parts, point it out and ask me for it.
My question: [question]

PICO questions are standard in health research and make database searches easier, since each part becomes a search concept.

Draft testable hypotheses

My research question is [question]. My study design is [design] and I will measure [variables]. Write one null and one alternative hypothesis for each of my main outcomes. Use the variable names I gave you.

Asking for the null and alternative together helps you check that each hypothesis can be tested with your design. Try our hypothesis generator for more options.

Brainstorm title options

Suggest 8 titles for a paper with this abstract: [paste abstract].
Give 4 descriptive titles and 4 titles that state the main finding.
Keep each under [number] words and include the main terms [terms].

Two kinds of titles give you a real choice, since journals differ on whether titles should state results. Our title generator gives more variations.

Prompts for reading and analyzing research papers

These prompts help you read faster and take better notes. Paste the paper's text, or upload the PDF where the tool allows it, and ask about that paper only. A model that answers from memory can mix up papers or invent details.

Summarize a paper into structured notes

Here is the full text of a research paper: [paste text or attach PDF].
Summarize it under these headings: research question, design and sample, main methods, main findings (with numbers), limitations stated by the authors, and how it relates to my topic, [your topic].
Quote the sentence from the paper that supports each finding.

The quotes let you check each point in seconds. Our AI summarizer and our guide to summarizing a PDF with AI cover longer documents.

Extract study details into a table

From the papers below, build a table with these columns: author and year, country, design, sample size, measures, main result and limitations.
Use only information stated in each paper. Write "not reported" where a detail is missing.
[paste or attach papers, each labeled with author and year]

"Not reported" stops the model from filling empty cells with guesses, and it shows you which papers to read more closely.

Explain a method or statistic

In this paper, the authors use [method or statistic, for example "mixed-effects logistic regression"]. Explain what it does, why they might have chosen it, and how to read the result they report:
[paste the result sentence or table].
Write for a [PhD student / master's student] in [field].

Naming your level gets you an explanation at the right depth, without basics you know or jargon you don't.

Critically appraise a study

Act as a careful methods reviewer. Here is a study: [paste text].
Assess its internal validity, external validity, risk of bias and whether the conclusions follow from the results. List the three most serious weaknesses and quote the passage behind each one.

Use the answer as a starting point for your own appraisal, and check it against a formal checklist if your review needs one.

Compare two papers

Here are two papers on [topic].
Paper A: [paste text]
Paper B: [paste text]
Where do their findings agree, where do they conflict, and what differences in sample, method or measures could explain the conflict?
Present the answer as a table, then a 100-word summary.

Conflicting findings are often the most useful material for a literature review, and this prompt finds them for you.

Ask questions about a paper, with quotes

Here is a paper: [paste text].
Question: [your question]
First, list the exact quotes from the paper that are relevant to the question. Then answer using only those quotes. If the paper doesn't answer the question, say so.

This follows Anthropic's advice to ask for quotes first. The answer stays tied to the paper, and you can see where each point comes from.

Literature review prompts

These prompts help with searching, organizing and writing a literature review. Paste your own notes and sources, because a model asked to find literature from memory can invent references.

Build a database search string

I am searching [database: PubMed / Scopus / Web of Science] for studies on [topic]. My main concepts are [concept 1], [concept 2] and [concept 3].
Build a Boolean search string with synonyms and related terms for each concept, combined with AND and OR. Use the database's syntax for phrase searching and truncation. List the terms you added so I can check them.

Test the string in the database itself and adjust it. A subject librarian can check it too, which helps if you need a reproducible search.

Synthesize sources into a thematic paragraph

I'm writing a literature review paragraph about [theme]. Here are my sources and their key findings:
- [Author 1 (Year)]: [finding]
- [Author 2 (Year)]: [finding]
- [Author 3 (Year)]: [finding]
Write a synthesis paragraph that identifies patterns and contradictions across these sources. Do not summarize each source individually. Use citation format: (Author, Year). Approximately 150 words.

The instruction "do not summarize each source individually" matters most here. Without it, the model writes one sentence per source, which is a pattern reviewers often criticize in literature reviews.

Build a synthesis matrix

Here are my notes on [number] sources: [paste notes, one block per source].
Create a synthesis matrix with sources as rows and these themes as columns: [theme 1], [theme 2], [theme 3]. In each cell, write a short phrase from my notes or "not covered". Then list the themes with the least evidence.

A matrix shows you where several sources agree and where you have only one, which is useful before you write a single paragraph.

Identify gaps in your review

Here is my literature review section: [paste text]. Based on what's covered, identify 3 to 5 potential gaps or underexplored areas that my study could address.
For each gap, explain in one sentence why it matters. Focus on methodological gaps, population gaps, or conceptual gaps.

Use this to brainstorm. The model can't read the papers you haven't pasted, but it can spot patterns in how you've described them and suggest angles you missed.

Write an annotated bibliography entry

Write an annotated bibliography entry in [APA 7 / MLA 9] style for this source: [paste the reference and the abstract or your notes].
The annotation should summarize the source in 2 sentences, assess its strengths and limits in 1 to 2 sentences, and explain in 1 sentence how it relates to my project on [topic]. Use only the information I gave you.

Check the formatted reference against the source itself. Our annotated bibliography generator formats entries for you.

Draft screening criteria for a systematic review

I am planning a systematic review on [question]. Draft inclusion and exclusion criteria covering population, intervention or exposure, comparison, outcomes, study design, language and publication dates.
Present them as a two-column table. Mark any criterion that needs a decision from my team.

Agree on the criteria with your co-reviewers before screening starts. Our guide to a PRISMA-compliant systematic review with AI covers the rest of the process, and the PRISMA 2020 statement has the official checklist.

Choose a structure for your review

Here are the themes from my literature review: [list themes with a 1-sentence summary each]. Suggest two ways to organize the review (for example by theme, by method or by time) and give a heading outline for each. Say which one suits a review that argues [your main argument].

Seeing two outlines side by side makes the choice easier than starting from a blank page. Our literature review guide explains each structure.

Methods and data analysis prompts

Methods sections suit AI well, because you supply every detail and the model only organizes them. Data analysis prompts need more care: run the code yourself, and check statistical choices with a statistician when the stakes are high.

Draft a methods section from notes

Write a methods section based on these notes: [paste bullet points of what you did]. Field: [field]. Use past tense, passive voice where conventional, formal academic register. Organize into subsections: Participants/Sample, Data Collection, Analysis. Include enough detail for replication.
Approximately [word count] words.

The content is procedural and factual, so the risk of invented detail is low when your notes are complete. Our methodology guide lists what each subsection needs.

Check methods for missing details

Here is my methods section: [paste text]. Review it as a peer reviewer would.
Identify any missing information that would prevent replication of this study.
List each gap as a specific question I should answer in the text.

This finds omissions authors often overlook, such as sample size justification, ethical approval statements and software versions.

Check your paper against a reporting guideline

My study is a [randomized trial / observational study / systematic review].
Here is the checklist of the reporting guideline that applies to it:
[paste the CONSORT, STROBE or PRISMA 2020 checklist].
Here is my manuscript: [paste text].
Go through the checklist items one by one. Say whether each one is reported, partly reported or missing, and quote where it appears.

Many journals ask for a completed checklist at submission. The EQUATOR Network library has the guideline for each study type.

Choose a statistical test

Here is my design: [describe groups, variables and their types, sample size, and whether measures are repeated]. My question is [question].
Which statistical tests fit, and what assumptions does each one need?
Explain how to check those assumptions, and flag anything I should confirm with a statistician.

The answer gives you options and the reasons behind them. Confirm the final choice with your supervisor or a statistician, especially for complex designs.

Write analysis code

Write [R / Python] code to run [analysis] on a data file with these columns: [column names and types]. Include comments, check for missing values, and print the results in a table I can report.
Don't simulate or invent data. I will run the code on my own file.

Describing the columns means you never have to paste your raw data, which helps when the data is confidential.

Explain your output

Here is the output from my analysis in [software]: [paste output].
Explain each number in plain terms and tell me which values I should report in the results section of a [field] paper, and in what format.

Check the reporting format against your target journal or style guide, since conventions differ between fields.

Suggest first codes for qualitative data

Here are excerpts from my interview transcripts, with names and identifying details removed: [paste excerpts].
Suggest initial codes, each with a one-line definition and one example quote. I will review and refine the codes myself.

Remove names and identifying details before you paste interview data, and check that your ethics approval allows it. The codes are a first draft for your own analysis.

Results and discussion prompts

These prompts turn results into clear prose and keep your discussion honest about what the results show. The interpretation stays your job.

Describe statistical results in prose

Convert these statistical results into academic prose: [paste your stats tables or key numbers]. Field: [field]. Use the format common in [target journal or style guide]. Report exact values (means, SDs, p-values, confidence intervals). Use past tense. Do not interpret the results, only describe them.

The instruction "do not interpret" keeps the model out of your discussion section. You want a plain description here, with numbers turned into sentences.

Structure a discussion section

My key findings are: [list findings]. The existing literature shows: [key prior results]. My study's limitations include: [list limitations]. Create a discussion section outline with these subsections: (1) Summary of key findings, (2) Comparison with prior research, (3) Theoretical/practical implications, (4) Limitations, (5) Future directions. Under each, include 2 to 3 bullet points of what to cover.

The discussion is where many papers lose their structure, and an outline helps. Fill it in with your own analysis. Our guide to writing a discussion section shows each part in more depth.

Write a candid limitations paragraph

My study has these limitations: [list them plainly]. Write a limitations paragraph that acknowledges each one directly without being overly apologetic or dismissive. For each limitation, briefly suggest how it affects interpretation of results. Academic tone, approximately 200 words.

Reviewers respect a candid limitations section. This prompt stops the model from playing weaknesses down or hiding them in hedges.

Check your claims against your results

Here is my results section: [paste text].
Here is my discussion section: [paste text].
List every claim in the discussion and say whether the results support it fully, partly or not at all. Quote the supporting result for each claim.

Reviewers often point out claims that go further than the results. This check finds them before a reviewer does.

Draft a conclusion

Here are my main findings and their implications: [paste notes].
Write a conclusion of about [word count] words that answers my research question, says what the study adds, and names one practical implication and one direction for future research. Add no new results.

"Add no new results" matters, because a conclusion should only draw on what the paper has already shown.

Figure and table description prompts

Captions and table text should describe, and the interpretation belongs in your results. These prompts keep that line clear.

Draft a figure caption

Here is what my figure shows: [describe axes, groups, and the key pattern].
Draft a caption for a journal in [field]. First sentence states what the figure shows. Following sentences define abbreviations, sample sizes, and statistical annotations. Keep it under [word limit] words.

Captions often fail because they interpret the data. This prompt keeps the interpretation in the results text.

Turn a results table into prose

Here is my results table: [paste table]. Write a results paragraph that reports the values relevant to [hypothesis], in past tense, without interpreting them. Report exact values with their units and statistics as given. Do not round or invent numbers.

The instruction not to round or invent numbers matters, because models tend to smooth figures. Check every value against the table after drafting.

Write alt text for a figure

Write alt text for this figure for screen-reader users: [describe the figure or attach it]. In 1 to 2 sentences, state the chart type, what it compares and the main pattern. Don't repeat the caption word for word.

Alt text lets readers who use screen readers follow your figures, and some journals ask for it at submission.

Introduction prompts

Many researchers write the introduction after the results, once they know what the paper shows. These prompts help with its structure and opening.

Build an introduction framework

I'm writing the introduction for a paper about [topic] in [field]. My research question is [question]. The gap in current literature is [gap]. Create an outline for a 4-paragraph introduction that moves from broad context to specific gap to research question to brief method overview. Include suggested citation points marked as [CITE].

Use this for structure and write the prose yourself. The [CITE] markers show where you need real references, which you add from your own reading.

Strengthen your opening sentence

Here's the first paragraph of my research paper introduction: [paste paragraph].
Suggest three alternative opening sentences that would better capture reader attention. Options should be: (1) a surprising statistic approach, (2) a real-world problem framing, (3) a knowledge gap statement. Keep each under 30 words. Academic tone.

Three options let you compare approaches before you commit to one. If you pick the statistic, use one from a source you've read.

Check the problem, gap and aim

Here is my introduction: [paste text]. Identify the sentences that state (1) the problem, (2) the gap in current research and (3) my aim or research question. If any of the three is missing or unclear, say what a reader would expect to see and where.

The introduction is where an editor first sees what gap your paper fills. This check makes sure it states the gap and aim clearly. Our introduction guide has examples.

Abstract prompts

Write the abstract last, when the paper is finished. These prompts draft and tighten it, then turn it into the summaries and keywords journals ask for.

Draft an abstract from your main findings

Write a 250-word structured abstract for a research paper in [your field].
The study used [method] to examine [research question]. Key findings include [finding 1], [finding 2], and [finding 3]. The sample was [describe sample].
Use past tense for methods and results. Use present tense for implications.
Follow the structure: Background, Methods, Results, Conclusions.

This works because you provide the substance, and the model handles structure and word count. Our abstract writing guide and free abstract generator go further.

Tighten an existing abstract

Here is my draft abstract: [paste abstract]. Reduce it to exactly [word limit] words. Keep all key findings and the main conclusion. Remove hedging language unless it's scientifically necessary. Maintain formal academic register.

First drafts of abstracts often run long. This prompt is useful when a journal has a strict word limit.

Write a plain-language summary

Write a plain-language summary of this abstract for readers with no background in [field]: [paste abstract]. Use short sentences and everyday words, explain any technical term you keep, and stay under 150 words.
Don't add claims that aren't in the abstract.

Some journals and funders ask for a lay summary. Our plain-language summary generator is another option.

Suggest keywords

Suggest 6 to 8 keywords for this paper: [paste title and abstract].
Avoid words already in the title. Include terms a researcher in [field] would type into a database search.

Keywords that repeat the title waste space in indexing. Our keyword generator suggests more.

Write journal highlights

Write 3 to 5 highlights for this paper. Each highlight states one finding in under [character limit] characters. Use only this abstract:
[paste abstract].

Check your target journal's character limit for highlights before you use this prompt.

Thesis and dissertation prompts

A thesis is long, so these prompts work at the level of chapters and the whole document. Our AI workflow for a PhD thesis shows how they fit together.

Plan a thesis chapter by chapter

I am writing a [master's / PhD] thesis in [field] with five chapters: introduction, literature review, methodology, results, and discussion and conclusion. My topic is [topic] and my research question is [question].
For each chapter, list its purpose, its main sections, a word count out of [total words], and what I need ready before I start writing it.

The five-chapter structure is common in many universities. Change the chapter list if your program uses a different one, such as a thesis by publication.

Outline a research proposal

Draft an outline for a [length]-page research proposal on [topic] for [program or funder]. Include background, aims, research questions, methods, timeline, expected contribution and references to add.
Under each heading, list what I need to write, based on these notes:
[paste notes].

An outline built from your notes shows which parts are still missing before you start drafting.

Check a chapter outline

Here is my thesis chapter outline: [paste outline]. My research question is [question]. Identify sections that do not serve the research question, sections that overlap, and any missing step in the argument chain from literature to method to findings.

Supervisors read chapters one at a time, so structural problems can show up late. This runs the check early, while restructuring is still easy.

Check terms across chapters

I will paste terminology and definitions from two chapters of my thesis.
List every term that is defined differently, used inconsistently, or abbreviated in more than one way. [paste excerpts]

Long documents drift over months of writing. A terminology check before submission saves you from examiner questions about inconsistent terms.

Turn supervisor feedback into an action list

Here are my supervisor's comments on chapter [number]: [paste comments].
Turn them into a numbered action list. Group related comments, mark each one as major or minor, and note any comments that seem to conflict.

An action list makes a long set of comments easier to work through, and the conflicts are worth raising at your next meeting.

Practice for your viva or defense

Act as an examiner for my [PhD viva / master's defense] in [field].
Here is my abstract and a summary of each chapter: [paste text].
Ask me 10 questions an examiner is likely to ask, one at a time. After each answer I give, tell me what was clear and what I should strengthen.

Answering out loud, one question at a time, is close to the real thing. Our master's thesis guide covers the defense in more detail.

General polishing prompts

These prompts tighten and adjust text you've already written. Run the result through a proofreader afterward, because shortening text can introduce small errors.

Reduce word count without losing content

Here is a section of my paper: [paste text]. Reduce the word count by 20% while keeping all key information and arguments. Remove redundancies, tighten phrasing, and eliminate filler. Do not remove any technical content or citations. Maintain academic register.

Journal word limits are strict, and this prompt usually cuts 15 to 25% without losing content. Run the result through our AI proofreader to fix any grammar slips from the cuts.

Change the register for a new audience

Rewrite this text for [target audience: e.g., a general science audience / an undergraduate textbook / a grant application panel / a different discipline]. Original text: [paste]. Keep the core findings and arguments but adjust vocabulary, assumed knowledge level, and example specificity.
Approximately [word count] words.

Researchers write for many audiences, from journals to grant panels to the public. This prompt adjusts the language, so you can focus on which details each audience needs.

Improve the flow between paragraphs

Here is a section of my paper: [paste text]. Check the order of ideas and the transitions between paragraphs. Suggest where to add a linking sentence or move a paragraph, and explain each suggestion in one line.
Don't rewrite the text.

Suggestions instead of a rewrite keep your sentences intact, and you decide which changes to make.

Balance your hedging

Here is my discussion section: [paste text]. Mark sentences where my claims are stronger than my evidence, and sentences with so much hedging that the point is lost. Suggest a balanced version of each.

Hedging that matches your evidence makes a paper easier to trust, and reviewers notice both overclaiming and vague wording.

Editing prompts that keep your voice

General chat assistants tend to rewrite whole paragraphs. Heavy rewriting makes your text sound more like typical AI writing, which is what AI detectors flag. These prompts limit the model to corrections and suggestions.

Line edit without rewriting

Edit the following paragraph for grammar, punctuation, and clarity only.
Do not change sentence order, vocabulary level, or my phrasing choices unless they are errors. Return the edit as a list of changes with reasons.
[paste paragraph]

A list of changes keeps you in control of every edit, the same principle as tracked changes.

Compare your register with a target journal

Here is a paragraph from my draft and a paragraph from a paper published in [target journal]. Compare register, sentence length, and hedging style, and list specific adjustments that would bring my draft closer to the journal's style. Do not rewrite my paragraph. [paste both]

For the final grammar pass, a correction-only mode such as ProofreaderPro.ai's Light editing fixes errors without changing your sentence rhythm.

Edit your English with explanations

English is my second language. Correct the grammar, articles and prepositions in this paragraph, and keep my wording otherwise. After the corrected version, list each change with a one-line reason so I can learn from it. [paste paragraph]

The reasons turn each edit into a short lesson. Our guide to academic writing for non-native speakers covers the most common errors.

Translation prompts for research papers

Many researchers draft in their first language and publish in English. These prompts keep terminology and citations correct across languages.

Build a glossary before you translate

I will translate my paper from [language] into English. First, list the 30 most important technical terms in this text, with the standard English term used in [field] journals for each, as a two-column table.
[paste text]

A glossary agreed in advance keeps the same term translated the same way in every section.

Translate a section into academic English

Translate this section of my paper from [language] into academic English for a [field] journal. Use the terms in this glossary: [paste glossary].
Keep all citations, numbers and abbreviations exactly as they are.
Don't add or remove content.
[paste text]

Our academic translator keeps citations and discipline terms intact, and our guide to translating a research paper into English covers the full process.

Check a translation against the original

Here is my original text in [language]: [paste text].
Here is its English translation: [paste translation].
Compare them sentence by sentence and list every place where the meaning changed, something was left out or something was added.

A sentence-by-sentence check finds meaning changes that are easy to miss when you read the English version alone.

Prompts for reviewing a paper

You can use AI to review your own draft before you submit it. For someone else's manuscript under review, the rules are strict. The NIH prohibits its peer reviewers from using generative AI to analyze or critique grant applications. Elsevier's policy says reviewers shouldn't upload a submitted manuscript, or any part of it, into an AI tool. Elsevier does allow private AI tools to improve the language of your own review report, as long as confidentiality is kept.

Review your own draft as a tough reviewer

Act as a critical but fair peer reviewer for [journal] in [field].
Here is my manuscript: [paste text].
Write a review with these parts: a summary of the paper, major concerns, minor concerns and questions for the authors. For each concern, quote the passage and suggest how to fix it. Be specific and skip praise.

Running this before submission gives you time to fix weak points. Our journal manuscript checklist covers the rest of the preparation.

Score your paper against the journal's review criteria

Here are [journal]'s criteria for reviewers: [paste the criteria from the journal's website]. Here is my manuscript: [paste text].
Score my paper on each criterion from 1 to 5 and explain each score in two sentences.

Many journals publish their reviewer criteria, and pasting them in keeps the feedback close to what the real reviewers will judge.

Polish your own review report

Here is my draft review report: [paste your report, with no text copied from the manuscript]. Improve its clarity, structure and tone so the comments are specific and constructive. Keep every point and my recommendation. Don't add new criticisms.

Check the journal's reviewer policy first, use a private AI tool, and keep the manuscript itself out of the prompt. Elsevier asks reviewers to disclose AI use in their report.

Journal submission prompts

These prompts help with the documents that go in with your manuscript. Check each one against your target journal's author guidelines.

Write a cover letter

Write a cover letter to the editor of [journal] for my manuscript titled [title]. Main finding: [finding]. Why it fits the journal: [scope match].
State that the work is original and not under review elsewhere.
Keep it under 300 words.

Editors read many cover letters, so a short one that states the finding and the fit works best. Our cover letter generator and cover letter guide have templates.

Check how well your paper fits a journal

Here is [journal]'s aims and scope statement: [paste it from the journal's website]. Here is my abstract: [paste abstract].
How well does my paper fit? List the points of fit and the points of mismatch, and suggest how to frame the cover letter.

Choose candidate journals with a journal database or our journal finder, then use this prompt on each one. A model asked to name journals from memory can get scopes and details wrong.

Draft a data availability statement

Draft a data availability statement for my paper.
Data: [describe the data].
Where it is stored: [repository and DOI, or "available on request"].
Restrictions: [ethics or privacy limits].
Follow this example wording from [journal]: [paste example].

Journals word these statements differently, and pasting an example from your journal keeps yours in the expected form.

Write an AI disclosure statement

Write an AI use disclosure statement for my manuscript. I used [tool and model] to [tasks, for example "edit grammar in the discussion"].
I reviewed and edited all output. Follow this format from [publisher]:
[paste the publisher's template]. Keep it to 2 or 3 sentences.

Elsevier, for example, asks for the tool, the purpose and your oversight in a separate declaration. Our guides to the AI disclosure statement and to each publisher's wording have examples.

Check your manuscript against the author guidelines

Here are [journal]'s author guidelines: [paste them].
Here is my manuscript: [paste text].
Check the manuscript against the guidelines: word limits, section order, reference style, figure and table rules, and required statements.
List each problem with what to change.

Desk rejections for formatting problems are avoidable. Our journal submission checklist lists what to check by hand.

Reviewer response prompts

A response letter needs a respectful tone and clear answers. These prompts help you draft both and keep track of every change.

Draft a point-by-point response

Here is a reviewer comment: "[paste comment]". Draft a professional response that: (1) thanks the reviewer for the observation, (2) directly addresses their concern, (3) describes what changes were made or provides justification for not making changes. Keep the tone respectful but confident. Under 150 words.

The tone of a response letter affects how the revision is received. Use this draft as a base and adjust it to your review.

Make a defensive response calmer

Here is my draft response to a reviewer: "[paste your response]". Rewrite it to be less defensive while still making the same points. The reviewer should feel heard, not attacked. Keep the technical content identical.

It's normal to write a frustrated first draft. This prompt keeps your points and changes the tone. Our post on editing defensive language out of a rebuttal has more examples.

Build a response table

Here are the reviewer comments: [paste all comments].
Here are the changes I made: [paste notes].
Create a response table with three columns: the reviewer comment, my response, and where the change is in the revised manuscript (page and line). Write [ADD LOCATION] where I haven't given one.

Editors find a table easy to check. Our response-to-reviewers generator builds one from your notes.

Revise a passage to answer a comment

Reviewer comment: "[paste comment]".
Here is the passage they refer to: [paste passage].
Revise the passage to address the comment, and keep my findings and citations unchanged. Then summarize what changed in 2 sentences for my response letter.

The 2-sentence summary goes straight into your response letter, so the letter and the manuscript match.

Grant proposal prompts

Funders set their own rules on AI use in applications, so check them before you start. These prompts help you draft from your own plan and test it against the review criteria.

Draft a specific aims page

Draft a one-page specific aims page for a [funder] grant on [topic].
The problem: [problem]. My long-term goal: [goal].
The objective of this proposal: [objective].
Aims: [aim 1], [aim 2], [aim 3]. Expected outcomes: [outcomes].
Use only these details and mark any gap as [ADD].

The [ADD] markers show what your plan still needs. Our guide to editing a grant proposal with AI covers NIH, NSF and Horizon Europe formats.

Score a proposal against the review criteria

Here are the review criteria for [funding scheme]: [paste criteria].
Here is my proposal draft: [paste text].
For each criterion, say how a reviewer would likely score it and what in my text supports the score. List the three changes that would raise the score the most.

Reviewers score against published criteria, so testing your draft against the same list shows where to focus your revision.

Prompts for sharing your research

After acceptance, the same findings go into abstracts, posters, talks and posts. These prompts adapt your text for each format.

Write a conference abstract from your paper

Turn this paper abstract into a conference abstract of [word limit] words for [conference], with the headings [headings the conference requires].
[paste abstract]

Conference abstracts often have different limits and headings, and this saves you rewriting from scratch.

Write the text for a poster

Write the text for an academic poster from this paper: [paste abstract and main results]. Sections: title, background (40 words), methods (50 words), results (3 bullet points with numbers), conclusion (30 words).

Fixed word counts keep a poster readable from a distance.

Write a talk script

Write a speaking script for a [length]-minute talk on my paper for an audience of [audience]. Plan one slide per minute, with a slide title and what I say for each. Base it only on this text: [paste text].

A slide-by-slide script helps you time the talk before you rehearse it.

Announce your paper on social media

Write a short post announcing my new paper for [platform]. Include the main finding in plain words, why it matters, and a placeholder [DOI] for the link. No hype words. Stay under [character limit] characters.
[paste abstract]

"No hype words" keeps the post accurate to the paper, which matters when other researchers share it.

How to check what AI gives you

AI output needs checking before it goes into a paper, and references need the most care. In a study in Scientific Reports, Walters and Wilder (2023) asked ChatGPT for short literature reviews and checked 636 citations. 55% of the GPT-3.5 citations and 18% of the GPT-4 citations were fabricated, and many of the real ones had errors.

Our hallucinated-citation audit explains the checks in detail. The prompts below help with the first pass.

Find claims that aren't in your sources

Here are my source materials: [paste notes, data or source text].
Here is a draft written with AI help: [paste draft].
List every claim, number and citation in the draft that does not appear in the source materials.

Anything on the list needs a source or needs to come out of the draft.

Check your reference list for formatting

Here is my reference list: [paste references]. For each reference, check that the authors, year, title, journal, volume, pages and DOI are present and formatted in [style]. Flag anything incomplete or inconsistent.
I will verify each reference in a database myself.

This checks the format only. A reference can be perfectly formatted and still not exist, so look each one up as well.

Journal and funder rules for AI use

Most publishers allow AI help with writing, as long as you disclose it and take responsibility for the content. The details differ, so check your target journal's author guidelines before you submit.

The rules are still new to many researchers. In Wiley's 2025 study, 57% of researchers said they are held back by a lack of guidelines and training.

How to cite AI and disclose your prompts

APA Style's 2025 guidance on citing generative AI says to cite a specific AI chat when the chat helps readers, using the chat's share link. When you used AI to edit or organize your own writing, you can cite the tool in general and describe how you used it. Here is APA's reference template for a chat:

AI Company Name. (year, month day). Title of chat [Generative AI chat]. Tool Name/Model. URL of the chat

Elsevier suggests this wording for the declaration in a manuscript:

During the preparation of this work, the author(s) used [NAME OF TOOL / SERVICE] in order to [REASON]. After using this tool/service, the author(s) reviewed and edited the content as needed and take(s) full responsibility for the content of the published article.

Keep a simple log of the prompts you use for each paper: the date, the tool, the task and what you kept. It makes the disclosure statement easy to write and answers any question an editor asks later.

How to get more from these prompts

Every prompt above follows the same principles. Be specific about your field. Provide the actual content, and don't ask the AI to invent substance. Specify format, tone and length. Tell the model what not to do.

A few more habits help.

Chain prompts together. Use the introduction framework prompt to build your outline, then ask the model to draft each paragraph from that outline. A sequence of small prompts gives better results than one large request.

Always give the model your data. It should format, not fabricate. Paste your actual numbers and real source notes, which keeps invented details out of your draft.

Write in your own language if it helps. You can write any of these prompts in your first language and ask for the answer in academic English, or the other way round.

Edit everything. These prompts produce drafts, not final copy. Run the output through our AI proofreader for grammar and style, and through our AI summarizer if you need to condense sections. Then read it yourself before it goes into your paper. If AI-drafted text needs to sound more like you, our academic humanizer is designed to lower AI scores while it keeps your citations and terms.

Frequently asked questions

Q: What is the best AI prompt for writing a research paper?

The best prompt for a research paper gives the model your own material and one clear task, along with your field, audience and the format you want. Prompts that ask the model to write a whole paper from a topic produce generic text with invented details. Work section by section, starting from your notes and data.

Q: Do these prompts work with any AI model?

Yes. The prompts are plain instructions, so they work in ChatGPT, Claude, Gemini and other chat assistants. Results depend more on how specific the prompt is and on the material you paste than on the model. For long documents, use a model that accepts long inputs, such as Claude Opus 5.5 or GPT-6 Astra.

Q: Which AI is best for academic writing?

Claude Opus 5.5 is a strong default for academic writing in October 2026, especially for long papers and careful editing. GPT-6 Astra is a good choice for data analysis and code, and Gemini 3 works well inside Google's apps. Our ChatGPT vs Claude comparison covers the differences.

Q: Will using AI prompts to write my paper get me flagged for plagiarism?

Using AI to help you write isn't plagiarism in itself, but text an AI generates can be flagged by AI detectors, and journals expect you to disclose substantial AI use. Use the prompts for structure, analysis and editing, and rewrite drafted text in your own voice. Our paraphrasing tool for academic writing keeps your citations while it rewords.

Q: Should I disclose that I used AI prompts to help write my paper?

Yes, in most cases. Elsevier requires a declaration for AI use in preparing a manuscript, and basic grammar, spelling and punctuation checks don't need one. Rules differ by journal, so check the author guidelines. Our AI disclosure statement guide includes wording templates for both cases.

Q: How do I prevent the AI from making up citations?

Never ask the AI to find or suggest references from memory, because it can invent them. Provide your real sources in the prompt and tell the model to use only those. Mark citation points as [CITE] and fill them in yourself, then check every reference in a database or with our citation checker.

Q: Can I use AI to review a paper for a journal?

Not with the manuscript itself in most cases. Elsevier tells reviewers not to upload a submitted manuscript into an AI tool, and the NIH bans generative AI in its grant peer review. You can use AI on your own draft before submission, and Elsevier allows private AI tools to improve the language of your review report.

Q: How do I cite an AI chat in APA style?

APA's 2025 guidance uses this template: AI Company Name. (year, month day). Title of chat [Generative AI chat]. Tool Name/Model. URL of the chat. Use the chat's share link as the URL, and describe how you used the tool in your method section.

Q: Can I write prompts in my own language?

Yes. ChatGPT, Claude and Gemini understand prompts in many languages. Write the prompt in the language you think in, and say which language you want the answer in, for example "Answer in academic English".