A Research Prompt Gallery

// how to use this gallery

How to use this gallery

Every prompt below is a starting point, not a script. Paste in your own material where it says to, change the audience and format to suit, and keep refining in the same chat until the answer fits. Two habits apply to everything here:

Paste the real text you want worked on, rather than asking the AI to recall facts. It invents far less when it is working from what you give it.
Keep data safe. Do not paste identifiable participant information, or community or cultural material, into a public tool. De-identify first, or use an approved tool. See the responsible-data page.

Across all of these, the pattern from the prompting page holds: task, context, role if it helps, and format. The examples just show it in action for research work.

// understanding and explaining

Understanding and explaining

Make sense of a dense paper.

Here is the abstract and results section of a paper [paste]. Explain in plain English what they did, what they found, and what the main limitation is. Then list three questions a careful reader should ask about it.

Why it works: it gives the AI the actual text, asks for plain language, and ends with a critical-thinking step rather than a summary alone.

Interpret a statistical output.

Here is a regression output [paste]. Explain step by step what the hazard ratio and confidence interval mean here, in language a non-statistician on our team could follow. Think step by step, and tell me where I should be cautious.

Why it works: "think step by step" improves statistical reasoning, and asking for caution surfaces the caveats.

// plain-language summaries

Plain-language summaries

Turn findings into something a community audience can read.

Here are the key findings from our study [paste]. Write a one-page plain-language summary for a community audience with no research background. Warm and respectful tone, short sentences, no jargon, and define any term you cannot avoid. Leave the actual figures as I have written them; do not change any numbers.

Why it works: it sets audience, tone and reading level, and explicitly protects the numbers from being altered.

Draft a plain-language statement for participants.

Draft a plain-language description of what taking part in this study involves, based only on the details I give you here [paste]. Keep it factual, do not add anything I have not stated, and flag with [VERIFY] anything you think is missing.

Why it works: it restricts the AI to your details and asks it to flag gaps rather than fill them.

// drafting

Drafting

A first draft of a report section.

You are helping me draft the background section of a report for [audience]. Using only the points I list here [paste bullet points], write about 300 words in plain, professional English. Leave a [CITATION NEEDED] marker anywhere a source is required; do not invent references.

Why it works: it builds from your points, sets a length, and blocks the most common failure (made-up citations) up front.

A grant or funding section.

Here are notes for the significance section of a grant [paste]. Draft it in about 250 words, persuasive but measured, for a health research funder. Do not add statistics or references I have not provided; mark places where evidence would strengthen it with [ADD EVIDENCE].

Why it works: it keeps the AI to your material and turns its tendency to invent into a useful "where to add evidence" map.

A professional email or letter.

Draft a short, polite email to a partner organisation asking to reschedule next week's meeting because of fieldwork. Keep it warm and brief. Here is the tone I usually write in [paste a past email].

Why it works: a sample of your own writing makes the AI match your voice instead of a generic corporate one.

// reworking your own writing

Reworking your own writing

Shorten without losing meaning.

Here is a section I have written [paste]. Cut it to half the length, keep every key point, and keep my voice. Do not add anything new.

Shift the audience.

Here is a paragraph written for researchers [paste]. Rewrite it for a general community newsletter, same facts, plainer language, friendlier tone.

Why both work: you give the AI finished text and a single clear change, which is the kind of editing it does well and safely.

// working with qualitative material

Working with qualitative material

Help structure your own coding.

Here are six de-identified interview excerpts I have cleared to use [paste]. Suggest a set of possible themes across them, with a short label and a one-line description for each, and point to which excerpt supports each theme. These are suggestions for me to check, not final codes.

Why it works: it treats the AI as a second opinion you verify, and the wording ("de-identified", "cleared to use", "for me to check") keeps the data and the method in your control.

Caution

qualitative data is often deeply identifying, and may be culturally sensitive or community-held. Only use material you are cleared to use, de-identified, in an approved tool. The decision about what is appropriate to use sits with the relevant people and communities, not with the AI.

// literature and evidence

Literature and evidence

Refine a search question.

Help me turn this rough question into a PICO framework, and suggest a draft Boolean search string for PubMed. I will have an information specialist check the string before I use it.

Why it works: it uses the AI for structure while keeping the human check that formal searching requires.

Synthesise papers you have verified.

Here are eight abstracts I have already found and checked [paste]. Summarise what they collectively suggest, where they agree, and where they conflict. Use only these abstracts; do not bring in outside studies or add citations.

Why it works: it confines the AI to verified material, which is the safe way to use it for evidence (see the literature page). Never ask it to "find papers".

// talking to services, funders and partners

Talking to services, funders and partners

Prepare for a conversation.

I have a meeting with [type of stakeholder] about [topic]. Based on what I tell you here [paste], help me prepare: likely questions they will ask, and clear, honest answers I could give. Flag anything I should check before the meeting.

Why it works: it rehearses rather than scripts, and the "flag anything to check" line keeps you from walking in with unverified claims.

// tips that make every prompt better

Tips that make every prompt better

Give context first. One or two sentences about your task, audience and what you have already done.
Ask for a format. A table, a word count, plain English, placeholders to fill in.
Block invention. "Use only what I provide. Do not invent citations or statistics. Flag uncertainty with [VERIFY]."
Refine, do not restart. Say what is wrong and ask for another go in the same chat.
Ask for confidence. "How sure are you, and where can I verify this?"
Protect data. De-identify, and keep community and cultural material out of public tools.

// glossary

Glossary

Prompt: the message you type to an AI.
PICO: a way of structuring a clinical or health question: Population, Intervention, Comparator, Outcome.
Boolean search string: a search built from terms joined by AND, OR and NOT, used in research databases.
Qualitative coding: labelling themes in text such as interviews; here, the AI suggests, you decide.
Placeholder: a marker like [name] or [CITATION NEEDED] that keeps sensitive details out or flags work still to do.
[VERIFY] flag: an instruction asking the AI to mark anything it is unsure of, so you can check it.

Take it with you: a one-page set of these prompt starters is available to download from the button at the top of this page.