Turn the question into a plan

Ask the tool to surface gaps before it writes a protocol. Compare what is feasible with what would provide stronger evidence.

Choose an assistant

Compare designs with an AI assistant

Use a tool already available to you, such as ChatGPT, Microsoft Copilot or Google Gemini. These are different services; responses and controls will vary. Sign-in, features, limits and organisational permissions vary. Use the prepared examples below if you cannot access a tool.

Open a new chat in your chosen assistant and paste the prompt below.
Compare the two proposals: what could you actually collect, and what else might explain a change?
Ask one follow-up about your most important concern. Compare the response with the prepared design options below.

Copy: compare study designs

For a hospital service, we want to evaluate whether introducing a pharmacist-led antimicrobial review process changes the proportion of adult treatment episodes with a review documented within 72 hours. Propose two options: A, a pragmatic before-and-after evaluation using routinely collected data over around three months; B, a potentially stronger longer-term design. Treat data availability and local circumstances as unknown. For each option set out the eligible population, intervention, comparator, primary outcome and timing, data source, feasibility and main threats to validity. Explain what each design can and cannot conclude, and what I should discuss with a statistician and local governance team. Do not invent details about the hospital.

Check: are population, outcome, comparator and timepoint specific? Have changes in documentation and other concurrent events been considered? The study classification and approvals must be decided locally.

Copy: ask for the missing details

I am a hospital pharmacist planning to evaluate: [INSERT QUESTION]. Act as a research methods adviser. Before proposing a design, ask me up to six concise questions, one at a time, about the population, current service, intervention, outcome, comparator, available data and practical constraints. Wait for each answer and accept “I don't know”. Do not assume approvals or data availability.

Here's one we prepared earlier

A / Pragmatic

Compare the proportion of eligible treatment episodes with a documented 72-hour review before and after introducing the process, using routinely collected records. Define the two periods, eligibility, review and documentation rules in advance. Feasible but changes over time and documentation practice could explain a difference.

B / Stronger

Consider an interrupted time-series study across multiple pre- and post-implementation periods, ideally with a comparable service as a control. It needs more observations, stable definitions and analytical support. Concurrent changes and choice of comparison still matter.

Both require local advice on whether the work is research, service evaluation or audit, and on any approvals needed. A before–after association alone does not establish causation.

Copy: challenge the plan

Critically review Design A as a methods reviewer. Identify the three most consequential weaknesses, how each could distort the findings, and one realistic mitigation for each. Separate changes I can make now from limitations I will have to report. Do not invent facts about my service.

Check: are the population, denominator, outcome and observation window precise? Is the comparison meaningful? Could other changes explain the result?

PreviousEvaluate articles

NextExplore with Gemini Notebook