How do you prompt AI to write in a register you don’t normally use, while ensuring statistical accuracy?

Context

CT coordinators must communicate results to participants in plain language (National Statement 3.1.72). This exercise demonstrates two techniques: (1) using AI as a statistical checker before translation, and (2) copy-editing the source to identify unclear sections—avoiding propagating confusion into the plain language version.

Materials

Setup (3 min)

  1. Download the Ramnarayan et al. study and RCH materials
  2. Open Claude or Gemini

Task - Step 1: Statistical Analysis and Clarity Check

In a new chat, upload the Ramnarayan et al. study and prompt the AI as follows:

Hi Claude. Please take the role of a highly meticulous copy-editor. We are preparing a plain language final letter communicating research results to participants. Before we do that though, we want to perform a final check of our preprint paper for the correct use of statistics, clarity, consistency, and flow.

You will take extra time and be extremely thoughtful in your discussions and considerations. The user, above all else, requires thoughtful and reasoned editing and discussion. We need to ensure that everything is in a single, consistent, and clear register. We also need to remove some words, where possible.

I would like you to highlight issues rather than making corrections. Then, for each issue, please thoughtfully discuss the issue. Do not suggest issues, but rather highlight your concerns, especially in items that may be confusing to a non-expert reader.

To begin, please provide a functional decomposition of these tasks and give me a readback of the headers of the document.

Attach the Ramnarayan et al. study or copy/paste from PubMed. Once you run it, make sure that you see all the headers!

Good.

Please make these specific checks:

  1. Extract statistical claims: List the main quantitative findings with their statistics (e.g., effect sizes, p-values, confidence intervals)

  2. Statistical appropriateness check: Are the statistical methods appropriate for the claims being made? Are there any red flags or overstatements?

  3. Methodology and results in plain language: How would you explain what they did to someone without statistics training? What are some concerns around communicating nuance that we should be aware of?

  4. Clarity issues in the original: Are there any sections of the original paper that are unclear, inconsistent, or have flow problems? Highlight these issues rather than making corrections.

Present this as a reference document I can check the final letter against.

Task - Step 2: Changing register to RCH Plain Language (2 min)

First, make a new system prompt. System prompt:

You are a plain language specialist helping draft final letters to research participants. All communications must meet RCH Plain Language Adviser standards: Grade 6-8 reading level, short sentences, active voice, no jargon without explanation. Be warm but not patronizing.

User prompt:

Now, I need to write a plain language final letter communicating research results to participants. This letter must meet RCH Plain Language Adviser standards.

Before writing, we will need to consider what makes an effective final letter, quote components from the attached RCH Plain Language Guidance, Example, and Style Guide. Please functionally decompose these documents with constraints to the appropriate register.

Attach the RCH Plain Language Guidance, Example, and Style Guide.

Task - Step 3: Draft with RCH Register (7 min)

Using the statistical analysis from Step 1 and the outline from Step 2, make an outline, then draft the letter in plain language appropriate for participants with no medical training. Use the register, tone, and structure from the attached RCH examples.

Debrief

  • How did the statistical analysis step change your approach?
  • What clarity issues did AI identify in the original paper?
  • How would you verify the final letter’s medical accuracy?
  • When would you use this multi-step approach versus a single prompt?