Do You Need a Prompt Framework Like RACE, RTF, or COSTAR?

Short answer: No — not for most research tasks. Prompt frameworks like RACE (Role, Action, Context, Expectation), RTF (Role, Task, Format), and COSTAR (Context, Objective, Style, Tone, Audience, Response) are useful for building AI agents, but they're unnecessary for everyday market research work. What gets you better output is context and structure.

What Is a Prompt Framework?

A prompt framework is a structured template for writing AI prompts. Common frameworks include:

RACE — Role, Action, Context, Expectation
RTF — Role, Task, Format
COSTAR — Context, Objective, Style, Tone, Audience, Response

These frameworks were designed to help users give AI models enough information to produce useful output.

Do Modern AI Models Need Prompt Frameworks?

No. Large language models in 2025 are capable enough that you don't need to tell them to "act as a professional market researcher with 20 years of experience" before they'll write a useful discussion guide or questionnaire. The model already knows what those outputs look like.

The gap isn't capability. It's context — the specific information about your situation that the model doesn't have.

What Actually Gets Better AI Output: Context and Structure

What Is Context in a Prompt?

Context is the information you would give a junior researcher before handing them a task.

Example: If you asked a junior team member to draft a questionnaire for a concept test, you wouldn't just say "write me a questionnaire." You'd tell them:

The industry and customer type
The business question the questionnaire needs to answer
The target audience
Where demographics should appear (beginning, end, or split)
Screener requirements
Desired length
Whether open-ended questions should be included

That information is context. Add it to your prompt and output quality improves immediately — no framework required.

What Is Structure in a Prompt?

Structure means specifying how you want the output formatted.

For a questionnaire: Do you need it in Word or Excel? What's your variable naming convention — do variable names appear in parentheses before or after the question text? Do your scales run positive-to-negative or use numerical values?

Examples are the fastest way to communicate structure. If you have a past questionnaire that represents good work in your organization, include it. The AI will infer your formatting preferences more accurately from one example than from three paragraphs of description.

How to Improve AI Output After the First Draft

After AI produces output, ask it to grade itself on a scale of 0–100.

The grade is rarely 100%. When it isn't, you have two options:

Tell it to redo the task and improve the grade
Ask: "What do you need from me to improve your grade on this task?"

Option 2 is more efficient. The model surfaces the specific gaps in your original prompt — missing context, unresolved format questions. You fill those in, it reruns, and output quality jumps. Ask for the grade again to compare.

When Should You Use a Prompt Framework?

Prompt frameworks are the right tool for AI agents — systems that need to execute the same type of task repeatedly and consistently.

When building an agent, you need to specify:

The role it plays
The task it performs
The format it should always follow

That structure matters for agents. For one-off research tasks, context and structure get you there faster.

A useful signal: if you ask an AI to help you write a prompt for an agent, it will use framework structure automatically — because that's what agents need.

Summary: Prompt Frameworks vs. Context and Structure

Situation	Best Approach
Everyday research tasks (questionnaire, discussion guide, report)	Context + Structure
Repeatable automated workflows / agents	Prompt framework (RTF, RACE, etc.)
Improving output after first draft	Self-grading technique

Have a question about AI in your research workflow? Reply to the newsletter and I'll address it in a future edition.