RACE Prompt Framework: Role, Action, Context, Expectation

Study guide

RACE Prompt Framework: Make Expectations Explicit

RACE is a simple four-part prompt framework for professional work: Role, Action, Context, Expectation. Its strength is the Expectation slot, which tells the AI what a good answer must contain before it starts. Most people skip this slot, and that is why their prompts produce answers that read well but hide their guesses. This course teaches you how to use all four parts, with a special focus on Expectation, so you get answers you can check, defend and reuse.

You will learn how to set a clear role, choose the right action verb, supply the material and constraints, and spell out quality checks such as listing assumptions, showing working, citing source lines and marking uncertainty. You will see how RACE compares with lighter frameworks like RTF and RISEN, and when to switch to CO-STAR for audience and tone. Finally, you will build a reusable RACE prompt for a task you do regularly, with an Expectation block that becomes your personal quality standard.

What is a prompt framework?

A prompt is the request you type into an AI tool such as ChatGPT, Claude or Gemini. A prompt framework is a reusable structure for that request: a short checklist of the parts a good prompt contains, usually named with an acronym so it is easy to remember. Why it helps: the AI fills in whatever you leave out with generic guesses. A framework makes sure you include the parts that change the answer, such as who it is for, what you want to achieve and the format you need.

Frameworks make prompts repeatable: you fill the same parts each time, save the prompt as a template and share it with colleagues. They are not magic words: the value comes from the detail you put in each part. They help most when a request is complex or open to interpretation; a quick, simple question can stay short. There are well over sixty named frameworks, and most of them remix the same ingredients: role, task, context, audience, format, constraints and examples.

In this course you learn RACE, which is a fill-in-the-slots framework: an acronym where each letter stands for one part of your prompt. It is best for: analysis and professional tasks where the expected quality must be spelled out.

What RACE adds: the Expectation slot

RACE stands for Role, Action, Context, Expectation, written in that order. It is one of the most used frameworks for everyday professional work because it covers the four things every prompt needs: who the AI should be, what it should do, what material and limits it has, and what a good answer looks like. The Expectation slot is the part most people skip. It tells the model what a good answer contains before it starts, such as a table, named assumptions, visible working, and marked uncertainty.

A prompt without Expectation often reads well but hides its guesses. A prompt with Expectation names the guesses and flags what is uncertain, so you can check the right things. Each slot does a different job. Role sets the point of view, Action states the task, Context gives material and limits, and Expectation defines the shape of the answer. Note that in marketing, RACE means Reach, Act, Convert, Engage. That is unrelated to this framework. Always check the four words, not just the acronym.

Role and Action: expertise and the verb

Role sets the expertise and standards the model should apply. A specific role such as a senior credit analyst gives it a clear professional lens, while a vague role gives it nothing to aim at. Action is the verb that defines the work. Analyse, compare, extract, rank and explain each point at a different job, so the verb you choose shapes the whole answer. Analysis verbs work best with material pasted in. Without the source document, the model fills the gaps from general knowledge and the answer stops being about your case.

Stacking several roles in one prompt blurs the standards. Pick the single role that matches the decision you are making, and keep the action focused on one job. Role and Action are the first two letters of RACE and they set up the rest. Context supplies the material and constraints, Expectation says how the answer should be shaped.

Context: material and constraints

Context has two halves: the material, meaning the data and documents, and the situation, meaning the deadline, audience and decision the answer feeds. Most weak prompts supply neither. Ordering matters most on long inputs. Put the long material at the top and the question at the end. Anthropic reports up to 30% better answers on long inputs with this order.

State what the analysis is for. The same document reviewed for a signature decision and for a training session should produce different emphasis, and only the purpose tells the model which one you need. Paste the real material rather than describing it from memory. A summary of a clause is not the clause, and the model can only flag risks that are actually present in the text you give it. Keep the situation honest and specific. A Friday deadline and a go or no-go decision change how deep and how long the answer should be, so vague situations produce vague answers.

Expectation: list assumptions, show working, mark uncertainty

Expectation is the fourth part of RACE and it spells out quality. It tells the AI what a good answer must contain, not just what topic to cover. The four quality checks are: list your assumptions, show your working, cite the source line, and mark what is uncertain. Together they make an answer auditable. Asking the AI to separate facts from interpretation makes errors visible. You can see which lines rest on evidence and which rest on the model filling gaps.

Expectation also covers format. Name the shape you want, such as a table with columns Risk, Likelihood, Mitigation, at most eight rows. Vague quality words like accurate or thorough do nothing. Name the specific checks instead, so the AI has something it can actually run before it answers.

RACE versus RTF and RISEN

RTF is the lightest of the three. It covers Role, Task and Format, which is enough for quick jobs like a short summary or a tidy list, but it has no place to state expectations. RISEN adds Instructions, Steps, End goal and Narrowing. That structure suits longer work where you need the AI to follow a sequence and stay inside a scope. RACE adds the Expectation slot. That is where you ask for assumptions, working and uncertainty marks, which is what makes research and analysis output easier to trust.

RACE has no audience or tone slot. For writing aimed at other people, such as client emails or board notes, use CO-STAR, which has dedicated slots for both. Add a chain-of-verification step when accuracy matters. Ask the AI to list its claims, then check each one against your source, and mark anything you cannot check.

RACE in your job: build a reusable prompt

The Expectation line is the part of RACE that most people skip, and it is the part that protects you. It tells the model to list assumptions, show working, and mark uncertainty, so you get an answer you can defend rather than one you have to trust blindly. Reuse the same Expectation block across every prompt you write. Because the standard never changes, you can compare answers across tasks and across weeks, and you stop reinventing your quality bar each time.

Review the answer yourself against the Expectation before you use it. Then ask the model to run the same check before it hands anything over. This gives you two passes over the same standard with almost no extra effort. Pick one regular analysis as your template. A weekly summary, a monthly review, a risk assessment. Writing RACE once for a task you repeat saves you time on every future version of that task. A prompt without an Expectation tends to produce a confident answer with no working and no warnings. Adding the Expectation line turns that into an answer that shows its reasoning and flags what it does not know. RACE works best when each slot stays short and clear. Role says who the model is, Action says what to do, Context gives the material and limits, and Expectation says what good looks like. Four short parts, one strong prompt.

Frequently asked questions