RISEN Prompt Framework: Multi-Step Work Done Right

Study guide

RISEN: Control Multi-Step AI Work Properly

Most prompt advice works fine until the task gets bigger than one instruction. Ask for a short rewrite and a simple prompt is enough. Ask for a one page policy, a rollout plan or a specification with five sections, and the model starts guessing at the order, the length and the level of detail. RISEN is a six part structure built for exactly that kind of work. Role names the expert whose standards the answer should meet. Instructions state the deliverable in a sentence or two. Steps set the order of work. End goal describes what done looks like and who will use it. Narrowing adds the fences on length, tone, format, sources and exclusions. The framework grew out of the simpler RISE structure, which stands for Role, Input, Steps and Expectation, by adding Narrowing for precise control over constraints and scope.

This course is for busy professionals who use ChatGPT, Claude or Gemini in a real job and need multi-step deliverables they can actually hand to someone else. You will see how RISEN differs from RISE and when the heavier frame earns its keep. You will learn to write a Role that changes the level of detail, Instructions that describe the destination rather than the route, numbered Steps that let you approve each stage, an End goal with success criteria, and a short Narrowing line that removes the most common guesses. Every lesson includes a worked example from ordinary office work and the mistake that trips most people up, such as conflicting rules in the Instructions or steps so broad there is nothing to review. By the end you will be able to turn one monthly deliverable into a reusable template and run a critique loop that checks the draft against the end goal you wrote.

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 RISEN, which is a fill-in-the-slots framework: an acronym where each letter stands for one part of your prompt. It is best for: plans, policies, specifications and other multi-step deliverables.

From RISE to RISEN: when the order of work matters

RISEN was created by Kyle Balmer and shared through his prompt content on TikTok and LinkedIn in 2023 and 2024. It is a practical structure, not a private formula. It grew out of the simpler RISE framework, which stands for Role, Input, Steps and Expectation. Those four parts still sit inside RISEN unchanged. The added part is Narrowing, which gives precise control over constraints and scope. It sets what to include, what to leave out, how long the answer should be and what format to use.

The reason to reach for RISEN is order. Use it when a single instruction is not enough because the order of the work matters. If step one must happen before step two, RISEN keeps that order visible in the prompt instead of leaving the model to choose its own sequence. A one line ask like write a summary of our return policy is fine with a light frame. A request like research the current process, outline the gaps, draft a one page policy for line managers and check it against our legal limits is not, because a wrong order produces a draft you cannot use.

Every part of the prompt should be short and specific. Long vague slots give the AI room to guess, which is exactly what RISEN is designed to prevent. The framework is heavy by design, so treat it as the right tool for multi-step deliverables and not a default for everything. The next five sections walk through each part in turn, starting with the pair that sets the foundation: Role and Instructions.

Role and Instructions: the standard and the target

Role names the expert whose standards the output should meet. It changes the level of detail, the priorities and the tone of the answer, because it tells the AI which kind of reader to write for. A prompt that opens with you are an experienced HR policy writer produces different work from one that opens with you are a facilities coordinator. Same task, different vocabulary, different assumptions about what the reader already knows. Pick the role that matches the standard you want, not the most impressive job title you can think of.

Instructions say what to produce in one or two sentences. They describe the deliverable, such as a one page policy or a five row table, and leave the method to the Steps section. Instructions and Steps are different jobs. Instructions are the destination, Steps are the route. If your Instructions describe how to work, you have written Steps in the wrong slot. A useful test is to read the Instructions aloud and ask whether they could be answered with a single finished document sitting on a desk.

Newer models follow instructions literally, so contradictions are costly. A rule like keep it short sitting next to a rule like cover every possible case will not both survive, and the model will quietly pick one. Before sending, read your Instructions aloud and ask whether any two rules pull in opposite directions. Conflicting rules waste effort and force a rewrite. Role and Instructions work as a pair. The Role sets the standard, the Instructions set the target, and together they give the rest of RISEN a clean foundation.

Steps: control the process

Steps are the order of work written into the prompt: research, then outline, then draft, then check. Without them the model chooses its own order and you cannot see the stages. Numbered steps make the model show each stage, so when the output is wrong you can point at the stage that failed instead of rewriting the whole prompt. That single habit saves more time than any wording trick, because it turns a mystery into a diagnosis.

For long tasks, ask the model to do step one only and wait. This turns one risky prompt into a short conversation where you approve each stage before the next one starts. The stop rule matters. If you do not say wait for my reply, the model will run every step in one go and you lose the chance to correct the early stages. A check step keeps quality inside the process. It asks the model to test the draft against the end goal before you read it, which catches gaps while the context is still fresh.

Steps that are too broad, such as research and write the specification, give you nothing to review. Split them into small actions you can judge in seconds. Research the current process becomes list the five steps in the current process and note where delays happen. Write the specification becomes draft the specification in four sections, one page maximum. If a step takes you more than a few seconds to check, it is probably two steps wearing one label.

End goal: what done looks like

The end goal is the E in RISEN. It describes the finished result and the person who will use it, such as a two page policy a line manager can apply without HR. A clear end goal lets the model make sensible trade-offs inside the steps. It can shorten, add or drop content because it knows what the final result has to do. Without it, the model optimises for looking thorough rather than being useful, and you get a long document nobody can act on.

Success criteria are the checks you run before accepting the work. What must be true for this to be good enough. Fits two pages. Covers three cases. No legal terms. Written so a new starter can follow it. These are short, testable statements, not aspirations. They also give the model something to aim at during the draft, which is why they belong in the prompt rather than in your head.

Name the reader in the end goal. The same content written for a line manager, a new starter or a customer will differ in tone, length and detail. Describe the result, not the topic. Write about the timetable change is a topic. A one page plan the office team can run is a result. Put the end goal near the top of the prompt, before or after the steps, so the model reads the target before it starts working.

Narrowing: constraints and exclusions

Narrowing is the N in RISEN. It defines length, tone, format, allowed sources and exclusions so the AI stops guessing and starts matching your real requirements. Exclusions are often the most valuable part. Lines like no legal advice, no new budget, or UK law only keep the answer inside limits you can actually use. A draft that recommends a budget you do not have is not a draft, it is a rewrite waiting to happen.

If an answer drifts, the fix usually belongs in Narrowing. Too long means set a length. Wrong tone means name the reader. Invented facts means set the sources. Off topic means add an exclusion. Sources are a fence too. Telling the AI to use UK law only, or your own policy document, prevents it from pulling in rules or figures that do not apply to your work. This is the single most common cause of confident but unusable answers.

Narrowing works best as a short, plain line at the end of the prompt. You do not need long explanations. Five or six clear fences are enough to shape the output. A good checklist prompt shows Narrowing in action. Length, tone, format, sources and exclusions each remove a guess, so the first answer is closer to final. When you review a prompt that keeps producing near misses, count the fences. There are usually fewer than you thought.

RISEN in your job: templates and the critique loop

A monthly deliverable is the best first candidate for a RISEN template because it returns on a schedule, its steps rarely change, and you already know what good looks like. Writing it once removes the need to rebuild the prompt every cycle. Pick the report, plan or policy you produce on a regular basis, write the six parts once, and save it somewhere you will find it. The second run takes minutes instead of an afternoon.

RISEN is heavy by design. It carries a role, instructions, ordered steps, an end goal and narrowing. For quick asks such as a short rewrite or a single fact, a lighter frame like RTF does the job with less scaffolding. Match the frame to the task. Using RISEN on a two-line ask adds effort without adding quality, and using a light frame on multi-step work leaves the process uncontrolled.

The critique loop is the step most people skip. Draft with RISEN, paste the output back, and ask the model to review it against the end goal you wrote. The review surfaces what the first draft missed. The end goal slot does double duty here. It tells the model what done looks like on the first pass, and it becomes the standard you measure the draft against in the review. Templates should evolve. When a review finds a recurring gap, fix the template rather than the individual prompt, so the improvement carries into every future cycle.

Frequently asked questions