The RTF Prompt Framework: Role, Task, Format Explained
The RTF prompt framework is probably the first structured prompting method most people run into, and it is easy to see why. It only asks for three things: a role, a task, and a format.
If you have typed a request into ChatGPT or Claude and gotten back something vague, generic, or oddly formatted, RTF is usually the fastest fix. It forces you to decide who the AI is acting as, what you actually want it to produce, and how the answer should be shaped, all before you hit enter.
How the RTF Prompt Framework Works
RTF splits a prompt into three labeled parts.
Role tells the AI who to be. This sets its assumed expertise, vocabulary, and point of view.
Task tells the AI what to actually produce. A clear task removes guesswork about scope, length, or deliverable.
Format tells the AI how to shape the answer, whether that is a table, a numbered list, a script, or plain paragraphs.
Here is what all three look like combined into one prompt:
Act as a nutritionist. Create a 3-day vegetarian meal plan for a busy student. Format it as a table with breakfast, lunch, dinner, and a grocery list.
Each part does distinct work. Drop the role and the plan reads like generic advice pulled from anywhere. Drop the format and you get a wall of text you have to reorganize by hand before it is actually usable.
If you find yourself rebuilding the same three-part prompt every time you open a new chat, it is worth turning that habit into a repeatable system instead of retyping it from memory. I put several practical AI resources in one place for exactly that.
Where the RTF Prompt Framework Actually Came From
RTF does not have a clear inventor attached to it, and that sets it apart from the frameworks it usually gets compared to. The RISEN framework, for instance, credits Kyle Balmer for creating and sharing it. The CLEAR framework traces back to a named academic paper by Leo S. Lo of the University of New Mexico.
One of the earliest public write-ups pairing RTF with other named frameworks is Moritz Kremb's newsletter, published October 13, 2023 on The Prompt Warrior, which lists RTF alongside RISEN and RODES, crediting RISEN to Balmer and RODES to another prompt writer. RTF gets no such credit, and it keeps showing up that way across newsletters and course slides published since.
That is worth knowing before you treat RTF as an authoritative method. It was not published by an AI lab, a university, or a peer-reviewed paper. It is a mnemonic that spread because three letters are easy to remember, not because someone tested it first.
Does RTF Actually Improve Your Prompts?
Most articles about RTF describe it as effective without offering evidence beyond a good-looking example. One recent benchmark actually tested that claim.
CodeSOTA, an AI benchmark registry, tested RTF and eight other prompting frameworks against a plain, clearly written prompt with no framework at all. The plain prompt scored 97 percent accuracy using 93 tokens. RTF scored 94 percent accuracy while using 119 tokens, meaning it cost more and delivered less.
RTF was not the worst performer in that test. STAR and TAG lost 17 to 19 percent accuracy by comparison. But RTF did not beat a prompt that simply states what you want, either. The benchmark's own takeaway was that the real prompting skill is not knowing acronyms, but communicating clearly.
That does not make RTF useless. It makes it a scaffold, something to lean on while you are still learning what a good prompt looks like, not a guarantee of a better answer.
When RTF Is the Right Call
RTF earns its keep on a specific kind of task: quick, low stakes, one shot, where you would rather not write four paragraphs of context first.
Good fits include:
- Drafting a first version of an email or social post
- Getting a quick comparison table
- Turning a rough idea into a structured checklist
- Any one-off request you will not reuse
It is the fastest framework to write, and that speed is the entire point. You are trading depth for turnaround time.
Where RTF Falls Short
RTF has no slot for context or audience, and that is the gap that shows up most often in practice. It does not ask who the answer is for, what the AI already knows about the situation, or what constraints matter beyond the output shape.
One widely shared breakdown of prompting frameworks makes the same point about RTF directly: it is simple, but it does not give you enough control for precise, high-stakes work.
That gap shows up fast on anything with real consequences. Ask RTF to write a job rejection email and you will get something generic, because nothing in the framework asks about tone or company policy. For tasks like that, frameworks with a context or audience slot, such as RACE (Role, Action, Context, Expectation) or RISEN (Role, Instructions, Steps, End goal, Narrowing), tend to produce more usable first drafts.
Lu's Final Thoughts
RTF earns its spot as the framework you learn first, not the one you lean on forever. It pulls you out of vague, one-line prompts fast, and for quick, low-stakes tasks, that is often all you need.
Once a task actually matters, add back the context and audience RTF leaves out. Reach for RACE or RISEN instead, or just write out what you would tell a smart coworker who knows nothing about the situation. Based on the benchmark above, that plain, specific instruction may beat any acronym you memorize.
RTF Prompt Framework FAQs
Does the RTF framework actually improve AI responses?
Not reliably. One benchmark found RTF scored 94 percent accuracy against a 97 percent baseline for a plain, clearly written prompt, while using more tokens to get there. RTF still helps structure a prompt, but it is not proven to outperform simply stating your request clearly.
Who created the RTF prompt framework?
No single person or organization is credited as its creator. Unlike RISEN or CLEAR, which trace back to named individuals, RTF appears across newsletters and courses starting in 2023 with no consistent attribution.
What is the difference between RTF and RISEN?
RTF covers Role, Task, and Format. RISEN adds Instructions, Steps, and an End goal on top of a role, giving the AI more to work with on complex or multi-part tasks. RTF is faster to write. RISEN gives you more control over the result.
When should I use a different prompt framework instead of RTF?
Switch frameworks once context, audience, or tone genuinely matter, such as client work, sensitive topics, or anything with real consequences if the AI misreads the situation. RACE and RISEN both add a context slot that RTF leaves out entirely.
Can I combine RTF with other prompting techniques?
Yes. Adding "think step by step" to an RTF prompt for a reasoning or math task, or asking for examples of the output style you want, works well alongside the Role, Task, Format structure without replacing it.