Prompt Frameworks (2026): RTF, CRISPE, RICE & CO-STAR
A prompt framework is a reusable structure that makes sure every prompt includes the parts that matter, role, task, context, format, and constraints. Learn RTF first (it covers most cases), then add CO-STAR for tone-sensitive content and RICE when you need the model to match a style. Here is each framework, when to use it, and a copy-paste example.
By GPTPrompts.AI · Last updated June 17, 2026
The major prompt frameworks, with examples
RTF
Role · Task · FormatBest for: The fastest everyday framework, quick, well-structured outputs.
CO-STAR
Context · Objective · Style · Tone · Audience · ResponseBest for: High-stakes content where tone and audience matter (marketing, comms).
CRISPE
Capacity/Role · Insight · Statement · Personality · ExperimentBest for: Exploratory work where you want multiple takes to compare.
RICE
Role · Instructions · Context · ExamplesBest for: When output quality depends on showing the model what 'good' looks like.
RACE
Role · Action · Context · ExpectationBest for: Action-oriented tasks with a clear deliverable.
APE
Action · Purpose · ExpectationBest for: The minimalist framework for simple, one-off requests.
Which framework should you use?
Start with RTF. Role, Task, Format covers the majority of everyday prompts and takes seconds to write. If you learn only one, learn this.
Use CO-STAR for content where tone matters, marketing copy, emails, social posts, anything audience-facing. Its explicit Style, Tone, and Audience fields are what make output feel on-brand.
Use RICE when quality depends on matching a style, paste 1-2 examples of what 'good' looks like and the model will follow them. Examples are the single most powerful lever in prompting.
Use CRISPE to brainstorm when you want several distinct options to compare rather than one answer. The real skill underneath all of them is the same, see what prompt engineering is and how to write effective AI prompts.
Frequently asked questions
What is the best prompt framework in 2026?
There is no single best framework, the best one depends on the task. RTF (Role, Task, Format) is the fastest and works for most everyday prompts. CO-STAR (Context, Objective, Style, Tone, Audience, Response) is best for marketing and communications where tone and audience matter. CRISPE suits exploratory work where you want multiple options. RICE adds examples for quality-sensitive tasks. The practical advice: learn RTF first because it covers 80% of cases, then reach for CO-STAR when tone matters and RICE when you need the model to match a specific style. A framework's job is just to make sure you include role, task, context, format, and constraints, any of them gets you there.
What is the RTF prompt framework?
RTF stands for Role, Task, Format. You tell the AI who to be (Role), what to do (Task), and how to return the answer (Format). Example: 'Act as a financial analyst (Role). Summarize this earnings report's key risks (Task). Return a 5-bullet list, most important first (Format).' RTF is the most popular framework because it is fast to write and forces the three things that most improve output: a clear persona, a specific task, and a defined output shape. It is the best framework to learn first.
What is the CO-STAR prompt framework?
CO-STAR stands for Context, Objective, Style, Tone, Audience, and Response format. It was popularized for content and marketing prompts because it explicitly separates tone and audience from the task. You give background (Context), the goal (Objective), the writing Style and Tone, who it's for (Audience), and the desired Response format. It produces more on-brand, audience-appropriate output than a bare instruction, which is why it shines for copy, emails, and social content. Use it when how the message lands matters as much as what it says.
What is the difference between RTF, CRISPE, and RICE?
They emphasize different things. RTF (Role, Task, Format) is the minimal, fast framework for clear everyday output. CRISPE (Capacity/Role, Insight, Statement, Personality, Experiment) is built for exploration, it asks for a personality and an 'experiment,' so you get varied options to compare. RICE (Role, Instructions, Context, Examples) centers on examples, which is the single most powerful lever when you need the model to match a specific style or format. In short: RTF for speed, CRISPE for ideation, RICE for quality matching. All three are just structured ways to include role, task, context, and constraints.
Do I need a framework to write good prompts?
No, but frameworks help you remember the parts that matter. A great prompt includes a role or perspective, a specific task, relevant context, the output format, and any constraints, frameworks like RTF and CO-STAR are just checklists that make sure you don't forget one. Once you've written a few hundred prompts, you internalize the pattern and stop needing the acronym. Until then, picking one framework and using it consistently is the fastest way to level up. For the underlying skill, see our guide on writing effective AI prompts.
Do these frameworks work for Claude and Gemini, not just ChatGPT?
Yes. Prompt frameworks are model-agnostic because they structure your intent, not any model-specific syntax. RTF, CO-STAR, CRISPE, and RICE all work on ChatGPT, Claude, and Gemini. Minor tuning helps: Claude responds especially well to explicit numbered instructions (a natural fit for RTF and RICE), Gemini benefits from clear context for grounding, and ChatGPT handles all of them comfortably. The framework gets you 90% of the way on any model; the last 10% is adapting tone and iterating.