RT-CROS is a six-part prompting framework - Role, Task, Context, Reasoning, Output Format, Stop Conditions - that turns vague ChatGPT requests into consistent, high-quality answers. We have tested prompt structures across 300,000+ readers at AI Central, and the pattern that holds up is always the same: the more of these six slots you fill, the less the model has to guess. This guide is pulled from the AI Central Library, where the full tutorial set lives.
What is the RT-CROS prompting framework?
RT-CROS is a checklist for writing a prompt. Each letter is one thing the model needs to know before it can answer well:
R - Role: the persona ChatGPT should adopt
T - Task: exactly what you want done
C - Context: background, constraints, and details
R - Reasoning: how the model should think before answering
O - Output Format: the exact shape of the response
S - Stop Conditions: what “done” looks like
Most weak prompts fail because they only cover the T. The other five slots are where the quality comes from. If you want the theory behind why this works, the 26 principles of prompt engineering cover the same ground from a research angle.
1. Role
Define a clear persona for ChatGPT to adopt. A role narrows the model’s vocabulary, assumptions, and standard of evidence before it writes a single word.
Act as a personal productivity coach specializing in effective, lesser-known learning methods for mastering a new skill in three months
2. Task
State exactly what you want the AI to do. Be explicit. Vague verbs like “help me with” or “look at” give the model permission to pick its own job.
Begin with a concise checklist (3-7 bullets) of steps to follow, focusing on conceptual planning. Identify and present the top 3 medium-commitment learning methods (not widely used) that enable strong progress in under 90 days. Ensure each method offers a unique advantage in efficiency, engagement, or adaptability.
3. Context
Provide all necessary background, constraints, and details for accurate responses. This is the slot people skip most often, and it is the one that prevents generic output.
Prioritize accuracy: method names must match official or widely recognized sources. Time and resource estimates should be realistic. Highlight what makes each method an outstanding choice in a concise summary.
4. Reasoning
Guide the AI’s thought process before it delivers the final answer. You are telling the model what to check, not just what to produce.
Internally vet all methods to ensure they are real, underused, and meet all parameters. Cross-check details with credible learning or productivity sources. Optimize for clarity, concise presentation, and practical value.
5. Output format
Specify exactly how you want the response structured. Naming the format is the fastest way to make an answer usable rather than merely correct.
Return results as a Markdown table with the following columns: Method Name, Main Resources, Weekly Time Commitment (Hours), Estimated Progress in 90 Days, Summary.
6. Stop conditions
Set clear boundaries for completion. Without them the model either stops early or keeps padding.
Task is complete when three verified, unique medium-commitment methods are returned in the specified table format, excluding overly common approaches, with full compliance to all requirements.
The complete RT-CROS prompt example
Here is the whole framework assembled into one prompt you can copy and adapt:
Role: Act as a personal productivity coach focused on recommending lesser-known, effective learning methods for mastering a new skill within three months.
Task: Begin with a concise checklist (3-7 bullets) of conceptual planning steps. Identify and present the top 3 medium-commitment, underused learning methods for strong progress in under 90 days. Ensure each method offers a unique advantage.
Context: Prioritize accuracy and realism in method names and time estimates. Highlight what makes each method outstanding in a brief summary.
Reasoning: Vet methods for authenticity and relevance. Cross-check with credible sources. Optimize for clarity and practicality.
Output Format: Markdown table with Method Name, Main Resources, Weekly Time, Estimated 90-Day Progress, Summary.
Stop Condition: Task complete when three verified methods are returned in the specified format, excluding common approaches, with full requirement compliance.
Swap the subject matter and the six slots stay identical. That is the point of a framework: you rewrite the content, never the structure. If your prompts still come back flat after this, the ChatGPT prompt optimizer walks through diagnosing and fixing a weak prompt line by line.
Frequently asked questions
What does RT-CROS stand for?
Role, Task, Context, Reasoning, Output Format, and Stop Conditions. Each letter is a slot you fill in before sending a prompt to ChatGPT.
Do I need to use all six parts every time?
No. Short factual questions only need the Task. Use the full framework when the output matters, when the answer has a required shape, or when you plan to reuse the prompt.
Does RT-CROS work with Claude and Gemini?
Yes. The framework is model-agnostic because it describes what any language model needs to know, not a ChatGPT-specific syntax. It works the same way in Claude, Gemini, and Copilot.
What is the difference between Reasoning and Task?
Task is what you want produced. Reasoning is how the model should get there - what to verify, cross-check, or weigh before it commits to an answer.
Why do stop conditions matter?
They define “done” in measurable terms, which stops the model from padding a short answer or cutting off a long one. They also make outputs consistent when you run the same prompt repeatedly.
