← Back to blog

RTF, CARE, CREATE, and Other Starters

RTF 🟢

Need: Absolute start — the simplest framework.

Explanation: Role → Task → Format. Three elements, zero complications.

“Act as [role]. Perform [task]. Return as [format].”

QA: “Act as a tester. Write 5 test cases for login. Return as a table with columns: ID, Description, Steps, Expected Result.”


CARE 🟢

Need: A specific task aimed at a measurable result.

Explanation: Context → Action → Result → Example.

“Context: [situation]. Action: [what to do]. Result: [goal]. Example: [template].”

QA: “Context: new sprint, 15 user stories. Action: generate a smoke test suite. Result: 1 test per story, max 3 steps. Example: US-101 → ‘Verify that the dashboard loads in <3s’.”


CREATE 🟡

Need: Advanced copywriting, highly personalized content.

Explanation: Character → Request → Explanation → Adjustment (tone) → Type of output → Extras.

“Take on the role of [character]. Task: [request]. Background: [context]. Tone: [which]. Type: [format]. Remember to: [extras].”

QA: “Take on the role of a QA manager. Write a sprint quality summary. Background: 3 critical bugs in production. Tone: factual, no panic. Type: email to stakeholders. Remember to: add action items.”


Zero-Shot 🟢

Need: Quick use without examples.

Explanation: Instruction without examples. The model performs the task based solely on its training.

“[Direct instruction]”

QA: “Write tests for the login API. Include both positive and negative scenarios.”


Role Prompting 🟢

Need: Control the style and expertise level of the response.

Explanation: Assigning the model an expert role — changes the “personality” and depth of the response.

“You are a [role with X years of experience]…”

QA: “You are a senior QA with 10 years of experience in fintech. Propose a testing strategy for the AML module.”


Skeleton-of-Thought 🟡

Need: You need a structured response — first a plan, then expansion.

Explanation: The model generates a skeleton (outline), then expands each point.

“First list the response plan (headings only). Then expand each point.”

QA: “Write a test plan. First list the sections (scope, approach, risks, schedule). Then expand each one.”


Full framework map

LevelFrameworks
🟢 EasyRTF, CARE, Zero-Shot, Role Prompting, ARC, ICE, TDS, Few-Shot, CoT
🟡 MediumCRISPE, CO-STAR, SCQA, RISEN, CREATE, Self-Consistency, Meta Prompting, Least-to-Most, Skeleton-of-Thought
🔴 AdvancedReAct+, ToT, P2P2, Reflexion, PAL

In the next post: AI Tools Review — ChatGPT, Claude, Copilot, Codex, Rovo, Devin, Gemini, and OpenClaw.