PRIME Framework
PRIME Framework
Sandeep Swadia's 5-element prompting rubric, from How To Use Claude Better Than 99% Of People (theMITmonk). The "full" version of his earlier AIM Protocol — a richer scaffold for setting up a high-quality conversation with any chatbot.
"Even with the full team of agents working for you, Claude is only as good as the input you give it. But the super users who get exceptional output from Claude, they direct Claude the way they would direct a smart person they've just hired. And the easiest way to do that is a framework I call prime."
The five elements
| Letter | Element | Purpose |
|---|---|---|
| P | Purpose | Precise goal. "Help me turn this messy proposal into a five-part organized document that looks like a sharp client memo." |
| R | Research | External research to ground the response or raw material from the user (notes, transcripts, files). "Anything that Claude needs so that it can stop hallucinating or guessing." |
| I | Interview | Let Claude ask clarifying multiple-choice questions before answering. "Based on what you clicked, it refines its approach." |
| M | Mechanics | Output shape: bullets vs paragraph, document vs table, concise vs detailed, strategic vs conversational |
| E | Examples | "Show Claude what good looks like." Past output, tone reference, an outline that worked. |
The distinctive piece: Interview
Most prompt frameworks (Sandeep's own AIM Protocol, standard prompt-engineering hierarchies in Intelligent Hill (Prompting Camps)) skip the elicitation step. PRIME names it explicitly:
"This is the hidden gem. I really like how Claude does it. In a lot of cases, you'll see that it will start asking you multiple-choice questions and you pick the right choice and based on what you clicked, it refines its approach to a specific way of responding to you."
That's the user becoming the interviewee instead of the interviewer. Compresses the "you don't know what you don't know" prompt-engineering gap by letting the model surface what's ambiguous.
Worked example — presentation prep
(From the video, slightly tightened.)
| Element | What the user provides |
|---|---|
| Purpose | "I have a meeting tomorrow. What's the outcome — approval, alignment, resources?" |
| Research | Audience, risks, data, internal docs, external resources |
| Interview | "Interview me before you respond" — Claude asks questions that sharpen scope and the user's own thinking |
| Mechanics | "10-slide deck plus talking points" |
| Examples | A past deck the user liked, or an admired style reference |
How it sits among Sandeep's other prompting frameworks
| Framework | Scope | Best for |
|---|---|---|
| AIM Protocol (Actor / Input / Mission) | 3 slots, fastest | Quick drafting prompts |
| PRIME | 5 slots, richer | Real deliverables; high-stakes single-shot |
| Intelligent Hill (Prompting Camps) | Hierarchy of techniques | Knowing which prompting style to climb to |
| DRAG Framework | Delegation decision | Knowing whether to delegate to AI at all |
PRIME is the content of a good prompt; the Intelligent Hill is the technique; DRAG is the gate.
Why it matters to this vault
- First 5-element prompting rubric in the vault. AIM and Intelligent Hill are both partial; PRIME is the most complete "set up a chat well" framework here.
- Interview step is genuinely novel for the vault. Pairs with Productive Resistance from the user side — instead of waiting for the AI product to ask clarifying questions by default (Gedeon), the user prompts the model to ask them. Same friction-in-the-right-place principle, opposite trigger.