# AI Prompt Framework

*This framework was generated using the Framework Builder methodology. It demonstrates the five-layer DNA architecture applied to AI prompting for a content team.*

Page: https://whereframeworks.com/frameworks/ai-prompt
Guide: https://howtoframework.com/ai-prompt-framework

---

## Layer 1: Principles Foundation

**Principle 1: Audience context comes before the instruction.** The model cannot write for your reader if it does not know who the reader is.

**Principle 2: Voice is a constraint, not a suggestion.** Each template carries concrete parameters: sentence length range, vocabulary level, tone descriptors and a short example paragraph.

**Principle 3: AI makes the first draft, a human owns the final.** Optimize for speed to a usable draft, not for something ready to publish.

**Principle 4: Build on communication principles, not model quirks.** Context, constraints, format and examples work across versions. A trick tied to one model version breaks when it updates.

**Principle 5: Share the reasoning with the template.** A prompt nobody can adapt is brittle. Each template says why it is built the way it is.

## Layer 2: Systematic Approach

**Step 1: List the recurring tasks.** Blog drafts, social copy, email sequences, briefs.

**Step 2: Route by content type.** Long-form content uses two steps: a structured outline prompt, then section-by-section drafting. Short-form content uses a one-shot template with platform constraints such as length limits and call-to-action format. Sequences use an iterative approach where each prompt includes the earlier messages as context.

**Step 3: Write the template in three parts.** Audience and brand context at the top, the instruction in the middle, output format and constraints at the bottom.

**Step 4: Score each output against a rubric.** Tone match, required elements, length and accuracy.

**Step 5: Revise the template from the scores.** Change the template, not the individual prompt.

## Layer 3: Force Multipliers

**Multiplier 1: A shared template library.** A dozen templates covering the most common tasks, stored where everyone can reach them.

**Multiplier 2: A reference-example field.** Paste a paragraph from a high-performing piece into each template. It often improves first-draft quality more than any other single change.

**Multiplier 3: Templates as the default.** Starting from scratch needs a one-line justification in the team channel.

**Multiplier 4: Keep prompts short and specific.** A 200-word prompt with clear constraints usually beats a 1,000-word one with elaborate role-play.

**Multiplier 5: Blind-rate outputs across people.** Have the team lead rate drafts from different writers without names. The spread between the best and worst writer shows whether the templates are carrying the quality or the individual is.

## Layer 4: Success Metrics

**Positive indicators:**
- First-attempt quality rate rises, where quality means a draft needing under fifteen minutes of editing
- Time to a first draft falls
- Blind-rated outputs from different people cluster closer together
- New team members reach good output in days, not weeks

**Failure signals:**
- Quality is judged by gut feel
- Everyone keeps their own private prompts
- Prompts grow longer and results do not improve
- Templates exist with no explanation of their design
- Output quality drops after a model update

## Layer 5: Implementation Guidance

**Start with one content type.** Week one, the strongest writer builds the first two templates. Week two, everyone uses them and scores the outputs. Week three, revise from the scores and add the reference field. Week four, extend to the next content type. A full rollout across four types takes about six weeks.

**Make the template the default from day one.** Otherwise it becomes shelfware.

**Common mistake: treating each prompt as a fresh start.** Capture every good prompt, annotate it and add it to the library.

**Edge case: a task with no template.** Write it from scratch once, score it, and add it if it works.

**Edge case: the model changes.** Rerun a fixed set of test prompts and rubric scores before updating templates. Revise only where scores drop.

**Common mistake: over-engineering prompts.** A short prompt with clear constraints often beats a long one built on role-play and multi-step reasoning chains. Complexity adds failure points, not quality.

**Common mistake: judging output by gut feel.** "This seems good" cannot be improved on. Score against criteria: tone match, required elements, length and accuracy.

**Common mistake: sharing prompts without the reasoning.** Someone facing a task that does not quite fit the template cannot adapt it without knowing why it is built that way. Put a short design note on every template.

**Common mistake: building around one model version.** Frameworks tied to a version's quirks break at the next update. Build on context, constraints, format and examples.

**Common mistake: no owner for the library.** Templates decay when no one maintains them. Name one person to review scores weekly, retire weak templates and add new ones that pass the rubric.

**Edge case: junior staff getting worse results.** Check the template and the reference example before the person. If the senior writer's output is consistently good, capture their exact context block and reference paragraphs in the template.
