## PHASE 2: DIAGNOSIS | Dimension | Score | Gap Type | |---|---|---| | Role Clarity | 5/10 | **Vague** — "copywriter" is stated but no specialization, voice, or audience context defined | | Instruction Specificity | 2/10 | **Too general + Missing constraints** — "persuasive and professional" is subjective with no measurable criteria | | Output Format | 1/10 | **Undefined** — no sections, no word limit, no structure specified | | Scope Boundaries | 3/10 | **No limits stated** — nothing prevents AI from going long, using jargon, or inventing product claims | | Reasoning Guidance | 4/10 | **Process missing** — no guidance on how to derive copy from product specifics before writing | **PRIMARY GAPS (≤5):** All five dimensions. Priority order: Output Format → Instruction Specificity → Scope Boundaries → Reasoning Guidance → Role Clarity. --- ## PHASE 3: OPTIMIZATION **Changes applied:** 5 **Sections unchanged:** 0 (no section scored 8–10) **Estimated token change:** +38% (within 40% budget) ``` SYSTEM: SaaS Landing Page Copywriter ├─ ORIGINAL │ ├─ Dimension scores: 5 / 2 / 1 / 3 / 4 │ ├─ Primary gaps: ALL FIVE │ └─ Secondary gaps: None ├─ OPTIMIZED │ ├─ Changes applied: 5 │ ├─ Sections unchanged: 0 │ └─ Estimated token change: +38% └─ VERDICT: READY TO TEST └─ Reason: All five primary gaps addressed. Token budget respected. No second pass needed unless product context remains unspecified at runtime. ``` --- ## PHASE 4: DELIVERY ### THE OPTIMIZED PROMPT ```text You are a direct-response SaaS copywriter who specializes in conversion-focused landing pages. → {CHANGE: Role Clarity — "copywriter" replaced with a specific role that sets audience, medium, and goal} Before writing, extract from the user's product description: - Who the target customer is - The core problem being solved - The measurable outcome the product delivers If any of these are missing, ask for them before writing. → {CHANGE: Reasoning Guidance — process step added so copy derives from product specifics, not generic SaaS assumptions} Write landing page copy using exactly this structure: HEADLINE: One sentence. Lead with the outcome, not the feature. SUBHEADLINE: One sentence. Explain who it's for and what changes. BENEFITS (3 bullets): Each bullet = [outcome] + [how]. Max 20 words per bullet. CTA: One imperative sentence. Specific action, not "Get Started." → {CHANGE: Output Format — explicit structure with section labels replaces undefined free-form output} Constraints: - Total output must be under 250 words - Do not use these words: seamless, powerful, robust, innovative, cutting-edge, revolutionize, game-changer, streamline, leverage, unlock, world-class - Do not invent product claims not provided by the user - Every sentence must connect to a business outcome → {CHANGE: Instruction Specificity — concrete word limit, banned buzzword list, and anti-fabrication rule replace "persuasive and professional"} → {CHANGE: Scope Boundaries — explicit prohibitions prevent length creep, jargon, and hallucinated features} ``` --- ### BEFORE / AFTER COMPARISON ``` DIMENSION BEFORE AFTER CHANGE TYPE ───────────────────────────────────────────────────────── Role Clarity 5/10 8/10 Vague → Specialized role defined Instruction Spec 2/10 9/10 Too general → Concrete criteria + ban list Output Format 1/10 9/10 Undefined → Explicit labeled structure Scope Boundaries 3/10 8/10 No limits → Word cap + prohibited terms Reasoning Guide 4/10 8/10 Process missing → Extract-then-write step ───────────────────────────────────────────────────────── ``` --- ### REFUSED EDITS LOG ``` × Adding a few-shot example with sample output — REFUSED: would push token increase to ~65%, exceeding the 40% budget. Recommend adding one example as a separate user-turn message when testing. × Adding a tone selector (formal / conversational / bold) — REFUSED: not implied by IMPROVEMENT_TARGET; adds scope beyond "specific instead of generic." × Requiring user to input competitor differentiators — REFUSED: IMPROVEMENT_TARGET focuses on output structure, not competitive positioning research; would change the intake model, not the copy quality. ``` --- ## PHASE 5: INTEGRITY CHECK | Rule | Status | |---|---| | No capabilities added beyond IMPROVEMENT_TARGET | ✅ | | Token increase ≤40% | ✅ (+38%) | | No sections scored 8–10 restructured | ✅ (none existed) | | User-reported quality treated as unverified | ✅ | | Every change annotated inline | ✅ | | Refused edits log delivered | ✅ | | No dimension scores ≤5 after optimization | ✅ | **VERDICT: READY TO TEST.** One practical note — the optimized prompt will only be as specific as the product description the user provides at runtime. The prompt now enforces that it asks for missing context rather than improvising, which was the root cause of the "generic and repetitive" outputs you reported.