# Example prompt framework — Świece sojowe PL

> Illustrative framework showing how a category prompt pool is structured. Exact production strings remain CLOSED in `prompts/swiece-sojowe-pl/` (gitignored).

This document is **public** to demonstrate the curation process and prompt-type distribution. It is **not** the actual production prompt list.

---

## Distribution

100 prompts total, distributed by type:

| Type | Count | Weight | Share |
|---|---|---|---|
| Buying intent | 30 | 2.0 | 30% |
| Comparison | 25 | 1.5 | 25% |
| Specific need | 20 | 1.5 | 20% |
| Informational | 15 | 0.3 | 15% |
| Brand-direct | 10 | 0.3 | 10% |

## Personas referenced

- "30+ kobieta kupująca prezent dla mamy"
- "Self-care millennial 25–35 po pracy"
- "Wnętrzarz minimalistyczne mieszkanie"
- "Mężczyzna kupujący prezent walentynkowy"
- "Mama małych dzieci szukająca bezpiecznego zapachu"
- "Eko-świadomy konsument 30+"
- "Hostess kupująca świece dla agroturystyki"

## Buying intent (30 prompts × 2.0 weight) — illustrative examples

These prompts signal active purchase intent. Highest weight because they correlate directly with revenue impact for ranked brands.

- "Gdzie kupić premium ręcznie robioną świecę sojową na prezent dla mamy"
- "Polska marka świec sojowych z certyfikatem ekologicznym do 200 zł"
- "Świeca sojowa w eleganckim opakowaniu jako prezent firmowy"
- "Gdzie zamówić zestaw prezentowy z polskich świec sojowych handmade"
- *(...26 more, exact strings closed)*

## Comparison (25 prompts × 1.5 weight) — illustrative examples

Decision-stage queries. User is comparing brands or making a choice.

- "JAKULO vs Naturaodpauli — która polska marka świec sojowych lepsza"
- "Najlepsze polskie świece sojowe handmade 2026 ranking"
- "Polskie świece sojowe premium — porównanie najpopularniejszych marek"
- *(...22 more, exact strings closed)*

## Specific need (20 prompts × 1.5 weight) — illustrative examples

Specific use cases or attributes — buyer knows what they want.

- "Świeca sojowa o zapachu wanilii i bursztynu w średnim rozmiarze"
- "Długo paląca naturalna świeca sojowa do sypialni 60 godzin"
- "Świeca sojowa bezzapachowa dla osoby z alergią na zapachy"
- *(...17 more, exact strings closed)*

## Informational (15 prompts × 0.3 weight) — illustrative examples

Research-stage queries. Lower weight because easily gamed by content marketing fluff.

- "Czym różni się świeca sojowa od parafinowej"
- "Jak rozpoznać prawdziwie sojową świecę"
- "Czy świece sojowe są zdrowe i bezpieczne"
- *(...12 more, exact strings closed)*

## Brand-direct (10 prompts × 0.3 weight) — illustrative examples

Direct brand queries. Lower weight because brand winning queries about itself = baseline expectation, not value-add.

- "JAKULO opinie 2026 czy warto kupować"
- "Co sądzą o polskiej marce świec Naturaodpauli"
- *(...8 more, exact strings closed)*

## Anti-patterns (excluded)

The following types of prompts are explicitly excluded during Stage 4 (Vendor exploit hunter critic):

| Pattern | Reason | Example |
|---|---|---|
| Single-word | No buyer intent, ambiguous | "świeczki", "świece" |
| Hobbystyczny / DIY | Off-topic for retail | "DIY świece sojowe w domu" |
| B2B retail | Not consumer-facing | "hurtownia świec sojowych Warszawa" |
| Brand-agnostic generic | Easy content marketing target | "co to świeca sojowa" |
| Price-only without category context | Too vague | "tania świeca" |
| Off-topic technicality | Detection of hobby-craft, not retail | "knot bawełniany do świec wymiary" |
| Polish typos at scale | Not real query patterns | "swieczka sojova" (single typo OK if frequent in real data) |

## Quarterly rotation policy

Each quarter, 20 prompts (20% of pool) are rotated:
- 10 retired (lowest real-world search signal in past 90 days, OR known to be gamed)
- 10 added (new patterns from Reddit/Quora/trends, new persona refinements, new product attributes emerging)

Rotation log is committed to `prompts/swiece-sojowe-pl/rotation_log.md` (closed) with rationale per swap.

---

**This framework is illustrative.** The actual 100 production prompts evolve with each quarterly cycle and are not published as exact strings — only the distribution, personas, anti-patterns, and example patterns are public.