What this dataset measures
The unit is a derived copy pattern, not a whole sales script. Each row labels what job the pattern performs, where it appears in the creative, which emotional tones it uses and how confident the extraction pass was. That makes the sample useful for structure analysis, taxonomy design, creative research and hypothesis generation without turning it into a verbatim swipe file.
Preview the actual fields
| Type | Position | Canonical pattern | Confidence |
|---|---|---|---|
| authority | problem agitation | Speaker has spoken at international medical congresses, contributed to scientific journals, and consulted for global health initiatives | 0.90 |
| avatar | problem agitation | Someone suffering from diabetes who wants to eat favorite foods, share family meals, and live without blood sugar anxiety | 0.84 |
| cta | problem agitation | Click the green button now to secure the package before time runs out | 0.97 |
| hook | problem agitation | Universal exposure: nearly everyone who eats common foods has already been affected by metabolic biofilm | 0.89 |
| mechanism | problem agitation | Desert-grown extract formulated to dissolve metabolic biofilm and restore pancreatic function | 0.95 |
| pain | opening | Blood sugar spiraling out of control despite healthy eating and medication adherence | 0.98 |
| promise | problem agitation | Comprehensive support to eliminate metabolic biofilm, repair pancreas, and maintain optimal blood sugar levels | 0.95 |
| social_proof | opening | Researchers from Imperial College London and University of Columbia Cairo have confirmed the metabolic biofilm discovery | 0.88 |
| tactic | problem agitation | Contrast between forbidden foods in Western medicine and thriving elders consuming those same foods | 0.89 |
| urgency | problem agitation | Time-limited stock with color-coded button that turns red when supply is exhausted | 0.95 |
How rows qualify
A source transcript must be active, English, reviewed, fully chunked and fully embedded. Its extraction pass must contain at least eight units, with at least 75% scoring 0.80 or higher. The export then rejects near-verbatim source language and deduplicates matching canonical patterns inside each niche and pattern type.
Counts describe observed corpus structure, not proof that a phrase causes conversion, and they do not validate medical or commercial claims made inside an advertisement.
Need the full cross-niche edition?
The commercial release combines every mature niche in one consistent CSV/JSONL schema, includes checksums and a data dictionary, and remains de-identified.
View the full dataset