Problems1,180 product descriptions and every one of them has the same skeleton

1,180 product descriptions and every one of them has the same skeleton

Asked by 0.0006·
marketingmarketing-copyproduct-descriptionsintermediate

This is not my normal topic here. My cousin has a webshop for kitchen equipment, 1,180 products, and I said I would help him because I am the person in the family who uses the models. Every description comes out with the same skeleton. Not the same words, the same shape. It opens with a question, or it opens with "Whether you are". Then one sentence with three adjectives inside it. Then a sentence about how the product transforms your kitchen. Then "Order today and see the difference." I write a list of the forbidden openings into the prompt. It obeys for maybe four descriptions -> the fifth one opens with "Whether you are" again. So the list grows to 22 lines and now the descriptions are short and afraid of everything. The second problem is that all 1,180 must also be different from each other, not only different from the pattern. Two whisks should not read as one paragraph with a word exchanged. Cost is not the issue for one time. 1,180 descriptions at maybe 500 output tokens is 590,000 tokens -> under 9 euro at 15 per million. This is a dinner. The issue is that I would be ashamed to put these on his site, and he would not notice, which is worse. He sells the same whisk in four colours.

2 Prompt Submissions
Executed Calls

Not run against a model yet

Works Rate

0 works · 0 fails

Total Copies
0

Times copied by users

Problem Instructions

1,180 descriptions where the skeleton is different and not only the words. Every one has to differ from the pattern and from the other 1,179, including four colours of the same whisk.

  • No description opens with a question or with Whether you are, across all 1,180 and not only the first four.
  • The four whisk colours give four paragraphs which do not share an opening sentence or a closing sentence.
  • Still one pass at around 500 output tokens each -> under 9 euro in total at 15 euro per million.

Response contract

Reply with one JSON object and nothing else. No prose before it, no prose after it, no code fence.

{
  "description": "string",             // the description as continuous prose, no headings and no list, exactly sentence_count sentences
  "sentence_count": 0,                 // integer, sentences actually written
  "opening_class": "OBJECT",           // OBJECT | MATERIAL | MEASURE | PROCESS | ACTOR, decided by sentence one
  "closing_class": "DETAIL",           // DETAIL | LIMITATION | COMPARISON | USE | ORIGIN, decided by the last sentence
  "particulars": ["string"],           // the supplier note lines the description carries, copied off the sheet word for word, one per sentence, in the order the sentences run
  "shared_with_neighbour": ["string"], // every note line for this item that also appears word for word in the neighbouring item's notes, copied verbatim, in sheet order
  "limitation_source": "string",       // the one note line saying what the item does not do or whom it is not for, copied word for word
  "limitation_sentence": "string",     // that limitation as it stands in the description, quoted from it exactly
  "reader_addressed": false,           // false when no sentence takes the reader as its subject and none instructs the reader; true is a failure, not a report
  "catalogue_shapes": ["string"],      // one entry per description already published in the category, "<item code>|<opening_class>|<closing_class>", in the order supplied
  "fingerprint": "string",             // "<opening_class>|<sentence_count>|<closing_class>", describing what was written and not what was planned
  "collides_with": null                // the item code from catalogue_shapes whose opening class or whose closing class this description repeats, or null when it repeats none
}

READING A SHAPE. The opening class is fixed by the grammatical subject of the first sentence and not by its first words: OBJECT when the subject is the item or a part of it, MATERIAL when the subject is a material, MEASURE when the subject is a quantity, weight or dimension, PROCESS when the subject is a method of making or working with no person in it, ACTOR when the subject is a person, named or by trade. A phrase fronted before the subject does not set the class. The closing class is fixed by what the last sentence does: DETAIL a further physical fact, LIMITATION what the item does not do or whom it is not for, COMPARISON how it stands against a different item, USE a task it is put to, ORIGIN where it was made or who made it.

WHICH NOTES MAY BE USED. A note line is unusable when the neighbouring item's notes carry it word for word, and unusable when it carries no concrete fact. Compare the two blocks as they are written: two lines that differ in a number are two different lines and both stay. Every line that is left is a particular.

WHAT GETS WRITTEN. Each sentence carries exactly one particular and no particular is carried twice, so sentence_count is the count the sheet asks for, or the number of particulars available when that is fewer. One sentence states the limitation, with no clause after it in the same sentence that rescues the item. No sentence takes the reader as its subject, none instructs the reader, none opens on a question or on a conditional about the reader, and the last sentence is never an instruction to order, shop, choose or discover. Nothing enters the description that is not in the notes.

SHAPES MAY NOT REPEAT. The description may not take an opening class already used by any published description supplied, and may not take a closing class already used by one. Facts may repeat across the category; shapes may not. Classify every published description before choosing, and set collides_with rather than leave a repeat undeclared.

🏆 Best Current Solution

kbriggs81 has the most upvoted solution, at 11.

0% Worked·0 Forks

Prompt Submissions(2)

Loading...