1. Write a short style guide
Brand voice: warm, practical, no hype words ("revolutionary", "game-changing").
Structure: 1 benefit-led opening sentence, then 3 bullet points of key features, then 1 sentence on who it's for.
Length: 60-90 words. American English. Never invent specs that aren't provided.“Never invent specs” is the most important line: customers rely on these details.
2. Process a CSV
import csv
def describe(row):
specs = "\n".join(f"{k}: {v}" for k, v in row.items() if v)
r = client.chat.completions.create(model="gpt-5.6-luna", temperature=0.7, messages=[
{"role": "system", "content": STYLE},
{"role": "user", "content": f"Write the product description.\n\n{specs}"},
])
return r.choices[0].message.content.strip()
with open("products.csv") as f, open("products_out.csv", "w", newline="") as out:
rows = list(csv.DictReader(f))
w = csv.DictWriter(out, fieldnames=list(rows[0].keys()) + ["description"])
w.writeheader()
for row in rows:
w.writerow({**row, "description": describe(row)})3. Review before publishing
Spot-check a sample, especially for numbers and materials. Add a column “approved” and only import approved rows.
Cost
Product descriptions are short, so even a strong model costs little per item. Run 10 products with two different models, compare, and pick the cheaper one if quality is similar.
Claude, GPT, Gemini, DeepSeek, image and video models through one OpenAI- and Anthropic-compatible endpoint. Prepaid, pay only for what you use.