The cost of doing nothing
James had been running his skincare and haircare brand for three years. The products were good — carefully formulated, well-reviewed, genuinely differentiated. But if you landed on one of his product pages, you'd see something like this:
"A deeply nourishing facial oil with rosehip, sea buckthorn, and vitamin E. Apply 2–3 drops to cleansed skin."
That was the supplier's copy. It was on his site. It was probably on a dozen others too.
He knew it was a problem. Generic descriptions meant weaker SEO — Google treats near-duplicate content as thin, giving it less visibility. They also meant lower conversion: shoppers who land expecting to be convinced, and leave because nothing spoke to them.
His catalogue had grown to 800 SKUs. Hiring a copywriter to do each one was expensive and slow. A freelance agency quoted him £3,500 and a six-week timeline. A junior freelancer was cheaper but still more time and budget than he could justify for a task that felt, at its core, mechanical.
What he actually needed
James didn't need beautiful prose. He needed descriptions that:
- Were unique to his brand, not a supplier template
- Highlighted the active ingredients and explained their benefits clearly
- Matched his brand voice — warm, informed, not clinical
- Were long enough for Google to work with (150–200 words per page)
That's a well-defined task. It applies the same way to every row in a spreadsheet. It was a good candidate for batch processing.
Setting up the CSV
His product data was already in a spreadsheet — he'd been maintaining it for inventory and Shopify imports. He added three columns he didn't have:
| product_name | category | key_ingredients | brand_claims |
|---|---|---|---|
| Radiance Facial Oil | Skincare / Face | Rosehip, Sea Buckthorn, Vitamin E | Brightening, Anti-ageing, Dry skin |
| Curl Define Cream | Haircare / Styling | Shea Butter, Argan Oil, Aloe Vera | Frizz control, Definition, Moisture |
Some products had sparse data. He spent about two hours filling gaps for the first 200 rows, then decided the rest were good enough to run — he'd fix the edge cases manually after reviewing the output.
The batch instructions
Written once, applied to all 800 rows:
James's batch instructions:
You are a copywriter for Luminos, a premium UK beauty brand. Write a product description for this item.
Tone: warm, knowledgeable, and reassuring — never clinical or salesy. Speak directly to the customer.
Format:
— Opening sentence: 1–2 sentences on what the product does and who it's for
— Ingredients paragraph: name the key ingredients and explain what each one does for the skin or hair
— Closing sentence: when/how to use it, or a gentle encouragement
Length: 160–200 words. No bullet points. Output the description only — no title, no extra commentary.
He ran a test batch of 30 rows on Gemini 2.5 Pro first. The output was solid — distinct, on-brand, ready to publish on most of them. He made one adjustment to his batch instructions (adding a note to avoid the phrase "skin-loving") and ran the full 800.
The job completed in just under three hours.
The result
James reviewed all the output. He found around 60 descriptions that needed meaningful edits — mostly products where he'd entered minimal ingredient data. The other 740 were publish-ready or needed only minor wording tweaks.
Total time on his side: about five hours across two days. Data prep, testing, reviewing output, uploading to Shopify. The cost in PromptMax credits was a few pounds — well under £10 for 800 descriptions on Gemini 2.5 Pro.
Against the agency quote of £3,500 and six weeks, the numbers are difficult to argue with.
Four things worth knowing before you try this
Your output quality depends on your input quality. Sparse ingredient data will produce generic descriptions regardless of how good your batch instructions are. The time you invest in the CSV pays back directly in the output.
Use Gemini 2.5 Pro for creative work. Flash is efficient for structured tasks like meta tags. Product descriptions benefit from the extra depth — especially the ingredient benefit explanations, which are where the quality difference is most visible.
Include a tone-of-voice anchor in your instructions. If you have one piece of existing copy you're happy with, paste it into your batch instructions as an example. It calibrates the model's output to your brand voice faster than describing it in the abstract.
Plan to review 10–15% of outputs. Especially for products where your source data is thinner. It's still far faster than writing from scratch — and reviewing is a different cognitive load than creating.