How to Write 500 Product Descriptions Without Hiring a Copywriter
A repeatable system for generating unique, on-brand, SEO-ready descriptions at catalog scale — and the mistakes that get pages deindexed.
Large catalogs have a quiet, expensive problem: most product pages are either the manufacturer's boilerplate or one line of text. Both are invisible to search, and both convert badly.
A freelance copywriter runs $25–$75 per description. At 500 SKUs that is $12,500 and about three months. Here is the system that gets you there in an afternoon without producing spam.
## Step 1: Fix the inputs before the output
AI writing is only as good as the structured data behind it. Before generating anything, make sure each SKU has:
- Real attributes: material, dimensions, weight, fit, ingredients, compatibility
- The customer problem it solves, in one sentence
- Who it is *not* for (this is what makes copy specific)
- A primary keyword, usually the head term plus one qualifier
Missing attributes are where hallucinations come from. The model does not invent a fabric weight because it is dishonest; it invents one because you left the field blank.
## Step 2: Define brand voice once, not per product
Write a short voice profile: three adjectives, two banned phrases, reading level, whether you use humor, whether you use "you." Apply it to every generation. Without it, product 1 sounds like a technical manual and product 400 sounds like a lifestyle blogger.
## Step 3: Use a consistent structure
The format that converts and indexes well:
1. **Hook** — one line naming the problem or the payoff
2. **Body** — two short paragraphs on how it works and who it is for
3. **Specs** — bulleted attributes, scannable
4. **Objection handler** — sizing, care, compatibility, returns
5. **CTA** — plain and direct
Consistency here is a feature. Shoppers comparing four of your products should not have to relearn the layout each time.
## Step 4: Generate in batches, review in batches
Do 25 at a time. Read all 25 together — that is how you catch repeated sentence openers, overused adjectives, and the model's favorite crutch phrases. Fix the prompt, not the individual outputs, and regenerate.
## Step 5: Handle near-duplicate variants
The same shirt in nine colors should not be nine near-identical indexed pages. Use one canonical product page with variant selectors, and let the color live as a variant attribute rather than a separate URL. Where variants genuinely deserve their own page, the copy must differ in substance, not just the color word.
## The mistakes that cost you rankings
- **Publishing unreviewed.** Every claim about safety, materials, or compliance needs a human check. Always.
- **Keyword stuffing.** Once in the H1, once in the first paragraph, naturally in the body. That is it.
- **Ignoring meta descriptions and alt text.** They are part of the job, not an afterthought.
- **Same intro sentence across the catalog.** The fastest way to look auto-generated.
- **Dropping the specs.** Specifics are what both shoppers and AI assistants extract. Vagueness is worthless to everyone.
## Step 6: Measure
Tag the batch with a publish date and watch impressions, average position, and add-to-cart rate for those URLs over the next 60 days against a control group of untouched pages. If the rewritten set does not beat control, your voice profile or your structure is wrong — not the idea.
## How ForgeMetric helps
Operator OS reads your Shopify catalog directly, so it generates from your real attributes instead of guesses. Point it at a collection and it drafts unique, on-brand descriptions for every product in it — with meta descriptions, alt text, and internal links included — then pushes approved copy straight back to Shopify.
Content is deduplicated against your email, blog, and social output, so you are not cannibalizing your own pages. And you review everything before a single page goes live.