A brand sounds like itself across hundreds of products when its voice is written down as a system: a copy spec for each product category, a shared vocabulary, and a small set of brand lines placed where they fit, all fed by accurate product data. Templates handle the structure and people handle the judgment. AI-assisted descriptions get the same accuracy review as anything else before they publish.
Voice usually holds up on the launch pages, where the best writer spent a week. It thins out by product four hundred, written on a deadline from a spreadsheet row, and by product nine hundred it reads like the supplier's catalog. Shoppers can notice the drop, and so can anyone comparing your pages with a competitor's.
This page is for brand owners and ecommerce leads planning product content. It is not legal advice, and claims inside product copy, such as ingredient or material claims, still go through whoever approves claims for your brand. The broader plan is on our retail, DTC and beauty brands page.
Voice at scale is a system, not a writer
A brand with a dozen products can rely on one person's ear. A brand with eight hundred cannot, because no single writer touches every page and each new hire learns the voice from whichever page they read first. A style guide is not a copy system. The guide says the brand is warm and direct; the system says what a warm, direct description of a linen duvet cover contains, in what order, and in which words.
The system has three parts. A category copy spec sets the fields each description must answer. A vocabulary list records the words the brand uses, the words it avoids, and the words that need approval because they are claims. A short bank of signature lines, the brand's real turns of phrase, gets placed deliberately rather than sprinkled on every page until it means nothing.
Write a copy spec for each category
The spec starts from what shoppers ask in that category, not from whatever the merchandising sheet happens to include. A candle shopper asks how strong the scent is and how long it lasts; a sweater shopper asks about weight, itch, and shrinkage. The brand's voice then decides how those answers sound. Here is a hypothetical spec for a fictional home goods brand:
| Field | What it answers | Source | Voice note |
|---|---|---|---|
| Opening line | What this is and who it suits | Merchandising brief | One plain sentence, no stacked adjectives |
| Feel and material | What it is like to use | Tech pack and wear test notes | Sensory words from the approved list |
| Size or fit note | How it sizes, or how it fits the space | Return reasons and customer service macros | Direct, second person |
| Care pointer | How to keep it looking right | Care label file | Short, with a link to the care page |
| Brand line | Why it belongs to this brand | Signature line bank | At most one per description |
Test the spec before it ships. Ask the merchandising lead to read ten finished descriptions with the brand name removed and answer one question: "Could a regular customer tell these were ours?" If the answer is no, the spec needs more of the brand's own words and fewer category defaults. Apparel brands can build the fit field from our page on fit, sizing and use content.
Feed the voice accurate product data
Voice cannot fix a wrong fact. A charming line about a "heavyweight" throw is a problem if the throw is the lightest in the range, and at scale the error repeats in every description built from the same bad field. The writer should not be guessing a fabric weight from a product photo or a fit from the style name; those facts come from the tech pack and the return data, and the description is assembled around them.
Variants need the same care. Color and size variants often share most of their copy, which is reasonable when the product is the same. Google's documentation on duplicate URLs describes specifying a canonical URL "for duplicate or very similar pages," and lists "To consolidate signals for similar or duplicate pages" among the reasons to do it. Ask your developer how your platform handles variant URLs, then write once for the product and add only what actually changes by variant, such as a color that runs sheer. A variant page that repeats the parent's copy word for word, with only the color name swapped, adds nothing a shopper can use.
Where AI drafting fits
AI tools can draft from a good spec and good data, and catalog volume is the problem many brands want them for. Google's guidance on generative AI content says it "can be particularly useful when researching a topic, and to add structure to original content," and also that "It is critical to manually factcheck and review all AI-generated content for accuracy." The same guidance warns that generating many pages without adding value for users may violate its spam policy on scaled content abuse.
Two working rules follow. The model drafts from your fields and your vocabulary list, never from its own sense of what a product like this usually has. Someone who knows the product reads every description before it publishes. If you send product data to Google Merchant Center, the same guidance says AI-generated product titles and descriptions "must be specified separately and labeled as AI-generated," so confirm with whoever manages your feed how that labeling is handled.
A fictional before and after
Here is a hypothetical description from a fictional home goods brand in Vermont with several hundred products, for a wool throw.
Before: This luxurious throw adds cozy elegance to any space. Crafted with premium materials for lasting comfort.
After: A midweight wool throw for the end of the bed or the back of the sofa. It is warm without being heavy, and our wear testers found it softened a little over the first few cold washes. Like everything we make, it is meant to be used, not saved for guests.
Every fact in the after version comes from a field: the weight from the tech pack, the washing note from the care file, the softening from the brand's own wear testing. The last sentence is a signature line, and the spec allows one per description, so it reads as the brand's point of view rather than a slogan. Our article on AI content that still sounds like your business works through the same idea for service businesses.
In Hudson South's process, the Authority Asset gathers the product data, claims list, and customer service answers the descriptions are built on. Hudson Twenty-One records how the brand communicates across 21 dimensions, so descriptions built from each copy spec sound like the brand. Market Intelligence researches what shoppers ask in each category, and AEO Creation writes the descriptions and category pages from all three.
Hear your brand on product nine hundred
On a free 30-minute call, Hudson South compares how your product pages and your competitors' pages answer the questions shoppers ask in your categories.
Get your free competitor analysisSources
- Google Search's guidance on using generative AI content on your website. Google Search Central. Read 2026-10-16.
- How to specify a canonical URL with rel="canonical" and other methods. Google Search Central. Read 2026-10-16.