TL;DR: Bulk descriptions read alike because they are written from the same few words. Fix the input, not the prompt. Build a spreadsheet with product-specific slots (material, measurement, use, origin, one detail only this product has), let AI write each description from its row, keep legal information in fixed blocks outside the generated text, and check repetition with a simple count rather than by reading every page.
A home-goods store with 300 products runs a bulk AI rewrite of its descriptions on a Sunday. On Monday the owner opens five pages at random. Four begin with "Transform your space with", three call the product "timeless", and a doormat and a lamp share a sentence about "effortless style for every room". Each page is fine on its own. Together they read like one page with the product name changed.
That is the core problem when you learn how to write product descriptions for hundreds of products. The prompt is rarely the issue. The input is: if each product has nothing but a name and a category, every description is built from the same raw material. This article is part of our guide to marketing strategies for a small Shopify store.
How this was made: written by Miriad with AI assistance and checked against the sources listed at the end.
How do you write product descriptions for hundreds of products?
Put each product's facts into a spreadsheet with fixed columns, write one template with slots for those facts, and let AI write each description from its own row. Keep legal and safety details in separate fixed blocks. Then check a sample by hand and the whole batch for repeated phrases. The spreadsheet does most of the work.
The order of work:
- Export the catalogue to a spreadsheet.
- Add the slot columns (below) and fill them, by hand or with a supplier's data sheet.
- Write the template and test it on ten products from different categories.
- Generate the batch.
- Run the repetition check and fix the top offenders.
- Read a sample by hand, plus every bestseller.
- Publish in groups, not all at once, so mistakes stay small.
Why do bulk descriptions all sound the same?
They sound the same because each one is generated from too little product information, so the model fills the gap with its most likely phrases. Fewer facts per product means more generic adjectives. Adding three or four specific facts per product changes the result more than any instruction to "be original" or "vary the wording".
You can see it in the openings. When most rows only contain a name, a category and a colour, the model has one thing to say: something general about the category. Give it a measurement, a material, a use case and a maker's note, and there are four different things to open with.
A second cause is the template itself. If it says "start with a benefit statement", every description starts with a benefit statement. Let the slot decide the opening instead: for one product the material, for another the use.
The slot template
A slot template has one column per fact and a rule for how each fact is used. The minimum is: product type, the one detail only this product has, material, measurements, main use, who it suits, care, and what is in the box. Add a column for the opening, chosen per product, so first lines differ.
| Slot | What goes in | Rule |
|---|---|---|
| Opening | One of: material, use, origin, detail | Different from the previous product in the same category |
| Unique detail | Something only this product has | Must be a fact, not an adjective |
| Material | What it is made of | Exactly as on the data sheet |
| Measurements | Size, weight, capacity | With units |
| Main use | One concrete situation | No "perfect for any occasion" |
| Suits | Who it is for | One line |
| Care | How to clean or store | From the maker |
| In the box | Contents | Complete list |
Example (illustrative, not a real case): one row from the home-goods store.
- Opening: use
- Product type: jute doormat
- Unique detail: rubber backing cut to the outline, so the edge does not curl
- Material: natural jute, rubber backing
- Measurements: 60 × 40 cm, 1.5 cm thick
- Main use: covered front door, outside
- Suits: entrances where wet shoes come in
- Care: shake out; brush when dry; do not soak
- In the box: one mat
Generated description: "For the covered front door where wet shoes come in. The jute top scrapes off grit, and the rubber backing is cut to the mat's outline, so the edge stays flat instead of curling. 60 × 40 cm and 1.5 cm thick. Shake it out weekly and brush it once it is dry; soaking weakens the fibres." No "transform", no "timeless", and nothing a doormat from another store would also say.
Which parts must never be generated?
Keep legally required and safety information in fixed blocks that the AI does not rewrite: manufacturer details, safety warnings, product identifiers, and any limits on the right of withdrawal. A rewrite should never be able to drop or change them. Generate the selling text around these blocks, not through them.
In the EU, the General Product Safety Regulation has required since 13 December 2024 that every online offer shows the manufacturer's name, postal and email address, an EU responsible person if the maker is outside the EU, a product picture, a type or identifier, and any safety warnings (CMS summary). Make each a column, and render it in the same place on every page.
Withdrawal exceptions need the same care. Hungary's consumer contract decree, for example, lists goods without a right of withdrawal, including items made to the buyer's specification, perishable goods and sealed goods unsealed after delivery for health or hygiene reasons (45/2014 Government Decree). If a product falls under one, say so in a fixed line, not in AI-written prose that may soften it.
How do you check hundreds of descriptions without reading every one?
Count instead of reading. List the most common opening words and three-word phrases across the batch, fix any that appear on many pages, then read a random sample and every bestseller by hand. A spreadsheet formula or a short script does the counting in minutes; the reading is where your judgement goes.
A practical check, in three passes:
- Openings: extract the first four words of every description and sort by frequency. If one opening appears on more than a handful of pages, rewrite those rows' opening slot.
- Phrases: count three-word phrases across the batch. Ban the top generic ones ("for every room", "add a touch") in the template.
- Facts: read 20 random descriptions and all bestsellers against the spreadsheet. Every number must match.
Sample size for the hand check = 20 per batch of 100 + every bestseller + every product with a safety warning. It is a starting rule, not a statistical sample; increase it if the first check finds errors.
How should feed titles and descriptions differ from page text?
Feed titles should be plain and structured: brand, product type and the distinguishing attribute, within Google's length limits and with no promotional text. The page description can be warmer and longer. Generate both from the same row, with different rules, so the facts match and the tone fits each place.
Google Merchant Center sets the limits: titles up to 150 characters, descriptions up to 5,000, and no promotional text such as "free shipping" in titles (Merchant Center Help). In the home-goods store's example, the AI rewrote each feed title within 150 characters (brand, product type, size or colour), removed "free shipping" and capitals, and the owner spot-checked 20 against the product pages before resubmitting.
For the rest of the store setup, see how to build a Shopify store with AI and still sound like yourself. For titles, structured data and the feed, see the Shopify store SEO checklist. For the full plan, return to the Shopify store marketing guide.
If you want tested description and product-page structures to put into your slot template, we are building the Conversion Copywriting Pack. Get notified when it launches.
FAQ
How long should a product description be?
Long enough to answer the questions a buyer asks before buying that product, and no longer. A simple consumable may need three sentences; a product with sizes, compatibility or care rules needs more. Lead with the facts that decide the purchase, keep measurements and care in a consistent place, and cut any sentence that could describe a different product.
Can I use the manufacturer's description?
You can, but many other stores selling the same product will have the same text, and it rarely answers your buyers' questions in your voice. Use the manufacturer's data sheet as the source for facts, such as materials, measurements and care, and write your own description from those facts. Keep the manufacturer's safety warnings and identification details exactly as supplied.
What if I do not have facts for every product?
Fill the slots you can and leave the rest empty rather than letting AI guess. A shorter description with true facts is better than a longer one with invented ones. Mark rows with missing facts, ask the supplier for a data sheet, and prioritise the bestsellers. Measuring and photographing a product yourself can be quicker than waiting for a reply.
Sources
- The new EU General Product Safety Regulation: what you need to know — CMS, 2024.
- 45/2014. (II. 26.) Korm. rendelet (consumer contracts decree, Hungarian) — Nemzeti Jogszabálytár, text in force 2026.
- Product data specification — Google Merchant Center Help, accessed 2026.