8 min read
AI Product Descriptions for a Shopify Catalog: What Actually Works
A product export out of an ERP reads like a warehouse manifest. Turning 4,000 of those rows into catalog copy is the one job where AI earns its keep — if you constrain it properly.
The problem with ERP product names
ERP names are written by operations for operations: `BLU-CTN-TSHRT-XL-2PK`. They are precise, sortable, and completely unsellable. Shopify search, Google, and your own on-site filters all index that string and find nothing useful in it.
Rewriting 4,000 of them by hand is a copywriting invoice most merchants never pay, so the catalog ships as-is and the SKUs quietly never get found.
What to generate, and what not to
Generate
- A title under 70 characters that reads like a product, not a part number.
- A 2-3 sentence description with the words a buyer would actually type.
- 3-6 lowercase tags for on-site search, filters, and collection rules.
Never generate
- Materials, dimensions, or weights not present in the source data.
- Certifications, safety claims, or compliance language.
- Origin claims, warranty terms, or anything a buyer could hold you to.
The rule is simple: a model may rephrase what it was given and may not add facts. Anything it invents becomes a product-liability claim on your store, not on the model.
Batching and cost control
Run in batches of about 25 products. Small batches fail cheaply, keep each request well inside provider timeouts, and let you inspect tone before committing to 4,000 rows.
Bill against a monthly quota rather than per call, and only count generations that actually returned copy. A failed batch that burns quota is the fastest way to lose trust in the feature.
The review gate is not optional
Nothing should be written to Shopify until a human approves it. Generate, show a diff against the current title and description, and let the merchant approve, edit, or discard per product.
Hasil's AI product enrichment works exactly this way: it generates, it shows you the result, and it writes to Shopify only after you approve. Quota is consumed per product that actually produced copy.
Where the copy pays for itself twice
The same generated fields feed more than the product page: Google Merchant Center and Performance Max feeds, on-site search relevance, collection automation rules, and email or LinkedIn product copy. One enrichment pass, four downstream uses.
Frequently asked
- Will AI-generated product descriptions hurt my SEO?
- Not on their own. Search engines penalise thin, duplicated, and inaccurate pages — not the tool that produced the text. Copy that is unique per SKU, factually grounded in your own product data, and reviewed before publishing performs like any other product copy.
- How many products can be enriched at once?
- Hasil generates in batches of up to 25 products per run, against a monthly quota that depends on your Hasil Sync plan.
- Does the AI write directly to my Shopify store?
- No. Copy is generated for review first. Nothing is written to Shopify until you approve it.
- What model is used?
- Generation runs on Google Gemini 2.5 Flash, chosen for the balance of catalog-scale throughput and copy quality.
Want this working on your store?
Hasil ships wholesale portals and Odoo ↔ Shopify sync. 14-day free trial.
Start free trial