TL;DR
Feed a model your specs, your customer’s top pains, and one differentiator, then make it write benefit-first copy inside the platform’s character limits. The win is not “AI writes faster” — it is that every SKU gets a unique angle so you avoid the duplicate-content trap that sinks marketplace listings. Use Shopify Magic or Amazon’s built-in AI generator for bulk drafts, then edit with the review checklist below. Plan on roughly 180-220 words for retail pages, and respect the hard limits: Amazon titles cap at 200 characters, third-party bullets at 255 characters each, descriptions at 2,000 plain-text characters (as of June 2026).
The task
You have a new SKU and need description copy for a product page, a marketplace listing, or a catalog. The copy has to translate specs into reasons to buy, fit your brand voice, and survive being read on a phone in five seconds. The trap is that AI makes it easy to generate a thousand near-identical paragraphs — exactly the “scaled, low-effort” pattern Google’s March 2026 core update demotes, and the duplicate copy that Amazon and Google both detect even after paraphrasing.
When AI earns its place
- You launch 5+ SKUs per week and hand-writing each one is the bottleneck.
- You already have a positioning angle and just need fluent first drafts.
- You want one consistent structure across a category (every chef’s knife uses the same five-section layout).
- You are adapting an existing description into a second voice, locale, or marketplace.
When not to lean on AI alone: flagship hero products where every word should be hand-tuned; regulated categories (supplements, medical, financial) where an unverified claim is a liability; and brands with an idiosyncratic voice the model has seen too few examples of.
Pick the tool for the channel
| Tool | Best for | Cost (June 2026) | Note |
|---|---|---|---|
| Shopify Magic | Bulk drafts inside Shopify admin | Free on every plan | 6 tone presets (expert, persuasive, supportive, daring, playful, sophisticated); pulls from product attributes |
| Amazon AI-Generated Listing | New Amazon listings (title, bullets, description) | Free in Seller Central | Catalog → Add Products → AI; fills 70%+ of attributes, adds Rufus Q&A formatting |
| Claude (Sonnet 4.6 / Opus 4.7) | Brand-voice matching, longer copy | Pro $20/mo, Free tier limited | Strongest at mimicking sample copy you paste in |
| GPT-5.5 in ChatGPT | Fast drafts + image-aware prompts | Plus $20/mo, Free $0 | Good default; Instant mode is fastest for batches |
| Gemini 3.1 Pro | Catalog-scale runs, sheet workflows | Google AI Pro $19.99/mo | 1M-token context fits a whole category spec dump |
Amazon reports independent sellers created more than 12 million listings with its AI tools in 2025, and over 1.3 million sellers have used them — so the channel-native generators are a real starting point, not a novelty. Use them for the draft, then edit; do not ship the raw output.
What to feed the model
- Product name, category, and hard specs (size, material, weight, capacity).
- The top 2-3 customer pains this product solves, in the customer’s own words if you have reviews to quote.
- The single biggest differentiator versus the closest competitor.
- Target audience (age, context of use, price sensitivity).
- 1-2 sample descriptions whose voice you want to match.
The differentiator and the pasted voice samples are what break you out of generic output. A model with no angle defaults to “premium high-quality” filler; a model with a pain and a competitor writes something specific.
Copy-ready prompt
You are a senior ecommerce copywriter for [brand].
Write a product description for [product_name].
Specs: [specs]
Audience: [audience]
Top pains: [pain_1], [pain_2]
Differentiator vs [competitor]: [differentiator]
Voice samples to mimic: [sample_text]
Output:
1. Hook line, max 12 words, benefit-first.
2. 3 benefit bullets — each starts with a verb, ties to a pain.
3. Two-sentence "who it's for" paragraph.
4. Spec table (plain markdown).
5. Three-question FAQ.
6. CTA line, 8 words max.
Rules:
- Use ONLY specs I provided. Do not invent measurements, materials, or claims.
- No superlatives without proof. Ban the words "revolutionary",
"game-changing", "unleash".
- Total length 180-220 words.The “use ONLY specs I provided” line matters: models will confidently invent a capacity or a fabric blend if you leave a gap, and a wrong spec on a live listing is a returns problem, not a typo.
Fit the copy to the platform’s limits
AI happily writes past the cutoffs, so trim to these before publishing (all current as of June 2026):
| Field | Limit | Notes |
|---|---|---|
| Amazon title | 200 chars (150 for apparel/jewelry) | First ~80 chars carry the SEO weight on mobile |
| Amazon bullet point | 255 chars each (3P sellers); up to 500 for Vendor Central | Keep all five bullets under ~1,000 chars total so mobile doesn’t truncate |
| Amazon description | 2,000 chars, plain text | No HTML for standard listings; A+ Content is separate |
| Amazon backend search terms | 249 bytes | Measured in bytes — one byte over silently de-indexes all of them |
| Shopify / own store | No hard cap | Aim 180-220 words retail, 250-400 for considered purchases |
How to check the output
- Read the hook out loud. Would a stranger want the next sentence?
- For each bullet, ask “so what?” If you cannot answer in one phrase, it is a feature, not a benefit.
- Verify every spec number against the source sheet — assume the model guessed any number you did not supply.
- Run the first paragraph through a duplicate check if you sell on marketplaces. Google and Amazon both flag near-duplicate copy even when it is reworded.
- Confirm you stayed inside the character limits above for the target channel.
Keep the SEO honest
Google does not penalize copy for being AI-assisted; it penalizes thin, scaled, duplicate output produced “to manipulate rankings.” Two practices keep you safe: give every SKU a unique pain angle and unique specs so no two descriptions paraphrase each other, and edit the draft as a human would — Google’s own guidance on generative AI content rewards genuine expertise, not raw output. If you publish AI-generated product data through a feed, label it as AI-generated where the attribute calls for it.
Common mistakes
- Listing features in bullets instead of benefits.
- Generic adjectives (“premium”, “high-quality”) with no proof behind them.
- Forgetting a CTA on the page even when the description is strong.
- Letting the model invent specs that are not in the source.
- Pasting the same AI draft across Amazon, Shopify, and your own store — three duplicates, not three listings.
Next steps to keep improving
Save your three best descriptions as voice samples and feed them back as exemplars. Track add-to-cart rate by SKU and re-prompt the laggards with a sharper pain statement. Build a category template library so a new SKU is a 5-minute job, not a 90-minute one.
FAQ
- Will AI descriptions hurt SEO? Not by themselves. Google demotes thin, duplicated copy regardless of how it was written. Pass unique specs and a unique pain angle per SKU and edit before publishing.
- How long should a description be? 120-220 words for most retail, 250-400 for considered purchases like furniture or electronics. On Amazon you also have the 2,000-character description cap and 255-character bullet limit to work within.
- Which tool should I start with? If you are on Shopify, Shopify Magic is free and built in. On Amazon, use the AI-Generated Listing flow in Seller Central. For brand-voice matching, paste samples into Claude or ChatGPT.
- Can I skip the FAQ block? Not recommended. FAQ blocks capture long-tail search and reduce pre-sale questions.
- What tone works best? Match the price point. Premium brands lean spare; mass-market leans warm and direct.
- What about the collection page that holds these SKUs? Different copy shape — see AI Shopify collection descriptions.
Related
For deeper prompt patterns, see product description prompts, Amazon bullet point templates, and beauty product description prompts.
Tags: #E-commerce #Workflow