Give the AI facts it can't guess (measurements, materials, who buys it, the questions customers ask before buying, and the doubt that loses the sale), then ask for a description that opens with what the buyer gets, answers those questions and ends with exact specifications. Check every figure against the product, and compare sales before and after.
Descriptions generated from a product name alone don't sell, because they're made of adjectives: "beautifully crafted", "perfect for any occasion", "a stunning statement piece". Buyers need answers, not adjectives. The difference between a description that sells and one that fills space is almost entirely in what you give the AI, which is why most of the work below happens before you open the tool.
The product fact sheet the AI can't write for you
For each product, fill in a short fact sheet. It takes five to ten minutes per product the first time, and much less once you have a spreadsheet template. Here is an illustrative filled-in sheet for a keepsake urn sold by a small funeral director through its online shop.
PRODUCT: Small keepsake urn, brushed brass
Measured: 9.5 cm tall, 7 cm wide at the base (measured, not
supplier sheet)
Capacity: 150 ml (supplier spec; checked by filling with rice)
Material: solid brass, lacquered; felt pad on the base
Closure: screw lid with a rubber seal
Who buys: family members sharing ashes between relatives; people
wanting a small urn at home alongside a scattering
Questions buyers ask before buying (from our inbox):
- Is it big enough for a share of ashes, and how much?
- Can it be engraved, and how long does that take?
- Will it arrive before [date]? How is it packed?
- Does the lid seal properly if it's moved or posted abroad?
Doubt that loses the sale: "Will it look cheap in person?"
Proof we have: close-up photos of the finish; 14 reviews mention
"heavier than expected"
Delivery and returns: dispatched in 2 working days, 5 if engraved;
unengraved items returnable within 30 days
Words to avoid: "perfect", "beautiful", "celebrate", exclamation
marksThree fields do most of the selling: the buyer questions, the doubt, and the proof. "Will it look cheap in person?" is the objection this shop hears on the phone, and the answer ("heavier than expected", from reviews) is the kind of detail an AI would never invent. Measuring the product yourself matters too. Supplier sheets are often wrong or list external dimensions for a capacity figure that belongs to a different model.
Mining the buying questions from your inbox and returns
You probably already know the questions, but they're scattered across emails, chat logs, phone notes and return reasons. Paste a few months of them into an assistant (with names and contact details removed) and ask it to organise them.
Below are customer emails, chat messages and return reasons from
the last 6 months for our [product category]. Customer details have
been removed.
1. List every question customers asked BEFORE buying, grouped by
product where possible, most frequent first, with a count.
2. List the reasons given for returns or complaints AFTER buying.
3. For each return reason, say which pre-purchase fact, if stated
in the description, might have prevented it.
Quote short phrases exactly; don't paraphrase customers' words.Illustrative output for the funeral director's memorial range, from about 120 messages:
Illustrative AI output: "Before buying: 1. Will it arrive by a specific date? (31) 2. Can it be engraved, and what font options? (22) 3. Is it big enough for [share / all] of the ashes? (19) 4. Is the lid secure for posting or travel? (9). Returns and complaints: 'smaller than I expected' (6) — prevented by stating height in centimetres with a common object for scale; 'lid was hard to open' (2) — prevented by explaining the sealed screw lid."
Check the counts against a sample of the messages yourself; assistants often round or merge categories. Even with approximate counts, the list tells you what each description must answer and in what order. Six returns for "smaller than expected" is a problem the description can fix before it happens.
A description prompt with a selling structure
With the fact sheet and questions ready, the prompt only has to impose a structure. This one works across most product types; change the tone line for your business.
Write a product description for our online shop using ONLY the
facts below. Don't add any claim, figure or feature that isn't in
the facts.
Structure:
1. Opening sentence: what the buyer gets or can do with it, in
plain words (not a list of adjectives).
2. One short paragraph answering the top 2 buyer questions.
3. One sentence that answers the doubt, using the proof provided.
4. A specification list: measurements, material, capacity/weight,
what's included.
5. One line on delivery and returns.
Tone: [e.g. quiet, practical, reassuring; no exclamation marks]
Length: 120-180 words plus the specification list.
Banned words: [your list]
FACTS:
[paste the fact sheet]The "ONLY the facts below" instruction is doing real work. Without it, assistants add plausible-sounding extras: "hand-finished", "dishwasher safe", "comes with a gift box". Each one is a promise, and some become returns.
A funeral director's memorial range, before and after
The illustrative funeral director sells about 40 memorial products online: keepsake urns, full-size urns, scattering tubes, and photo frames. The original descriptions came from supplier catalogues, lightly edited. Here's the keepsake urn before:
Before: "This beautiful brass keepsake urn is the perfect way to celebrate and remember your loved one. Elegantly crafted with a timeless design, it makes a treasured memorial for any home. A wonderful tribute."
It answers none of the questions buyers ask and uses three of the banned words. After running the fact sheet through the prompt, and one round of editing:
After: "A small brass urn for keeping a share of ashes at home, or for dividing ashes between family members. It holds up to 150 ml, which is a share rather than a full set of ashes; our team can advise on what size you need if you call us.
The screw lid has a rubber seal, so it stays closed if the urn is moved or posted. We can engrave a name and dates on the front; engraved urns are dispatched within five working days, others within two.
Customers often tell us it's heavier than they expected. It's solid brass with a brushed, lacquered finish that doesn't show fingerprints.
Height 9.5 cm · Base 7 cm · Capacity 150 ml · Solid lacquered brass · Felt base · Screw lid with seal
Unengraved urns can be returned within 30 days."
What the editor changed in the AI draft: it had written "holds a portion of ashes" (vague) and the owner replaced it with "a share rather than a full set", plus the offer to call, because sizing is the question families find hardest to ask. It had also written "a lasting tribute", which was cut. Everything else came through from the fact sheet.
The time cost, as an illustration: fact sheets for 40 products took about five hours spread over two weeks (measuring, checking the inbox, pulling review quotes), and generating and editing the descriptions took another three. The team's view afterwards was that the fact sheets were worth having regardless, because staff now answer phone questions from the same sheet.
Spec-led buyers: a packaging supplier's mailer box
Business buyers read descriptions differently. They scan for the numbers that decide whether the product fits their use, and adjectives slow them down. An illustrative before and after for a packaging supplier's printed mailer box:
- Before (AI, from the product name): "Our premium eco-friendly mailer boxes are perfect for brands that want to make a lasting impression. Sturdy, stylish and sustainable, they're ideal for all your shipping needs!"
- After (AI, from the fact sheet): "A one-piece folding mailer for shipping clothing, cosmetics and small gifts. Internal size 250 × 180 × 80 mm, which fits a folded hoodie or two 200 ml jars with padding. Made from 3 mm kraft corrugated board with a tuck-in lid and no tape needed. Printed one or two colours on the outside, from 250 boxes; plain boxes from 50. Most orders ship within 7 working days of proof approval."
The fact sheet for this product had a "who buys" line ("small clothing and cosmetics brands shipping direct to customers") and a "doubt" line ("will it survive the post?"), answered with the board thickness and a line about crush testing only after the owner confirmed the supplier's test data existed. The first AI draft had "eco-friendly" and "sustainable" in it; both went, because the fact sheet didn't support them. If a box is recyclable or made with recycled content, say exactly that, with the percentage from the supplier. Vague green claims are exactly what buyers and regulators have learnt to distrust.
Technical products: a laboratory's home test kit
Where a product makes a performance claim, AI is at its most dangerous, because it knows what such claims usually look like. An illustrative laboratory sells a home water testing kit: the customer takes a sample and posts it back, and the lab tests it for a set list of contaminants.
Asked for a description with only the product name and a few features, the assistant wrote: "Our lab-grade kit delivers 99.9% accurate results in just 48 hours, giving you total peace of mind about your family's water." Every part of that sentence was a problem. The lab had never stated an accuracy figure for the kit, turnaround was three working days from receipt, and "total peace of mind" implies the test covers everything, when it covers twelve named contaminants.
The rewrite, from a fact sheet the lab manager approved: "Take a water sample at home, post it back in the prepaid pack, and we'll test it for the 12 contaminants listed below. Results are emailed within three working days of the sample reaching us, with each result shown against its guideline limit and a plain-English note on what to do if anything is high. This kit doesn't test for [list]; if you need those, call us about a full analysis." For anything with health, safety or legal implications, have the person who knows the product sign off each description, and don't publish accuracy, detection limits or health claims unless they come from your own validated data.
One product, four channels
A description that works on your own shop usually needs cutting down for marketplaces and feeds. The limits below are the ones that most often trip people up.
| Channel | What to watch | How to prompt for it |
|---|---|---|
| Your own shop (e.g. Shopify) | Room for the full structure; Shopify Magic can draft for free on any plan | Use the full prompt; Magic or a chat assistant both work if fed the fact sheet |
| Amazon | Titles capped at 75 characters in all categories except media since 27 July 2026, plus Item Highlights | "Title under 75 characters: product type, key size, material. No adjectives." |
| eBay | Titles up to 80 characters | "Title under 80 characters with the words a buyer would search." |
| Google Merchant Center feed | Descriptions up to 5,000 characters; key details in the first 160-500; no price, shipping, sale dates, links or company name | "Remove anything promotional; put size and material in the first two sentences." |
The feed rules deserve attention, because a description copied from your shop often breaks them. "Free delivery this week!" and "Only at [shop name]" are both against Google's product data rules for the description attribute. If you use Shopify Magic rather than a chat assistant, the same principle applies: it can only work with what you give it, so paste the fact sheet into its input rather than a product name. Shopify Magic versus ChatGPT for descriptions compares the two in detail, and writing eBay listings with AI covers that marketplace's quirks.
Writing 200 descriptions without them all sounding alike
For a large catalogue, work in a spreadsheet: one row per product, one column per fact-sheet field, and a final column for the draft. You can paste 10 to 20 rows at a time into a chat assistant, or use a spreadsheet's built-in AI function if your plan has one. The risk at scale is sameness: forty descriptions that all open with "Crafted from…" read as machine-made, and near-identical pages don't help search either.
Add three rules to the batch prompt:
- "Vary the opening: never start two descriptions in this batch with the same word."
- "Open with the buyer's use or outcome from the 'who buys' column, not the material."
- "If a fact-sheet field is empty, leave that part out rather than filling it."
An illustrative set of three openings from one batch of the packaging supplier's range, after those rules: "A one-piece folding mailer for shipping clothing…", "Shipping candles or glass jars? This double-wall carton…", "The smallest box in our range, sized for jewellery, cards and…". Spot-check at least one in five drafts against the fact sheet, and every draft that includes a number. The third rule prevents the most common batch error: the AI filling an empty "capacity" field with a guessed figure because the row above had one.
Errors that quietly cost sales
None of these is dramatic, and each shows up as returns, questions or abandoned baskets rather than complaints.
- Invented dimensions. An illustrative shop's AI described a photo frame as "holds 6 × 4 inch photos" because that's common; it held 5 × 7. Eleven returns in a month before anyone noticed.
- Wrong material. "Solid oak" for an oak-veneer product. Customers notice, and some ask for refunds on principle.
- Features from a different variant. The engraving option described on a product that can't be engraved, because the fact sheet was copied from its sibling.
- "Perfect for" lists. "Perfect for weddings, birthdays, anniversaries and more" is filler, and it pushes the useful facts below the fold on mobile.
- Keyword stuffing. Repeating the product type six times reads badly and doesn't help modern search.
- Delivery promises that went stale. "Dispatched next day" written into 200 descriptions, then the courier changes. Keep delivery lines in one shared block your shop can update in one place, where your platform allows it.
More of these, and how they erode trust across a whole store, are collected in online shop AI mistakes that hurt trust and conversions. If you publish any claim with a number in it, run it past the checks in fact-checking AI marketing copy first.
Proving the new copy sells
Most small shops can't run a proper A/B test on product descriptions, so use a before-and-after comparison by product group. Pick a group of similar products (say, the 12 keepsake urns), rewrite them all on the same day, and compare four weeks before with four weeks after on three numbers from your shop analytics: product page views, conversion rate (orders divided by product page views), and returns or complaints mentioning size or description.
A quick sum shows how to read it. If the urns had 1,600 product page views and 24 orders in the four weeks before (1.5%), and 1,500 views and 33 orders after (2.2%), that's an encouraging change. It's still worth checking that nothing else changed, such as prices, photos, a promotion or a seasonal effect, and it's worth repeating with a second product group before rewriting the whole catalogue. Returns tell you something too: if "smaller than expected" disappears from the return reasons, the description is doing a job the photos couldn't.
Whatever the numbers show, keep the fact sheets. They're the reusable asset: the next time a supplier changes a product, a marketplace changes its rules, or you want ad copy, you rewrite from facts you've already checked, in minutes. For a store on Shopify specifically, writing Shopify product descriptions that convert covers the platform's own tools and settings in more depth.
Product description questions
How long should an AI-written product description be?
Long enough to answer the questions buyers ask before they buy, and no longer. For a simple item that might be 80 words plus a specification list; for an expensive or technical product, 250 to 400 words. Put the most important details first: Google's Merchant Center guidance, for example, says to put key details in the first 160 to 500 characters.
Will using the same AI prompt for every product hurt my search rankings?
The prompt isn't the problem; identical output is. Search engines and shoppers both struggle with near-duplicate pages, so give each product its own facts and buying questions, and vary openings deliberately. Variants of one product, such as colours, can share a description, but separate products should not.
Should I say that my descriptions were written with AI?
There's generally no rule requiring shops to label AI-assisted product copy, and most don't. What matters is that every statement is true and checked by someone who knows the product. If AI helped with images that change how the product looks, that's a different question and is worth disclosing.
Further reads
- How to Build a Brand Voice Guide That AI Can Follow — Give every description the same recognisable voice.
- How to Create Product Photos With AI on a Small Budget — Pair the new copy with honest, well-lit product photos.
- How to Turn Customer Reviews Into Marketing Copy With AI — Pull buyers' own words from reviews into your descriptions.
- How to Use AI for Etsy Titles, Tags and Descriptions — Adapting descriptions for Etsy's titles and tags.
- Taking a Local Shop Online With AI: Product Listings in a Weekend — A weekend plan for listing a whole range from scratch.
- How to Use AI to Find Out Why Website Visitors Don't Buy — Find other reasons product pages aren't converting.
- Does Writing Etsy Listings With AI Hurt Your Search Ranking? — How Etsy search actually ranks listings, the specific ways AI-drafted copy loses ground, and a simple test to run on your own shop.
- Is It Cheaper to Use AI or a Freelancer for Marketing Content? — A per-piece cost comparison of AI and freelance writers for marketing content, counting your own editing time, with three worked cases.
- AI Product Photos vs a Photographer: Cost and Quality Compared — Per-image costs of AI product photo tools against a photographer's quote, a shot-by-shot quality verdict and a 40-product worked example.
- How to Write Facebook and Instagram Ad Copy With AI — A creative brief, angle-based prompts, Meta's recommended text lengths and the policy traps AI walks into, worked through for a real-looking cookery class ad.
- Does Google Penalise AI-Written Content on a Business Website? — What Google's rules actually say about AI-written pages, how to tell a penalty from a page that never ranked, and a value check for every draft.
- Can a Small Online Shop Use AI to Compete With Bigger Brands? — Compete through clearer product choices and dependable service, with a practical AI trial and an honest margin calculation.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Shopify help pages (Shopify Magic); Google Merchant Center product data specification (description attribute); Amazon and eBay title limits as published by each marketplace.