How to Write Product Descriptions That Rank, Convert and Get Picked by AI (with Examples)
A working guide to the product description from people who rewrite them for a living. What to include, how to structure it, how long it should be, plus how to write thousands with AI without losing accuracy.
By Alberto Barberis, Founder and CEO, AndromedAI · Updated October 2026
The short answer
A product description is the text that explains what a product is, what it's made of, how it fits or works, who it's for and when to use it. A good one opens with the main benefit, states every buying attribute in plain sentences, uses the words shoppers search with and answers pre-purchase questions, so shoppers, search engines and AI agents can match it to a need.
Product descriptions in 30 seconds
A product description is the evidence every channel uses to decide whether your product fits a need. Write it for people who skim, for search engines that rank pages and for AI agents that filter on facts.
- 01
Every description now has three readers: a shopper skimming on a phone, a search engine deciding where the page ranks, and an AI agent deciding whether the product makes the shortlist at all.
- 02
People skim. In Nielsen Norman Group research 79% of users scanned new pages and only 16% read word by word. The first lines and the bullets do most of the selling.
- 03
A good product description states every fact a buyer needs, in the words buyers actually use, and answers the questions they ask before they buy.
- 04
The structure that serves all three is plain. Name the product and its benefit in the opening line, tie benefits to features, write attributes and use cases as sentences, and close with specs and a short FAQ.
- 05
At catalog scale AI does the writing. What decides whether you can publish it is the product data behind it, your written brand rules and the checks that run before anything goes live.
What a product description must do in 2026
It has to let a shopper decide in seconds, give search engines a clear match and hand AI agents facts they can verify. Thin or inaccurate copy now costs you in sales and returns, and it keeps you out of AI answers.
of US and UK shoppers returned items because of inaccurate descriptions
Bad copy keeps costing money after checkoutSource: Salsify 2025 Consumer Researchof shoppers returned an online purchase because of incorrect or misleading information
A year later, accuracy is still the top complaintSource: Salsify 2026 Consumer Researchof users scan a new page; only 16% read word by word
Write for the skim firstSource: Nielsen Norman Grouphigher conversion for AI-referred visitors to US retail sites than non-AI traffic in July 2026
People who arrive from an AI answer have mostly decidedSource: Adobe AI Traffic Trends, Aug 2026customers used Amazon's Rufus assistant in 2025, helping drive nearly $12B in incremental annualized sales
An assistant reads your listing before the shopper doesSource: Modern Retail, Amazon earningscharacters: the opening of a Merchant Center description shoppers see before they click to expand
Anything important goes up frontSource: Google Merchant Center HelpThree readers, one text
The shopper
Wants to know fast whether the product suits their body, their room, their device or their budget. Reads the opening and the bullets, glances at the specs, then buys or bounces.
The search engine
Matches the page to queries by reading the title and headings first, then the description and structured data. Pages that cover a product completely, in their own words, have the edge.
The AI shopping agent
Turns "a breathable linen shirt for a summer wedding, under $90" into a set of constraints, then drops every product whose data can't confirm one of them.
What changed with AI shopping
Search engines rewarded keyword matches. Agents need more than that. They have to confirm facts, then explain the pick to the shopper in a sentence or two.
The platforms say so in their specs. OpenAI's product feed specification defines the description field as a "Factual product description for this item", in plain text, up to 5,000 characters. In 2026 Google added optional conversational attributes to Merchant Center, such as question_and_answer, which it says help "AI systems and conversational agents better understand your products' specific nuances" (Search Engine Roundtable).
Here's how that plays out in a chat. The agent quotes attributes straight from the listing.
ChatGPTShoppers already use these tools. McKinsey and ICSC found that 68% of surveyed consumers used at least one AI-enabled tool in the past three months (Luxury Daily). Salsify's 2026 research names detailed descriptions and clear specifications as the top triggers of trust in AI-recommended products (Salsify).
The product description is now the evidence an AI agent uses to justify recommending your product. A fact that isn't written down can't be used.
That's the idea behind Agentic Commerce Optimization (ACO). Search engines and shopping ads read the same product data as AI agents, so one better description pays off in every channel.
What should a product description include?
Cover the product's identity and main benefit, its materials, dimensions and fit, compatibility and use cases, who it's for, what's in the box, care, and answers to common questions. Prices, policies and promotional claims belong elsewhere on the page.
In practice a complete description comes down to ten elements.
- Product identity
Product type and brand, then the model and variant in plain words: "men's merino crew hiking sock, size L, charcoal".
- Primary benefit
The one outcome the buyer cares about most, in the first sentence.
- Materials and composition
Exact composition with percentages, full ingredients for beauty and food, plus what's deliberately absent ("BPA-free").
- Dimensions and fit
Measurements with units and fit guidance. Give a second unit system for international buyers.
- Compatibility
Specific model numbers and standards. "Fits most models" is a return waiting to happen.
- Use cases
The situations it's made for: commuting, a summer wedding, a small balcony, induction hobs.
- Who it is for
Skill level, age range, skin type or household size where it matters. Also who it isn't for.
- What is included
Everything in the box, and whether items in the photos are sold separately.
- Care and maintenance
Washing and storage instructions, plus expected lifespan. These cut returns and support tickets.
- Answers to top questions
The three to five questions customer service hears most, answered on the page and in the FAQ.
Most checklists stop at "features and benefits". When we audit a catalog, what's missing far more often is use cases and the honest "not for" line, exactly what an agent needs to include you in an answer or rule you out before a return.
Where each element lives
The description is one field among several. Strong product pages repeat the same facts across the visible page, the structured data and the feed, and keep them identical. Salsify found that 54% of shoppers dropped purchases because of conflicting product information (Salsify).
| Element | On the product page | In the Google feed | Why AI agents care |
|---|---|---|---|
| Identity | Title and H1 | title, brand, gtin, mpn | Lets the agent match the exact product and variant |
| Benefits | Opening sentence and bullets | product_highlight (2 to 100 per product, up to 150 characters each) | Gives the agent a reason to recommend it |
| Attributes | Description body and spec table | material, color, size, pattern, product_detail | Confirms the constraints in the shopper's request |
| Use cases and fit | Description body | description, product_type | Matches situational requests |
| Questions | FAQ block | question_and_answer (conversational attribute) | Answers follow-up questions in AI chats |
Google's product highlight guidance recommends four to six highlights focused on selling benefits, with technical, verifiable data kept in product_detail.
One Shopify trap: a product has a single description shared by every variant, so a fact that applies to one color has to read correctly for all of them. And if your size guide lives in a theme block or a custom metafield, it probably isn't in the description your feed app sends to Merchant Center. For the full set of page elements, see the product page optimization guide, or check a live page against our product page checklist for AI readiness.
What to leave out
Google's description attribute rules list what doesn't belong in a feed description: comparisons with other products, links, capital letters used for emphasis, promotional text about price, sales or shipping, your company history and your policies. Your own product page gives you more room. The same logic still holds for the description block itself. Describe the product. Prices, delivery terms or returns info each get their own component.
How to write a product description: structure and template
Open with a sentence that names the product type and main benefit, add four to six benefit bullets, then a short body with attributes and use cases. Specs and a FAQ close it.
Write from the facts outward. The skeleton is the same for one hero product or ten thousand SKUs.
- 1
Gather the facts
Pull the spec sheet and supplier data: composition, dimensions, certifications. Add reviews and the questions customer service gets. Every claim you write must trace back to one of these.
- 2
Define the buyer and the intent
"A hiker planning multi-day trips who gets blisters" is a buyer. "Outdoor enthusiasts" is a demographic, and it won't help you write. List the three to five searches or prompts that buyer would actually type.
- 3
Write the opening sentence
Name the product type and key attribute, then the main benefit, in 160 characters or fewer. Feeds and AI answers show this line before anything else.
- 4
Turn features into benefit bullets
Four to six bullets. Each starts with the benefit and ends with the feature that proves it: "Dry feet on long climbs: 80% merino wool wicks moisture."
- 5
Write the body
Two to four short paragraphs. State attributes as full sentences, describe use cases, explain fit or compatibility, and say who the product isn't for.
- 6
Add specs and FAQ
A scannable spec table covering materials and dimensions, weight and care, and what's included. Then three to five real questions with direct answers.
- 7
Check, then publish everywhere
Verify every number against the source and read it against your brand rules. Then push the same facts to the store and the feed, and to your marketplaces.
A lot of copywriting advice says to open with a story. We disagree for most SKUs. Nobody reads a brand story on a $14 pair of socks, and feeds cut it off anyway. Keep storytelling for hero products and luxury, after the facts.
A reusable product description template
- Opening sentence: [Product type] in [key material or feature] for [use case or buyer], with [primary benefit].
- Benefit bullets (4 to 6): [Benefit]: [feature that proves it].
- Body paragraph 1: what it's made of and how it feels or performs, with exact values.
- Body paragraph 2: when and where to use it, two or three concrete situations.
- Body paragraph 3: fit, sizing or compatibility, and who it isn't designed for.
- Specs: composition, dimensions, weight, care, included items, country of origin.
- FAQ: three to five questions shoppers actually ask, answered in a sentence or two each.
If you'd rather start from a filled-in version, our product description templates by category adapt this skeleton for fashion, beauty, home, food and automotive.
Before and after: one product, rewritten
The example uses a fictional product, the Trailmark Merino Hiking Sock. The "before" column is typical of what we find on supplier-fed catalogs.
| Element | Before | After |
|---|---|---|
| Title | Hiking Sock Grey | Trailmark Men's Merino Wool Hiking Socks, Cushioned Crew, Charcoal, Size L |
| Opening | Our best sock for the outdoors. Premium quality you can trust. | Cushioned crew hiking socks in 80% merino wool that keep feet dry and blister-free on long, multi-day hikes. |
| Attributes | High quality wool blend | 80% merino wool, 17% nylon, 3% elastane. Medium cushioning under heel and toe, flat toe seam. |
| Use cases | Great for adventures | Built for day hikes and multi-day treks in boots, from spring trails to cool autumn mornings. |
| Fit | Available in several sizes | Size L fits US men's shoe sizes 9.5 to 12 (EU 43 to 46). Arch band holds the sock in place. |
| Care | (missing) | Machine wash warm, inside out. Do not tumble dry. Odor-resistant between washes. |
| FAQ | (missing) | Are they warm enough for winter? They suit cool conditions; for snow, choose the heavy cushion version. |
Every vague phrase in the "before" column is a question nobody can answer from the page. The rewrite carries the words people search for (men's, merino, hiking socks, cushioned, crew). It also gives an agent attributes to filter on and use cases it can quote when it explains the pick.
SEO product descriptions: keywords and shopper intents
For product description SEO, use the exact terms shoppers search, put them in the title, the opening line and the FAQ, and cover the intents behind them, from use cases to fit and care. Specific, unique copy beats repeated keywords.
Keywords get a page matched. Intent coverage gets it chosen, by a shopper scanning results or by an agent assembling an answer.
Find the words shoppers actually use
Start from demand data. A brainstorm tells you how your team talks, which is rarely how customers search.
- Google Search Console, Performance report: filter Page by your product URL pattern (on Shopify, URLs containing
/products/) and read the queries that already earn impressions. Lots of impressions with few clicks usually means the title or snippet doesn't match what people typed. - Keyword tools such as Google Ads Keyword Planner: check volumes for product type plus attribute combinations ("linen shirt men", "linen shirt for wedding").
- Your site search logs: what customers type on your own store, especially the searches that return zero results. In GA4 these show up through site search tracking in enhanced measurement.
- Reviews and customer service tickets: how buyers describe the problem the product solves, and what they ask before buying. This is the most underused source.
From keywords to shopper intents
A keyword is what people type. An intent is what they need to know before they'll buy. Agents make the difference visible. Google describes a technique called query fan-out in AI Mode, where one request is split into many sub-queries. Each sub-query is an intent your description either answers or misses.
| Intent type | Example query or prompt | What the description must state |
|---|---|---|
| Product type and attribute | "merino hiking socks men" | Product type and gender, with the material, in the title and opening |
| Use case | "socks for multi-day hiking in boots" | The situations the product is made for, in words |
| Problem to solve | "hiking socks that prevent blisters" | The benefit plus the feature that delivers it (flat toe seam, cushioning) |
| Fit and compatibility | "hiking socks size 11 men" | Size chart mapping, written in the text and set in the size attribute |
| Comparison | "merino vs synthetic hiking socks" | A factual sentence on why this material, with no competitors named |
| Care and lifespan | "do merino socks shrink" | Care instructions and what to expect over time |
Where keywords go
| Placement | Guidance |
|---|---|
| Title and H1 | Primary keyword plus the attributes that define the product. Google shows about the first 70 characters of a Shopping title, so put the key terms first (Google) |
| Opening sentence | Primary keyword and the main benefit, in natural language |
| Bullets | Secondary keywords tied to features, one idea per bullet |
| Body | Use cases and long-tail phrases as full sentences |
| FAQ | Question-form searches, answered directly |
| Meta title and meta description | Primary keyword with the product type, plus a reason to click |
Product description SEO without keyword stuffing
- Use each keyword where it reads naturally. If a sentence exists only to hold a keyword, cut it.
- Cover variations through meaning. "Rain jacket" and "waterproof shell" can each appear once in context, alongside a phrase like "jacket for wet weather".
- Prefer specifics to repetition. "80% merino wool" ranks and converts better than "premium merino" three times.
- Keep keywords out of
product_highlight. Google's guidance says to avoid SEO keywords there and focus on selling benefits.
Keyword density targets are still floating around in SEO checklists. Ignore them. We've never seen a description win because it hit 2% of anything.
Duplicate product descriptions
Plenty of retailers publish the manufacturer's text unchanged, so the same paragraph sits on dozens of sites. Despite the folklore, that doesn't trigger a penalty. The cost is quieter. Google's documentation on duplicate URLs explains that it decides which pages are duplicates "based on similarity of content" and then picks one version to show. If your page says exactly what twenty others say, it's competing for that slot with nothing to set it apart.
So rewrite manufacturer copy around your buyers, with use cases, fit notes, answers to their questions and the keywords of your market. We wouldn't start with the whole catalog, though. Begin with best sellers and with pages that get impressions but few clicks, where a rewrite shows up fastest.
title
BeforeNORDVIK SHELL JACKET 2.0
AfterNordvik Women's Waterproof Rain Jacket, 2.5-Layer Packable Shell with Hood, Moss Green
description
BeforeThe Nordvik Shell Jacket 2.0 combines innovative technology and timeless design for the modern adventurer.
AfterA packable 2.5-layer rain jacket for women with taped seams and a 10,000 mm waterproof rating, made for wet commutes and day hikes. It folds into its own chest pocket and weighs 290 g in size M.
product_type
BeforeJackets
AfterApparel > Women > Outerwear > Rain Jackets
attributes
Beforecolor: green
Aftercolor: moss green; material: 100% recycled polyester; waterproof rating: 10,000 mm; fit: regular, room for a fleece
faq
Before(missing)
AfterCan I wear it over a fleece? Yes, the regular fit leaves room for a midlayer. Size up only if you plan to wear it over a down jacket.
Variants need care too. Keep the color and size variants of a product on one page, or point them to one canonical URL, so near-identical variant pages don't compete with each other. The ecommerce SEO guide covers canonicals and variant handling in depth, and structured data for ecommerce shows how to mark up the facts your description states.
Attributes in prose, tone of voice and localization
Write buying attributes as full sentences, keep one documented brand voice across the catalog, and localize with per-market keyword research instead of word-for-word translation.
Why attributes belong in the prose too
Structured fields like material, color and size power your filters and feeds. The description gives those values context: what the material means to the buyer, how the size actually fits, which situations the color suits. Agents read the product page as well as the feed, and a sentence hands them the fact along with the reason it matters.
The common advice to "just use bullets" is half right. Bullets help people skim. A bare "Material: linen" bullet still doesn't say the shirt stays cool at an outdoor ceremony in July, and that's the sentence an agent can repeat to a shopper.
| Attribute | Feed value | Sentence in the description |
|---|---|---|
material | Linen | Made from 100% European linen that gets softer with every wash and stays cool in summer heat. |
size | M | Size M fits a 38 to 40 inch chest; the relaxed cut leaves room to layer a T-shirt underneath. |
color | Sage | The muted sage green works with navy or white, and with stone chinos. |
pattern | Solid | A solid weave with no print, easy to dress up for a wedding or down for the beach. |
Baymard recommends labeling every measurement with its unit, adding a second measurement system and translating dimensions into plain language such as "fits in the palm of your hand" (Baymard). In Baymard's testing, 50% of users needed ingredient information when shopping for beauty products.
Tone of voice that survives scale
A voice is easy to hold across ten pages. Across ten thousand it only holds when it's written down as rules any writer can follow, human or AI.
Tone rules
A handful of principles, each with a do and a don't. "Confident, never boastful: say 'keeps you dry in heavy rain', not 'the ultimate rain protection'."
Reference examples
A few approved descriptions per category. Examples teach rhythm and sentence length better than any rule.
Glossary
How your brand names its materials and fits in every language, including collections and technologies. "Relaxed fit", never "loose fit".
Banned words
Words you never use. Legal risks such as "cures" or an unproven "eco-friendly", tired filler such as "high quality" and "must-have", and competitor terms.
Localization: translate the meaning, research the keywords
A translated description that uses the wrong word for the market is invisible there. A US shopper searches for a "sweater", a UK shopper for a "jumper". An Italian shopper may search "felpa" for a sweatshirt and "maglione" for a knit. Good localization starts with per-market keyword research and applies the brand glossary in each language. It also converts units and sizes, along with care conventions (US shoe sizes vs EU, inches vs centimeters).
That last part is where we see the most embarrassing errors. A size chart translated into Italian that still says inches. A care label converted to Celsius in the spec table but left at 86°F in the prose. Check numbers separately from words.
Google's description rules add a technical point: write in the target language's alphabet, and use foreign words only when they're widely understood. The guide to ecommerce localization covers the full workflow for multi-market catalogs.
How many words should a product description be?
Most product descriptions need 50 to 150 words of core copy, plus bullets, a spec table and a FAQ, with the essentials in the first 160 characters. Technical and high-consideration products need more. Facts should set the length, never adjectives.
There's no ideal word count. Long enough to cover every fact a buyer needs, short enough to scan.
Some guides still insist on 300 words minimum "for SEO". That rule does real damage. Teams hit the number by padding, and the first 160 characters, the part feeds and agents show, end up saying nothing.
What the platforms tell you
- Google Merchant Center accepts up to 5,000 characters in
descriptionand asks you to put the most important details in the first 160 to 500 characters, because shoppers must click to see the rest (Google). - Nielsen Norman Group found that concise text, about half the word count, improved measured usability by 58% compared with promotional writing, and a scannable layout improved it by 47% (NN/g).
Working ranges by product type
Rules of thumb for the core description only, excluding specs and FAQ.
| Product type | Core description | Why |
|---|---|---|
| Basics and consumables (socks, pantry staples, refills) | 40 to 80 words | Few decision factors; specs and size do most of the work |
| Fashion and accessories | 60 to 150 words | Fit and occasion need sentences, so does material; styling adds value |
| Beauty and personal care | 80 to 180 words | Ingredients and skin type, plus usage steps and a results timeline |
| Home and furniture | 100 to 250 words | Dimensions and materials, assembly, room fit, care |
| Electronics, tools, auto parts | 150 to 300 words | Compatibility, technical specs, what's in the box |
| Jewelry and luxury | 80 to 200 words | Materials and craftsmanship, sizing, the story behind the piece |
Signs a description is too short or too long
- Too short: customer service answers the same question every week, returns cite "not as described", or the page has no words for the use cases shoppers search.
- Too long: the opening line says nothing specific, paragraphs repeat the bullets, or adjectives outnumber facts.
Product description examples by industry
Each category has its own deciding attributes: fit and occasion in fashion, ingredients in beauty, dimensions in home, origin in food, materials in jewelry, compatibility in electronics. These examples put them first.
Buyers ask different questions in every category, so good descriptions look different too. All products below are fictional.
Fashion: material, fit and occasion
Ardena Linen Shirt, Sage. A relaxed-fit men's shirt in 100% European linen, made to stay cool at summer weddings, garden parties and long beach days. The breathable weave softens with every wash. Size M fits a 38 to 40 inch chest; for a closer fit, size down. Pair it with stone chinos or wear it open over a white T-shirt.
Four sentences carry the fabric and fit guidance plus three occasions. Those are the exact constraints people type into prompts like "linen shirt for a summer wedding".
Beauty: ingredients, skin type and routine
Lumea 10% Vitamin C Serum, 30 ml. A lightweight daily serum with 10% ascorbic acid and hyaluronic acid, formulated for normal, dry and combination skin. Apply two to three drops in the morning after cleansing and before moisturizer and SPF. Fragrance-free and alcohol-free. Patch test first if you have sensitive skin.
Concentration and skin types come first, then how to use it. Note what it leaves out: no medical claims, only what the formula contains.
Home and furniture: dimensions and room fit
Holm Two-Seater Sofa, Oat Bouclé. A compact two-seater for small living rooms and studio apartments: 160 cm wide, 85 cm deep, 78 cm high (63 x 33.5 x 30.7 inches), seat height 45 cm. Solid oak legs, removable bouclé covers, machine washable at 30°C. Arrives in two boxes and assembles in about 15 minutes with no tools.
Both unit systems are there. It answers "will it fit" and "how hard is it to set up" before anyone asks.
Food and drink: origin, taste and use
Colle Verde Extra Virgin Olive Oil, 500 ml. Cold-extracted extra virgin olive oil from Coratina olives grown in Puglia, Italy, harvested in October. Peppery finish with notes of green almond and artichoke. Best raw on bean soups, grilled vegetables and bread. Store away from light and use within 12 months of opening.
Origin and cultivar come first, with the harvest month. An agent can match it to "peppery olive oil for drizzling" without guessing.
Jewelry: materials, sizing and craftsmanship
Mira Signet Ring, 18k Gold Vermeil. A slim signet ring in 18k gold vermeil over sterling silver, with a 9 mm oval face ready for engraving up to three initials. Available in US sizes 5 to 9. Hypoallergenic and nickel-free. Remove before swimming to keep the finish bright.
Base metal and plating are both named. Shoppers who know vermeil from plain gold plating look for exactly that line.
Electronics and auto parts: compatibility first
Voltline Magnetic Car Phone Mount. A vent-clip magnetic mount compatible with iPhone 12 and later with MagSafe, and with other phones using the included metal ring. Fits horizontal and vertical vents up to 15 mm thick. Holds phones up to 280 g on rough roads. Not compatible with round vents.
The last sentence matters most. A clear "not compatible" line prevents returns and lets an agent exclude the product when it won't fit.
The ACO industry playbooks go further, with attribute lists and examples for five categories, from fashion and jewelry to food.
Writing product descriptions at scale with AI
AI writes good product descriptions at scale only when it has complete product data, written brand rules and a check before publishing. Pick a tool by catalog size and data sources, then by languages and how it publishes.
Without those three things you get fluent text with invented details, which is worse than no description at all.
The failure we see most is small. A dry-clean-only shirt gets described as machine washable because the model filled a gap, and nobody notices until the returns come in.
AI is already the default writer
Shopify reports that in its November 2025 merchant survey 75% of business owners used AI tools, and content generation was the most common use case at 69% (Shopify).
Google's guidance on generative AI content is explicit on three points:
- Generating many pages "without adding value for users may violate Google's spam policy on scaled content abuse."
- Publishers should "manually factcheck and review all AI-generated content for accuracy and trustworthiness before publishing", including titles and meta descriptions, structured data, and alt text.
- In Merchant Center, AI-generated product data such as titles and descriptions must be labeled separately. The description attribute offers
structured_descriptionfor this, with adigital_source_typesuch astrained_algorithmic_media.
Best AI product description generator: what to compare
Teams spend weeks tuning prompts. In our experience that's the wrong place to look: a perfect prompt can't recover a material that was never in the source file.
A generic writing assistant, or our free product description generator, is fine for a handful of products. Thousands of SKUs in several languages, with a feed to maintain, need a platform that reads your data, follows your rules and publishes back. Test every option on the same 20 real products and score the output on accuracy and attribute coverage, then on brand voice. That's the method behind our comparison of the best AI product description generators.
| Criterion | Manual copywriting | General AI writing assistant | Catalog AI platform |
|---|---|---|---|
| Best for | A few hero products | Small catalogs, one-off drafts | Hundreds to thousands of SKUs, many markets |
| Source of facts | Writer's research | Whatever is pasted into the prompt | Imports from PIM, store, spreadsheets, spec sheets, feeds |
| Keyword and intent data | Manual research | Usually none unless provided | Built into the workflow per market |
| Brand voice | Strong, depends on the writer | Varies prompt to prompt | Enforced with tone rules, examples, glossary, banned words |
| Accuracy checks | Human review | Human review of every output | Automated checks plus human approval |
| Languages | One writer per language | Good translation, generic keywords | Localized with per-market keywords and glossary |
| Publishing | Copy and paste | Copy and paste | Pushes to store, PIM, Merchant Center |
The four guardrails of AI product copy
Data in
Does the AI have every fact it needs?
Every fact imported per SKU (spec sheets, attributes, supplier data, reviews, FAQs), with gaps flagged rather than guessed
Brand rules
Does it sound like the brand?
Tone of voice and examples, plus the glossary and banned words, applied to every output
Checks
Is it accurate and complete?
Automated checks catch invented facts and missing attributes, flag banned words and score keyword coverage page by page
Approval and publishing
Who signs off, and where does it go?
Human review for low scores, auto-approval above a threshold, then publishing to every channel
Skip one and scale multiplies errors instead of results.
Product descriptions with AndromedAI
AndromedAI creates and rewrites product descriptions from your product data, applies your Brand Kit, checks every page and publishes to your store and PIM, and to Google Merchant Center, in up to 12 languages.
We built AndromedAI to turn product data into complete, on-brand descriptions that rank and convert, and that AI agents pick. It has optimized more than 500 catalogs.
From data to published description
| Step | AndromedAI | What it does |
|---|---|---|
| Score your pages | AI Readiness Audit | Scores product pages on Product Data Completeness, Keyword Coverage, Customer Intent Match and Shopping Metadata, and shows what is missing |
| Create new descriptions | Creator | Creates complete product pages from brand or supplier data, in 12 languages |
| Rewrite existing descriptions | Optimizer | Rewrites titles, descriptions, bullets and FAQ, extracts attributes, and adds use cases and shopper intents |
| Cover category demand | Category page optimizer | Builds category pages around real search demand |
| Publish everywhere | Integrations | Imports from CSV, Excel, Google Sheets, Shopify, Akeneo, SAP, PDF spec sheets, XML or JSON feeds and REST API; publishes to Shopify, Google Merchant Center, Salesforce Commerce Cloud, Adobe Commerce, Shopware, Akeneo, Plytix and WooCommerce |
Guardrails built in
Every output follows your Brand Kit: tone of voice and rules, examples, glossary, banned words. AI checks score each page. Approval workflows send anything uncertain to your team, and pages scoring above 4.0 on the AI Checker can be approved automatically.
AndromedAI also generates Merchant Center conversational attributes, so the facts in your descriptions reach Google's AI surfaces as structured data too.
Descriptions are written from your data and checked before they go live, so scale never comes at the cost of accuracy.
Not sure where descriptions fit in the wider catalog? Start with product feed optimization or the guide to getting products recommended by ChatGPT.
to the first sales from ChatGPT, and -95% catalog generation costs
Matassasales and +160% organic traffic, with 483 hours saved
Semprefarmaciaclicks from AI chats
Bomboogieadd-to-cart in two months
Altaforma Milanobounce rate on product pages
AusiliumFAQ
Start from facts: specs, dimensions, materials and the questions customers ask. Open with a sentence naming the product type, key attribute and main benefit, then add four to six benefit bullets. Follow with a short body that states attributes in full sentences and describes use cases, then a spec block and a short FAQ. Use the words buyers search with and check every number against the source.
Start with what the product is and its primary benefit. Then cover materials or ingredients, dimensions and fit, compatibility, use cases, who it's for, what's included and care instructions. Answer the most common pre-purchase questions too. Leave out prices, shipping, store policies and comparisons with named competitors.
There's no fixed number. Most products need 50 to 150 words of core copy plus bullets, a spec table and a FAQ; technical and high-consideration products often need 150 to 300. Put the most important facts in the first 160 characters. Feeds and AI answers often show only the opening.
SEO product descriptions use the primary keyword in the title and opening sentence and cover the intents behind related searches, such as use cases, fit or care. They're unique to your site and backed by matching structured data. Specific facts do more than repeated keywords.
Google groups pages with very similar content and usually shows one version. If your page uses the same manufacturer text as many other sites, it competes for that slot with nothing to distinguish it. Rewriting descriptions around your buyers and their use cases, with your market's keywords, gives your page a reason to be chosen.
Agents such as ChatGPT, Google AI Mode and Amazon Rufus break a shopper's request into constraints, then look for products whose data confirms each one. They read descriptions and attributes, along with feeds, to match and justify a recommendation, so a fact that isn't written down can't be used.
It depends on catalog size. General AI writing assistants work for a few products, but hundreds or thousands of SKUs need a catalog platform that imports your product data and applies brand rules such as tone, glossary and banned words. It should also check accuracy, handle multiple languages with per-market keywords and publish to your store and feeds. Test any tool on 20 of your real products.
Yes, if the content is accurate and adds value. Google says AI content produced at scale without value can count as scaled content abuse, and asks publishers to fact-check AI output, including metadata, before publishing. In Merchant Center, AI-generated descriptions should be labeled using structured_description.
Keep a brand glossary in every language for how you name materials and fits. Add a banned-words list per market and approved examples, and do keyword research for each market. Adapt units and sizes to each country, along with everyday vocabulary, instead of translating word for word.
Glossary
- Product description
- The text on a product page and in a product feed that explains what a product is, what it's made of, how it fits or works and who it's for
- PDP
- Product detail page, the page dedicated to a single product
- Product attribute
- A structured value describing a product, such as material, color, size or pattern
- Product highlight
- A Merchant Center attribute for short benefit statements, 2 to 100 per product and up to 150 characters each
- Product feed
- A structured file of product data sent to channels such as Google Merchant Center or ChatGPT
- Conversational attributes
- Optional Merchant Center attributes, such as question_and_answer, that help AI systems understand product nuances
- structured_description
- A Merchant Center attribute for labeling AI-generated descriptions with their digital source type
- Shopper intent
- The need behind a search or prompt, such as a use case, a fit question or a problem to solve
- Query fan-out
- The technique by which an AI system splits one request into many sub-queries
- Keyword stuffing
- Repeating keywords unnaturally to influence rankings, which makes text harder to read and less trustworthy
- Duplicate content
- Text that appears on several pages or sites, such as unchanged manufacturer descriptions
- Canonical URL
- The version of a page you want search engines to treat as the main one among duplicates
- Brand Kit
- A written set of tone of voice rules and examples, plus a glossary and banned words, that guides AI writing
- AI Readiness Score
- AndromedAI's score of how ready a product page is to be recommended by AI agents
Keep reading
Draft a description from your product data in seconds
GuideBest AI Product Description Generators (2026): Tested on Real CatalogsHow the main tools compare on accuracy, attributes and brand voice
GuideProduct Description Templates by Category (Free)Copyable templates for fashion, beauty, home, food and automotive
GuideProduct Page Optimization: The Complete Guide to PDPs That Rank, Convert and Get Recommended by AIEvery element of a PDP that ranks, converts and gets picked by AI
Sources (15)
- Google Merchant Center Help: Description [description]
- Google Merchant Center Help: Title [title]
- Google Merchant Center Help: Product highlight [product_highlight]
- Google Search Central: Guidance on using generative AI content
- Google Search Central: Consolidate duplicate URLs
- OpenAI Commerce: Product feed reference
- Search Engine Roundtable: Google Merchant Center conversational attributes
- Baymard Institute: Product description research
- Nielsen Norman Group: How users read on the web
- Salsify 2025 Consumer Research Report
- Salsify 2026 Consumer Research
- Adobe AI Traffic Trends Report, August 2026
- Modern Retail: Amazon Rufus users up 115%
- Luxury Daily: McKinsey and ICSC, Shopping in the Age of AI
- Shopify: How to write product descriptions that sell
Parlano di noi
