To write a product description that converts, answer the question the images could not. Shoppers scan rather than read, so structure it as labeled highlights, not prose. Then write it twice more. The Google Shopping feed does not accept the urgency and promotional text that work on the page, and Amazon sets bullet limits by product type. One product, three versions.
The paragraph on your product page, the description in your Google Merchant Center feed, and the bullet points on your Amazon listing are three separate assets with three separate rulebooks. Page copy sent unchanged to the other two can get them flagged or edited.
That is the expensive part. A brand writes one lovely description full of urgency and brand personality, sends it everywhere, and then wonders why Merchant Center shows warnings and Amazon edits the bullets.
This covers what each surface accepts, how to write the on-page version so people read it, and how to stop writing the same sentence three times.
Key Takeaways
- Shoppers scan, they do not read. In a 1997 Nielsen Norman Group study, 79% of test users always scanned any new page, and only 16% read word by word (nngroup.com, accessed September 2026). Write for the scan.
- Highlights beat paragraphs. Baymard Institute found descriptions structured by feature highlights lead to deeper exploration. In its 2018 benchmark, only 22% of sites did it (baymard.com, accessed September 2026).
- Google’s feed rules block much of the usual copywriting advice. Its description attribute rules say not to include price, sale dates, shipping, your company name, links, all-caps emphasis or comparisons to other products (support.google.com, accessed September 2026).
- Front-load the feed. Google allows up to 5,000 characters in the description and recommends the most important details in the first 160 to 500 (support.google.com, September 2026).
- Marketplaces set their own limits. Amazon asks for at least three bullet points and says character limits vary by product type (sellercentral.amazon.com, accessed September 2026).
- The description answers what the gallery could not. Fix the image set first and the description gets shorter and works harder.
How to write a product description that converts
A product description converts when it removes the specific doubt that stops the purchase. That doubt is almost never “what is this product”. It is size, fit, material, compatibility, what is in the box, or what happens if it is wrong. Shoppers arrive at the description after the images, carrying one unanswered question. The description that converts answers that question in the first two lines.
Baymard Institute’s benchmarking found that 10% of ecommerce sites have product descriptions that are insufficient for users’ needs (baymard.com, 2021, accessed September 2026). In separate testing, Baymard found users often called descriptions “bland” or “boring”, and some spent less time on products that suited them (baymard.com, 2018, accessed September 2026).
So the job is narrow. Not “describe the product”. Answer the objection the pictures left open.
Why shoppers never read the description you wrote
Shoppers scan a product description instead of reading it. Nielsen Norman Group’s 1997 research on how people read web pages found that 79% of test users always scanned a new page and only 16% read word by word. A 2008 NN/g analysis found users have time to read at most 28% of the words on an average page, and 20% is more likely (nngroup.com, accessed September 2026).
The same analysis found the share drops fast as text gets longer. On an average visit, users read half the information only on pages of 111 words or less. A 180-word product description is already past that point. If your differentiator sits in the middle of the third sentence of a paragraph, the scan can miss it.
This is why the feature-versus-benefit debate matters less than people think. A benefit buried in prose and a feature buried in prose both go unread. Position beats phrasing.
Structure the page description as highlights, not paragraphs
Break the description into labeled highlights instead of a block of text. Baymard’s testing found users given highlight-structured descriptions performed more in-depth exploration than users given conventional text blocks. Its 2018 benchmark found only 22% of sites structure descriptions this way, even on their top five selling products (baymard.com, accessed September 2026).
In Baymard’s testing, a highlight pairs one feature with an image, graphic or icon, a short header and a short text. Feature icons did less well than images, but they still did much better than text-only descriptions. If you cannot add an icon per feature, a bold label plus one line is the text-only fallback. The table shows that fallback.
| Weak | Label plus one line |
|---|---|
| ”Crafted from premium stoneware with a durable glaze, this mug is perfect for your morning routine.” | Material: Stoneware, 320ml, 0.4kg |
| ”Available in a range of beautiful colors to suit any kitchen.” | Finish: Reactive glaze, so no two are identical |
| ”Dishwasher safe for your convenience.” | Care: Dishwasher and microwave safe |
The right column is shorter, less charming, and answers more questions per second of attention. That is the trade the scan forces on you.
Keep the page description between roughly 60 and 200 words of actual highlight content. Longer is fine below the fold for the minority who expand it, but everything load-bearing goes above.
Write the description against your images, not a blank page
Open your gallery before you open the text field. Every question the images already answer is a sentence you do not need to write, and every question they leave open is a line the description owes the shopper.
A gallery that shows scale, texture, the product in use and one frame per variant retires most of the description’s workload. A gallery of five near-identical front-on shots pushes all of it back onto the text, where most readers skim past it. We covered the sequence that works in what to show in a PDP gallery and in what order.
For most catalogs that is the higher-return fix. Adding a scale shot and a detail crop often does more than another rewrite of the same paragraph, because it moves the answer from a surface people scan to a surface people look at. AI product photography in DesignerBox starts that set from one photo of the real product, and a multi-angle template covers the multi-view part.
Build the shot list once and it stops being a per-product decision. You set your brand rules once and the workflow reads them on every run. Save the set you approved as a workflow, and it runs the same way on the next SKU, so each new gallery answers the same questions as the one you signed off. Batch runs one workflow over a whole sheet of products. The description work shrinks by the same amount every time.
The full workflow from the first product photo to the finished ad, in one subscription. Two limits worth knowing before you plan around it: the commercial license starts on the Pro plan, and AI video and virtual try-on start on the Premium plan. Plans and credits are on the pricing page.
Before you commit an AI image to a listing, run the accuracy checks in our guide on what to verify in an AI product photo. A description that contradicts the picture costs more than a weak one.
Google Shopping feed rules for descriptions
The description you submit to Google Merchant Center follows different rules to the one on your page, and the standard copywriting checklist violates several of them.
Google’s product data specification tells you not to include promotional text in the description attribute, naming price, sale price, sale dates, shipping, delivery date, other time-related information and your company’s name. It also says not to include links to your store, capital letters used for emphasis, or comparisons and details about other products. Descriptions should cover the product itself, not accessories, business history or store policies (support.google.com, September 2026). Not following these rules can lead to disapprovals or limit how your ads serve.
One newer rule matters if AI writes your copy. Google says descriptions created with generative AI must go in the structured description attribute, marked as trained_algorithmic_media, instead of the plain description attribute (support.google.com, September 2026). Whatever tool drafts the copy, check every claim against the real product. AI tools for dropshipping lists the tools that write product copy and the check each one needs.
Read that against the usual advice. “Create urgency” is time-related information. “Use social proof” tends to arrive as review counts and superlatives. “Show your brand personality” is where the company name and the all-caps emphasis creep in. “Add a call to action” is a link or a store reference. Four common tips, four ways to trip the spec.
The length rules also differ from the page. Google accepts up to 5,000 characters and recommends the most important details in the first 160 to 500 characters, because customers must click to see the longer description (support.google.com, September 2026).
What the feed version looks like: plain, factual, front-loaded, no brand voice. Size, material, fit, contents, compatibility, in that order. It reads like a spec sheet because that is what it is.
Marketplace listings run on hard character caps
Amazon is a third rulebook again. Its bullet point requirements ask for at least three bullet points and say character limits vary by product type. Amazon asks for a header with a colon, then the detail, as in “Cotton fabric: Made from 100% cotton”. Bullets with all caps, emojis, special characters, guarantee wording, company information or website links may be removed or updated (sellercentral.amazon.com, accessed September 2026). Check the limit for your product type in Seller Central before writing to a number.
The practical effect of a short bullet is one claim per bullet, stated plainly. There is no room for a setup sentence. Lead each bullet with the attribute, then the detail.
Marketplaces also have their own rules on AI-generated imagery and on what a listing photo may show, which is a separate compliance question from the copy. Our breakdown of what each marketplace requires for AI product images covers Amazon, Etsy, Walmart, eBay and TikTok Shop.
What AI shopping surfaces read
Google’s shopping surfaces work from the product data you submit: the feed attributes, the price, availability and return policy. Google says the description attribute can be a foundational input for its AI-powered ad experiences and formats (support.google.com, September 2026). A well-written paragraph on the page does less if the feed behind it is thin or non-compliant.
That points to the same conclusion from a different direction. The feed version of the description is the version that machines read. Google says a product with no description gets a warning and may see limited performance (support.google.com, September 2026).
Keep both accurate and keep them consistent on the facts. Google asks for the description to match the rest of your product data and your landing pages. A size stated one way on the page and another way in the feed is the kind of mismatch that produces returns. The other jobs a store keeps for AI shopping are in the guide to agentic commerce optimization.
One product, three versions: the working checklist
Write the facts once, then cut three versions from them.
- Build the fact block. Dimensions, weight, material, fit, contents, compatibility, care, country of origin. Plain values, no sentences. This is the source for all three versions.
- Audit the gallery. Mark every fact the images already answer. Those drop down the priority list for the page version.
- Write the page version. Bolded highlight labels plus one line each. Lead with the fact that answers the most common objection. Keep the load-bearing content above the fold.
- Write the feed version. Strip brand voice, urgency, company name, links, all-caps and comparisons. Front-load the first 160 to 500 characters with size, material and fit.
- Write the marketplace version. At least three bullets on Amazon, one claim each, attribute first, inside your product type’s character limit.
- Check them against each other. Same numbers, same names, same variant labels in all three.
Steps 1 and 2 take the time. Steps 3 to 5 are the quick part once the fact block exists, which is the whole argument for building it first.
If you want to know whether the changes worked, run the traffic math before you run the test. Most stores cannot detect a real lift at their session volume, which we cover in ecommerce image testing.
Images first, then the description
Every question your gallery answers is a line the description does not have to carry, and images get read where text gets skimmed. Build the full set from one photo of the real product, then write the three versions against what is left.
Start from a template, add your brand and your products, and run it. The cost is shown before the run. See the templates.
FAQ
How long should a product description be?
There is no single answer because there are three descriptions. On the page, aim for 60 to 200 words of highlight-structured content, with the load-bearing facts above the fold. In the Google Merchant Center feed, the limit is 5,000 characters, and Google recommends the important details in the first 160 to 500 (support.google.com, September 2026). On Amazon, at least three bullets, with a character limit that varies by product type.
Should product descriptions focus on features or benefits?
Both matter less than position. In a 1997 Nielsen Norman Group study, 79% of test users always scanned a new page (nngroup.com, accessed September 2026), so a benefit buried in a paragraph is as invisible as a feature buried in a paragraph. Put the fact that resolves the shopper’s main doubt in the first highlight, then decide how to phrase it.
Can I use the same product description on my site and in Google Shopping?
Only if the page version already follows Google’s rules. Google’s description attribute rules say not to include price, sale dates, shipping information, your company name, links, all-caps emphasis or comparisons to other products (support.google.com, September 2026). Urgency, social proof and CTAs usually break at least one of those.
Do I need a different product description for Amazon?
Yes. Amazon asks for at least three bullet points, with character limits that vary by product type, so a paragraph written for your own site does not fit the field. Write one claim per bullet with the attribute first, and check Seller Central for your product type’s limit.
Does the product description need to repeat what the images show?
No, and repeating them wastes the attention you have. Audit the gallery first, then write the description to cover what the images left open. If the gallery answers scale, texture and use, the description gets shorter and converts harder.
Does AI-written listing copy hurt SEO?
Thin, duplicated or inaccurate copy is the risk, whoever wrote it. Accuracy against the actual product is the thing to check. One feed rule does depend on how the copy was made: Google requires AI-written descriptions to go in the structured description attribute, marked as made with generative AI (support.google.com, September 2026).
How many bullet points should a product listing have?
On your own site, Baymard notes that a highlights structure is often used for 2 to 6 core features (baymard.com, 2018). List the rest in a plain spec list below them. On Amazon, the requirement is at least three bullet points.
Sources
- Google Merchant Center Help, description and structured description attributes (support.google.com, accessed September 2026)
- Amazon Seller Central, product bullet point requirements (sellercentral.amazon.com, accessed September 2026)
- Baymard Institute, structuring product page descriptions by highlights (baymard.com, April 2018, accessed September 2026)
- Baymard Institute, 10% of ecommerce sites have product descriptions that are insufficient (baymard.com, March 2021, accessed September 2026)
- Nielsen Norman Group, How Users Read on the Web (nngroup.com, September 1997, accessed September 2026)
- Nielsen Norman Group, How Little Do Users Read? (nngroup.com, May 2008, accessed September 2026)
- DesignerBox plan gating and pricing (DesignerBox pricing page, designerbox.ai/pricing, September 2026)
Product description requirements verified from Google’s Merchant Center product data specification and Amazon’s bullet point requirements as of September 2026. Amazon character limits vary by product type and change; confirm in Seller Central before writing to a number. Individual results vary.
