Product descriptions are usually treated as filler between the photograph and the buy button, written last, quickly, and frequently copied from whatever the manufacturer supplied.
That is a mistake with two measurable costs. The page does not rank, because every competitor selling the same item published identical text and there is nothing to distinguish yours. And the item comes back, because the description did not tell the buyer what they needed to set expectations correctly.
Both are fixable with a structure rather than with creative writing.
Descriptions for marketplaces versus your own store
If you sell on marketplaces as well as your own store, the temptation is to write once and publish everywhere. That is usually the wrong call and the reason is structural.
Marketplace listings compete inside a search result where dozens of near-identical products sit side by side, so the first line and the specification block carry almost all the weight. Buyers scan rather than read, and the platform frequently truncates anyway. Write those tight, lead with the distinguishing specification, and accept that brand voice has little room.
Your own store is a different surface. The visitor arrived deliberately, usually from search, and is further along. There is room to answer the questions a marketplace listing cannot: who it does not suit, what it is like to use, what happens if it is wrong. That is also where your search relevance lives, since the marketplace listing ranks on the platform rather than on Google.
Publishing identical text in both places wastes the second surface and, on your own store, means the page competes with a marketplace listing of the same words on a domain with far more authority than yours. You will lose that comparison.
Measuring whether the rewrite worked
Product description work is unusually measurable, which is a reason to treat it as an experiment rather than a content project.
Establish the baseline before you change anything: organic sessions to the product page, conversion rate on that page, and return rate on that product with reasons where you capture them. Take a three-month window if volume allows, because a single month on a low-volume product is noise.
Then rewrite a set of products rather than one, so you have enough signal, and leave a comparable set untouched as a control. Products of similar price, category and traffic, changed at the same time.
What you are looking for after six to eight weeks is movement in three numbers. Organic sessions rising means the description is now distinguishable enough to rank. Conversion rate rising means it answers what the buyer needed. Return rate falling means expectations are now being set correctly, and this is usually the largest financial effect because a return costs the outbound shipping, the return shipping, the handling and frequently the resale value.
If none of the three moves, the description was not the constraint on that product, and the next thing to examine is photography or price positioning rather than more copy.
In practice
Why manufacturer copy fails
Using supplied manufacturer text is the single most common reason product pages fail to rank, and the reason is simple: dozens of retailers published the same words, so no page has a claim to being the useful one.
Search engines are not penalising you for duplication in any punitive sense. They are choosing which of the identical pages to show, and that choice goes on other signals, which for a smaller retailer means it does not go to you.
The second problem is that manufacturer copy is written to sell the product in general rather than to help a specific buyer decide. It is full of capability claims and empty of the practical detail a buyer needs: dimensions, weight, what is in the box, what it does not do, and who it is not suitable for.
That last category matters disproportionately. Telling a buyer honestly that something is unsuitable for a particular use prevents a return, and returns cost you the outbound shipping, the return shipping, the handling, usually the resale value and frequently the customer.
What to include, in order
A structure that serves both search and the buyer, roughly in the order a buyer wants it.
Writing at catalogue scale
The objection to all of this is volume: nobody is writing five hundred words for four thousand products, and they should not try.
Prioritise by demand. A small proportion of products carry most of the search volume and most of the revenue. Those get full treatment. The long tail gets a structured template populated with accurate specifications, which is still far better than manufacturer text because the specifications are genuinely useful and at least partly unique.
AI drafting is genuinely useful here and needs a human pass. Generating a first draft from your specification data and then editing is several times faster than writing from scratch, and unedited AI product copy is recognisable, repetitive and does your brand no favours.
Use your own returns data to decide where to invest. Products returning at an above-average rate for reasons related to expectation are telling you exactly which descriptions are failing, and fixing those has a calculable return.
And revisit the top pages periodically. The questions customers ask in support are the questions the description did not answer.
In practice
Key takeaways
- A description has two jobs: get the page found, and stop the item coming back.
- Manufacturer copy is the most common reason product pages fail to rank; everyone published the same words.
- Naming who the product does not suit prevents returns and builds more trust than any claim.
- Specifications prevent returns. For apparel, measurements per size rather than a size label.
- Prioritise by demand and by return rate rather than trying to rewrite the whole catalogue.
- AI drafting plus a human pass is fast; unedited AI copy is recognisable and unhelpful.
Frequently asked
You can, and it is the most common reason product pages fail to rank. Dozens of retailers published identical text, so search engines choose between them on other signals, which rarely favours a smaller retailer. It is also written to sell in general rather than to help a specific buyer decide, so it omits the practical detail that prevents returns.
Long enough to answer what a buyer needs and no longer. For a simple item that may be a hundred words plus complete specifications. For a considered purchase it may be four hundred. Length itself is not a ranking factor; completeness of the answer effectively is, because a page that fails to answer sends the visitor back to search.
Yes, as a first draft from your specification data, with a human editing pass. That is several times faster than writing from scratch and produces usable copy. Unedited AI product text is recognisable, repetitive across your catalogue, and does your brand no favours, which undermines the point of writing unique descriptions at all.
Those carrying the most search demand and revenue, then those with above-average return rates. A high return rate for reasons related to expectation tells you precisely which description is failing, and fixing it has a calculable return in avoided shipping, handling and lost resale value.