Discussion starter: an ecommerce SEO question, posted with an answer from Niraj Raut to open the thread. If you have dealt with this on a site, reply with what you saw, especially where it differs.
- eCommerce SEO
Manufacturer copy duplicated across 50 retailers: how much rewriting did you need before rankings moved, and was AI rewriting enough?
Disclosure: Niraj Raut, who posted this answer, runs the SEO consultancy linked at the end of it.
1 reply
-
adminAdmin
There is no rewriting threshold after which product pages start to rank, and no public evidence points to one. Rewording manufacturer copy, whether by hand or with AI, mostly changes the words and leaves the information the same. Google filters near-identical results rather than penalising them, so when 50 retailers carry the same description, that description is rarely what separates the retailer that ranks from the other 49. The deciding factors sit elsewhere: the retailer’s authority in the category, price and availability, product data quality, reviews, and whatever the page adds that nobody else has.
So AI rewriting on its own is usually not enough, and at catalogue scale with nothing added it drifts towards a pattern Google’s spam policies describe. A more useful question than “how much do I rewrite?” is “which SKUs can I realistically win, and what information can I add to those pages that the other 49 retailers cannot?”
Why rewording the description rarely moves a product page
Duplicate text across retailers is handled in two broad ways. At indexing time, Google groups duplicate or near-duplicate pages and picks one canonical per cluster; Gary Illyes described this on the Search Off the Record podcast in 2020 as reducing the main content to checksums and clustering pages whose checksums match or come close. Fifty retailer pages rarely collapse into one cluster, because each has its own template, price, stock, reviews and navigation. The competition happens at ranking time instead, where Google also tends to avoid showing several results that look near-identical for the same query.
In that competition the shared description is a small part of what differs between the pages. For a model-number query, the manufacturer’s own site and the largest marketplaces often hold the top positions whatever the copy says, and the remaining slots go to retailers with stronger category signals, better prices or better product data. A reworded paragraph gives Google almost nothing new to evaluate. Unique copy matters more for long-tail attribute queries (“quiet dishwasher for open-plan kitchen”, “compatible with model X”), but only when the new copy contains information those searchers want.
A useful way to picture it:
Fifty bookshops sell the same novel with the same back-cover blurb. Rewording the blurb in your window does not make yours the shop people choose. A staff review, a signed copy, same-day delivery or a reputation for the genre might. Search ranking among retailers works in a similar way: what separates the pages is the information and signals beyond the shared text, not a different arrangement of it.
What AI rewriting changes, and what it doesn’t
Google’s scaled content abuse policy lists “Using generative AI tools or other similar tools to generate many pages without adding value for users” as an example, and separately describes scraping content to generate pages through “automated transformations like synonymizing, translating, or other obfuscation techniques”. A retailer using licensed manufacturer copy on real, purchasable product pages is not what that policy targets, and using AI is not a violation in itself. The practical risk of a mass paraphrase is that it costs money and editing time, adds nothing Google can use to prefer your page, and can introduce factual errors.
Approach What changes New information for searchers Main risk Manufacturer copy as supplied Nothing None You stay one of 50 near-identical pages AI paraphrase of manufacturer copy Wording only None Spec errors, cost, a scaled low-value pattern across the catalogue Human rewrite of manufacturer copy Wording and tone Little, unless the writer knows the product Cost with a weak payoff AI-assisted drafting from your own data (reviews, return reasons, customer questions, fitment data) Information Real Needs fact-checking and a clear source for each claim Structured unique modules (compatibility tables, sizing guidance, comparison with the previous model) Information and query coverage Real Engineering and editorial effort up front Watch for this:
AI rewrites of product copy can change specifications: voltage, dimensions, materials, compatibility. On a product page that is a consumer law and returns problem, not just an SEO one. Keep specifications coming from the structured product feed, and let any generated prose reference those fields rather than restate them freely.
What tends to separate the retailer that ranks
Google does not publish the weight of each factor for product pages, so treat the following as the areas worth investing in rather than a ranked list.
Information the manufacturer didn’t write
Compatibility and fitment, sizing guidance drawn from return reasons, real-use photos, answers to the pre-sale questions your customer service team keeps getting, and how the model differs from the one it replaces.
Reviews and questions
Genuine reviews and Q&A add unique text that grows over time and answers the questions buyers search for. They only count if they are real.
Product data quality
GTINs, price, availability, shipping and returns in structured data and in Merchant Center. Google requires a price on every Offer for merchant listing eligibility, so do not strip prices to work around data problems.
Commercial position
Price competitiveness, stock and delivery speed shape merchant listing visibility and which result a searcher clicks. No amount of copy fixes an uncompetitive offer.
Category strength
Internal links from well-linked category pages, and a site that is recognisably a specialist in the category, carry more weight on generic product queries than the copy on one page.
How much work, and on which SKUs
Treating the whole catalogue the same is where most budgets go. A tiered approach puts effort where a ranking change is plausible and commercially worth having.
Tier Which SKUs Treatment A Real search demand, and you can compete on price, stock or reviews Full enrichment: unique modules, edited copy written by someone who knows the product, original images, reviews prompted after purchase B Moderate demand, or strong category fit Templated unique modules generated from your own data, with manufacturer copy kept or lightly edited C Long tail with little demand Manufacturer copy, clean data, solid category linking. Spend nothing more until demand shows up in Search Console Hypothetical example:
A power-tools retailer carries 12,000 SKUs, most sold by dozens of other stores with the same copy. Search Console shows that about 800 SKUs earn most product-page impressions, and the retailer is price-competitive on roughly half of those. Tier A is those 400 SKUs. Tier B is the next 2,000, where a compatibility module built from the battery-platform data and a “what’s in the box” check from the warehouse add real information at template cost. The remaining 9,600 stay as they are. A mass AI paraphrase of all 12,000 would cost more than the Tier A work and add less.
How to find out whether AI rewriting is enough on your catalogue
The question you asked can be answered directly with a three-arm test, which beats relying on anyone’s anecdote, because product-page results depend heavily on the retailer’s position in its market.
- Select several hundred SKUs from the same categories with similar demand, similar current positions and stable prices, and split them into three matched groups.
- Arm 1: AI paraphrase of the manufacturer copy only. Arm 2: unique modules built from your own data, manufacturer copy kept. Arm 3: no change (control).
- Primary metric: organic clicks per group against the control, from Search Console.
- Secondary metrics: impressions on model-number and attribute queries, the Merchant listings and Product snippets search appearances in Search Console, conversion rate, and revenue per session.
- Duration: six to ten weeks. Long-tail product pages are recrawled slowly, so check in URL Inspection or logs that the changed pages were actually recrawled before reading results.
- Exclude SKUs that go out of stock, change price or join a promotion during the test, and note any core update that lands mid-test.
If Arm 1 matches the control and Arm 2 beats it, you have your answer on paraphrasing and a case for the enrichment budget. If both match the control, the constraint is probably commercial or site-level (price, authority, category strength) rather than content, and copy work of any kind is the wrong lever. If Arm 1 beats the control, check whether the old copy had a real problem, such as missing key attributes or truncated text, that the rewrite fixed by accident.
Where I would start
If this were my catalogue, I would not commission a rewrite of anything until I knew which SKUs had demand I could realistically win. For those, I would add information that comes from the business itself: return reasons, customer questions, fitment data, warehouse checks and reviews. AI is useful for turning that data into readable text at scale, provided every claim traces back to a field someone can check. Rewording the manufacturer’s paragraph would be the last thing I spent money on, because it is the one change that gives Google nothing new to prefer.
Need help with this on your own site?
Niraj Raut works with businesses in Nepal, Australia, the UK, Europe and the US on technical SEO, ecommerce SEO, local SEO and AI search.
Add a reply Cancel reply
Replies are open to approved members. Log in or apply to join.