Most advice on keywords in product descriptions assumes the text is yours to write. If your catalog comes from a supplier feed, it isn’t – and neither is it exclusively yours to publish. This article covers what keyword optimization looks like when the source text arrives with the feed, the catalog runs into thousands of items, and rewriting product by product is not an option.
The problem isn’t only scale – it’s that the text isn’t yours
Advice on optimizing product descriptions usually starts the same way: research what your competitors rank for, then work those phrases into your copy. With fifty products that is sound. With three thousand it stops being advice and becomes a multi-month project.
But scale is only half of it. A shop selling from a supplier feed has a second problem, and it is the one that actually decides how the page performs. The description arriving in the feed reaches every shop selling that product, in exactly the same form. You and your competitors publish identical text under different addresses. No amount of keyword placement fixes that, because the phrase is not the issue – the duplication is.
So the question worth answering is not "how do I write a good product description". It is "how do I produce three thousand of them without publishing three thousand copies of somebody else’s text".
Start with the category, not the product
The one shift that makes catalog-wide optimization possible: the unit of work stops being the product and becomes the category.
Categories and products carry different things. A category carries buying intent – someone is looking for hiking backpacks, waterproof jackets, gel nail polish. That is the level where the real competition for search results happens, and where you decide which phrases you want to be visible for. A product carries specifics: capacity, material, color, size, intended use. That data already sits in your supplier feed.
The practical consequence is that one decision covers hundreds of descriptions. You settle the phrases for "hiking backpacks" once, and the distinguishing details of each model come in automatically from product data. You do not sit down with every backpack separately, because at the level of a single product there is nothing left to decide.
This also explains why advice written for a single description leads nowhere here. It describes an editor working on a text. What you need is work on a system that will produce a thousand texts.
Where keywords actually belong in a description
Once you know which phrases you are playing for in a category, the question becomes where they should appear. The places that matter have not changed in years: the product name, the heading, the opening paragraph, and the technical data and attributes. If your phrase shows up there naturally, the job is done.
What you do not need to do is count. Keyword density – the percentage of times a phrase appears in a text – is not a factor that decides rankings. Google has said so consistently for years, and there is no such thing as an optimal density. What matters is whether the phrase is there at all and whether the text reads sensibly, not how many times you managed to squeeze it in.
Worth remembering when you read older guides, because some still repeat rules from the days when search engines matched strings instead of understanding meaning. Advice like "add a heading every five hundred characters" or "the phrase should appear a set number of times" has no basis today. At catalog scale it is actively harmful, because it turns work on content into filling quotas.
How AI builds a description – the four inputs
Automated description generation is often pictured as dropping a product name into a chat window and copying the answer. That is how you get text that fits everything and says nothing. How Google treats AI-generated content and where the duplication risk comes from is covered separately in our piece on AI product descriptions and SEO. What matters here is different: what a description that works for a shop is actually built from. There are four inputs.
The first is supplier data: features, attributes, the source description and the product category. That is the raw material – the specifics nobody has to guess at, because they arrive with the feed.
The second is category context, supplied by the shop. This is the knowledge no feed contains: who you are talking to, what matters to your customer, what sets you apart. This input usually decides whether the descriptions sound like yours or like anyone else’s.
The third is SEO guidance. This is where keywords enter – at category level, following the logic from the previous section. It covers not only the body of the description but also the meta title and meta description, which is what a user sees in search results. At catalog scale this counts twice over, because filling meta fields by hand for three thousand products is exactly as unworkable as writing three thousand descriptions by hand.
The fourth is a template, populated with the individual data of each product. It keeps descriptions within a category structurally consistent while each one carries its own content.
This shows exactly where keywords sit in the process. They are not appended to finished text, and they are not something the AI goes looking for on its own. They enter as one input among several, alongside product data and category context.
Where the phrases come from if you don’t have an SEO agency
Two opposite concerns tend to come up here.
The first: I don’t know SEO, so this probably isn’t for me. A misunderstanding – instead of arriving with a finished list of phrases, you can get a proposal from us.
The second is the reverse: I have an SEO agency and my own strategy, and I don’t want anyone interfering. Equally unnecessary – if you have your phrases, they go into the process and those are the ones used.
Both paths work in practice. Keywords are either supplied by the client or proposed by us. The choice belongs to the shop, and neither is a precondition – not having done your own research doesn’t block you from starting, and having your own strategy doesn’t mean giving it up.
What AI won’t do for you
Automating descriptions solves the scale problem. It does not take the decisions off your hands.
AI will not decide which phrases you want to compete for. That is a business call: whether you want visibility on broad, competitive queries or narrower ones you can realistically win. The answer depends on who you want to be in search results, and that cannot be derived from product data.
AI does not know your customer better than you do. Without category context a description comes out correct and completely impersonal. The difference between "a 30-liter backpack with adjustable straps" and a description that tells the reader what kind of trip this backpack suits and who it is for comes from your knowledge, not from the feed.
Nor does AI replace watching what actually sells in your shop. The categories generating revenue deserve more attention than the rest of the catalog – and you are the one who knows which those are.
Quality control – who clicks "approve"
Automation does not mean content goes live unsupervised. A generated description can be accepted, edited by hand, or sent back for regeneration with your own comment – pointing out, for instance, that the tone is too technical or that something important for this category is missing.
At catalog scale the goal is not to read three thousand descriptions. A more sensible approach: start with the categories that sell the most and set the quality bar there. Once descriptions in a category look the way you want them, the rest follow the same pattern.
The decision about what reaches the shop stays with a person. That matters all the more because a product description is part of selling, not just content for a search engine.
Summary
One shift puts the rest in order: you optimize the catalog, not the individual description. Phrases are chosen at category level, specifics come from product data, a template keeps things consistent, and a person decides what stays. Nobody counts density, because there is nothing to count.
If you sell from a supplier catalog and publish the same descriptions as your competitors, take a look at how AI product description generation works.