While a store has thirty products, descriptions are written by hand. At three thousand that is already months of work, at fifteen thousand it is a full-time job. So most catalogues load descriptions straight from the supplier's feed. That looks fine right up until one simple question: how many other stores are loading the same text?
This article is about how a bulk catalogue rewrite works, what it actually removes and what not to expect from it. We explain it through Textory, our product that does exactly this.
Why a supplier description is technical loss
A supplier hands the same description to all of its partners. The same paragraph stands word for word in dozens of stores, and the search engine sees a page that adds nothing new to the index. This is not a penalty or a sanction. It is an absence of grounds: the page has nothing to stand out with among identical ones.
Structure is the second half of the problem. A supplier description rarely has headings, a specification list or a separate meta description — that is, the things that make a page a document rather than a paragraph of text.
The limit, right away, so there is no misunderstanding later. A unique description removes technical loss: duplicates, thin content, missing structure. It does not create demand and does not lift rankings by itself — those depend on assortment, prices, links and how people behave on the page. We are writing about a mechanism, not about a result.
The conveyor: a feed in, a feed out
A bulk rewrite only makes sense when it does not create a second round of manual work in the form of spreadsheet merging. So the cycle is closed at both ends.
Import. A feed link or a file: XML, YML, CSV, XLSX. The parser reduces any of these formats to a single product model — name, description, price, currency, category, brand, images, specifications.
Selection. A product with a description of twenty characters or more gets the «to be processed» status. A product without a description is marked as skipped, and that is deliberate: there is nothing to rewrite, and inventing a description from nothing is a different task with different risks.
Queue. Each product goes as its own job. Each one takes a single structured call to Gemini that returns the rewritten description at once, plus a new name if needed, keywords and a meta description. Not four requests but one: it costs less that way and the fields do not contradict each other.
Export. The finished catalogue is exported in the platform's format — Prom.ua, Rozetka, generic XML, CSV or JSON — with a choice of which fields go into the feed. The file is ready to import back without manual edits.
Processing statuses are visible per item: pending, processing, done, skipped, error. Processing can be paused and resumed — on a catalogue of tens of thousands of items that is not a convenience but a way to control spend.
Why the preview matters more than the settings
The most expensive mistake in a bulk rewrite is launching the whole catalogue and only then seeing that the tone is wrong. So in Textory one typical product from your niche is rewritten first, and it already shows what the rest will look like.
You can set the tone — sales, neutral, technical — the length and the emphases, and switch keywords, the meta description and name uniqueness on or off. For Prom.ua the target name_ua and description_ua fields are generated together with the description and go into the export, so the feed does not have to be finished by hand after unloading.
GEO mode: structure that AI systems find easier to cite
A separate toggle optimises the description for citation in AI search: the first sentence becomes a self-contained factual answer, subheadings are phrased as questions, fact density rises, entities are named explicitly.
Alongside it works a GEO score from zero to a hundred. An important detail: it is a deterministic heuristic, not an AI grading itself. It counts the presence of a direct answer, question-shaped headings, fact density, named entities, a specification list and length. That makes the score reproducible — the same text always gives the same number.
And the literal limit we repeat everywhere AI search comes up: nobody, ourselves included, can promise a mention in an AI answer. GEO mode gives structure that systems find easier to parse and quote. Whether they quote it does not depend on us.
What we measured
The measurements concern the service itself, not its users' sales.
Between 31 July and 2 August 2026 we measured our own sites. Textory responded in 0.22 seconds at 141 KB of HTML — the fastest result among the 16 agency sites measured. Eight public pages, JSON-LD markup for the application, the organisation, the site and the questions block, a working PWA manifest. That is the state of the site on the date of measurement.
To check behaviour at volume there are demo catalogues of up to 11,806 products, so you can watch the queue work before uploading your own feed.
What still deserves a human hand after the rewrite
An automatic pass is a good draft for the whole catalogue and the final text for most of it. But two cases stay with a person.
The first is flagship items that carry the main traffic and that you know better than any model. Here the rewrite gives structure, and the substantive emphases are added on top.
The second is categories with legally sensitive wording: anything involving composition, permits, compatibility or contraindications. Automatic text must not invent such facts, and they have to be checked by eye.
Where this plugs in
If the catalogue is updated constantly, manual export stops being acceptable by the second month. For that there is a REST API, access keys for external systems, webhooks on processing completion and permanent public links to the generated feed with a chosen set of fields. For OpenCart there is a separate synchronisation module.
On how that feed then behaves on marketplaces we wrote in an analysis of why Rozetka rejects products, and on connecting the marketplace itself, on the Prom integration page.
Data
A catalogue is commercial information, so about how it is handled, briefly and plainly: there is an NDA mode, catalogues are stored encrypted, the data is used only to provide the service, is not passed to third parties and does not go into AI training.
In summary
A bulk rewrite removes technical loss that cannot be closed by hand at this volume: identical descriptions, missing structure, empty feed fields. It does not replace strategy and does not promise rankings, and selling it inside a store is more honest exactly that way — as a tool against a specific problem.
You can try it on your own catalogue at textory.com.ua, where the pricing model follows the volume of text and the current terms are on the product site. Our other products are in the own products section. If you would rather have the work done than configure it yourself — description generation on your catalogue, a custom integration or automated export — we describe that format of task on the AI product description generation page. We reply within two hours and work under a contract.
