Case study: SEO for an agricultural ecommerce catalog of 10,000+ products
SEO for a 10,000+ product store on ShopRenter: category structure, content, two-week sprints, and 38% more impressions in October than a year earlier.
Traffic held while the changes rolled out in two-week steps, and in the month after the engagement ended, impressions were up 38% on the same month a year earlier.
| Result | Metric | Note |
|---|---|---|
| +38% | Google impressions | |
| +28% | Clicks | |
| +15% | Google organic sessions (GA4) |
Google Search Console, GA4
The store
agrobook.hu sells agricultural parts and accessories: tractor glass, cab mats, seed drill parts, sprayer nozzles. The catalog has more than 10,000 products in a tree structure, and the store runs on ShopRenter, a Hungarian ecommerce platform. Demand is strongly seasonal; the season starts in February and March.
I worked on the store from March to September 2026. It wasn’t a full SEO optimization: we didn’t get to the end of the list during the engagement. What was left out is covered at the end of this article.
How we worked
The work had five steps:
- Measurement. First, I set up data collection in BigQuery: Search Console, GA4, Bing Webmaster Tools, Microsoft Clarity and Screaming Frog. Every later change could then be measured against the same data.
- Keyword research. What buyers search for at category and product level, and how the catalog is structured against that.
- Audit. A review of the technical and content state. The audit found some duplicate SKUs, non-indexable URLs and thin pages.
- Fixes in two-week sprints. Plan, implement, measure, then the next step.
- Performance tracking. I followed the effect of the changes week by week and compared it with the same period a year earlier.
Why two-week sprints
Google doesn’t crawl every page of a site at once. How many URLs it requests in a given period is set by the crawl budget, which depends on how much load the server can take and on how important and fresh Google considers the pages. For a store with more than 10,000 products, processing a full restructure would take months. During that time the old and new states mix in the index, and the first consequence can be that the search engine temporarily ranks the pages lower.
On top of that, if many URLs and the internal links change at once and traffic then drops, there’s no way to tell which change caused it. In the middle of the season there’s no time to find out. So we moved in small steps and measured after each one.
Communication and division of work
We met with the client every week. Implementation was shared: I did part of it, and the client’s team did the rest on my instructions. They know the products, so their expertise went into the copy. In the weekly meeting both sides looked at the same measurement, which made it quick to agree on the next two weeks.
What we changed
- Category structure. We rebuilt the categories around search intent and a logical hierarchy. Tractor glass, for example, moved into subcategories by brand, because searches usually name the brand and model.
- Category pages. We rewrote the category page copy.
- Product pages. We rewrote the product content and reorganized it at category level.
- Title, H1, meta description. These made up most of the technical SEO work.
- Unneeded URLs. We deleted products that couldn’t be sold and a half-finished Croatian language version; those URLs now return 404. They brought no revenue and only took up crawling.
- Infinite scroll. We had it turned off, so pagination runs on separate, crawlable URLs. Whether this affected indexing I can’t tell in hindsight: well-implemented infinite scroll doesn’t necessarily hurt it, and the implementation depends on ShopRenter.
- Homepage. We fixed a technical issue that was slowing the homepage down. It’s the densest page in the store, with the most images, so that’s where the work made sense. Product and category pages carry far fewer images, and ShopRenter leaves little room for speed work. I have no measurement from before the fix, so I don’t give a number.
The results
Weekly clicks started to rise at the end of January 2026, before the work began. I don’t claim that rise: it’s when the season starts. What the data does show is that traffic didn’t drop from March to June, while the structural changes were going in.
Over the summer, traffic stayed around the previous year’s level. Clicks were up 11% in July and 3% in both August and September compared with a year earlier.
Own measurement · Google Search Console · agrobook.hu · June 2025 – October 2026 · weekly
In October, a month after the engagement ended, growth appeared. Compared with October of the previous year:
- impressions were up 38%,
- clicks were up 28%,
- Google organic sessions were up 15% (GA4).
Growth picked up in the second half of the month. From October 16 to 31, compared with the same weekdays a year earlier, impressions were up 55% and clicks 38%. In the last week, Google organic sessions were 51% above the same week of the previous year.
Google Search Console · agrobook.hu · October 2026 vs. October 2025 · daily
Because the comparison is with the same month a year earlier, the growth isn’t seasonal. October 2026 is the strongest month in the 16 months of Search Console data, in both clicks and impressions.
Impressions grew faster than clicks because the store is showing up for more and more queries. On a new query, clicks usually only come with a better position, so CTR fell from 3.98% to 3.69%. Average position improved from 9.2 to 7.6 over the same period, but the two figures don’t cover the same queries, because the searches the store appears for have changed.
Own measurement · GA4, Google organic sessions · agrobook.hu · September – early November 2026, against the same weeks a year earlier · weekly
This wasn’t a one-off day or week: growth held through the second half of October, and the November data shows the trend continuing.
Signs from AI search
From July, conversational, prompt-like queries started to show up in Search Console, for example (translated from Hungarian):
- “what Fendt tractors are on the market?”
- “where can I get it and how much does it cost?”
- “what’s the difference between the two”
- “recommend a reliable place to buy agricultural machinery.”
Phrasing like this may point to fanned-out queries coming from AI surfaces. GA4’s AI Assistant channel brought between 13 and 26 sessions a month from June to October. SEO and GEO rest on the same foundations; the difference is the search surface: Google, Bing, ChatGPT or AI Overviews. These numbers are still too small to draw conclusions from. They show that visitors are also arriving at the store from AI assistants.
What was left out
- I didn’t work on crawl budget optimization or log file analysis, and no systematic indexing rules were put in place.
- I left faceted navigation alone: on ShopRenter, filter pages are set to noindex, nofollow by default.
- I didn’t work on structured data, the XML sitemap, canonical URLs, product variants, or internal linking beyond the category restructuring.
- Speed work covered the homepage only. Rendering and most of the templates are set by ShopRenter.
In hindsight
With what I know now, I assume crawl budget also slowed down crawling during the work: Google had to recrawl a lot of rewritten pages, and it kept checking the deleted 404 URLs for a while too. Because we moved step by step, the changes had more time to settle, and the traffic effect only showed up in October. This is a hypothesis: without log analysis I can’t prove it.
Key takeaway
A large catalog can be restructured in two-week steps without traffic dropping along the way. The cost is time: here, the effect showed up in the month after the engagement ended. That’s why, on a project like this, measurement is worth keeping running after the work is done.