Case studies / B2B agricultural export
856 indexed product pages for a B2B exporter — and named by ChatGPT and Gemini to international buyers
An agricultural exporter's website went from 58 to 385 search clicks per 90 days and from 489 to 1,852 ranking queries, by generating close to 900 product pages from structured catalogue data. 856 are indexed. Asked buyer questions with web search on, ChatGPT named it in 28 of 147 checks and Gemini in 16 of 75.
Radiance Overseas · India, selling internationally · radianceoverseas.com
Built and run in-house — this is our own property, not a client engagement, which is why every number can be produced on request.
- 6.6×
- search clicks, 58 → 385 per 90 days
- 5.8×
- impressions, 3,213 → 18,702
- 856 of 899
- pages indexed
- 21%
- of Gemini buyer checks name it
Source: Google Search Console, 90 days to 22 May 2026 vs 90 days to 30 Sep 2026 · index coverage 12 Oct 2026 · AI probes 19 Sep – 12 Oct 2026.
What was the problem?
International buyers search for exactly what they need: a rice variety, a tea grade, a private-label manufacturer, an importer in a specific country. A catalogue site with a dozen category pages cannot answer hundreds of precise questions like that.
Hand-writing a page per product and market would take a content team months, and would drift out of date the first time a product changed.
What did we build?
Pages generated from the catalogue, not typed
Product hierarchies live as structured data. Python generators turn them into static pages — one per product, grade and use — with consistent specifications, structured data and internal links. Adding a category is a data edit, not a rebuild, and every page is reviewable in a diff.
Index coverage treated as a metric
Coverage is checked against Search Console on a schedule. A page that is generated but not indexed is a page that does not exist, so the ratio — now 856 of 899 — is tracked like revenue.
A closed loop from measurement back to content
Search Console measures the pages, a keyword pipeline reads those measurements and writes briefs, and the briefs change the generator's inputs. The site improves on a weekly cycle without anyone re-planning it.
What changed?
- Clicks from search rose from 58 to 385 per 90-day window, impressions from 3,213 to 18,702, ranking queries from 489 to 1,852.
- Non-brand buyer queries now rank: “private label tea manufacturers in India” (average position 9.8), “basmati rice importers in Saudi Arabia” (15.5), specific rice grades.
- With web search on, ChatGPT named the exporter in 28 of 147 buyer-intent checks and Gemini in 16 of 75.
- The enquiries this produces are handled by a separate desk — see the enquiry-desk case.
What is not working yet?
Most clicks are still people searching the company's name; the non-brand queries rank but sit mostly on page one's lower half and page two. Moving them up is the current work, and it is slower than the first jump was.
Common questions
Is programmatic SEO safe, or will Google penalise generated pages?
Google penalises thin pages, not generated ones. Each page here answers a distinct buyer question with real specifications; 856 of 899 are indexed, which is the clearest evidence Google considers them useful.
How do AI assistants find a B2B exporter?
The same way buyers do — by searching. A page that answers “who manufactures private-label tea in India” directly and carries clean structured data is something an assistant can quote. We measure it: 28 of 147 ChatGPT checks and 16 of 75 Gemini checks named this exporter.
Could this work for your business?
It probably fits if:
- You have a catalogue of products, grades, SKUs or locations
- Buyers search for specifics your category pages do not answer
- You sell B2B internationally and get found by importers rather than shoppers
What this used
More case studies
Student accommodation
From invisible to 577 search clicks a quarter — and recommended by ChatGPT in 73% of student questions
38.5× search clicks, 15 → 577 per 90 days
Student accommodation — paid search
The Google Ads dashboard said ₹243 a lead. The real figure was ₹1,100. Here is what we found and fixed.
~40% of recorded conversions were fraud
B2B agricultural export — sales operations
1,824 buyer enquiries from email and WhatsApp, deduplicated, scored and moved through one pipeline
1,824 enquiries captured, deduplicated