Cases / IWH

53 visits from ChatGPT to an English-language B2B spare-parts site

A site for foreign parts buyers. 1,359 visits in five months. Google brings 91% of search traffic, and ChatGPT became the main referral source: 53 of the 59 visits that came from AI assistants.

The description above is the case snapshot. Current Metrica figures are shown separately.

1,495non-ad visits since 1 June 2026 · Yandex Metrica
Niche
B2B export of spare parts for marine, mining and power-generation diesel engines from Istanbul
Project dates
Since the first visit · 144 days Launch · June 2026 First visit in Metrica ·
Site languages
English
Statistics
We publish aggregate data from the Metrica API
Open data
JSON

IWH reconciliation for 1 June to 28 October 2026. The archive lists 1,359 visits and 208 actions. A fresh Metrica query for the same dates returned 1,386 non-ad visits and 201 visits meeting at least one of the four contact goals. The archived source report for 1 June to 29 October lists 1,394 visits; its filters were not preserved. The non-ad API query for those dates returned 1,397 visits and 203 visits with a contact goal. These figures do not establish how many enquiries were confirmed.

Contacts208contacts recorded by Metrica goals: forms, messengers, email
Visits by month, 2026 57 Jun 200 Jul 350 Aug 339 Sep 413 Oct*

* October up to the 28th · 1,359 visits in five months

Where visitors come from
Search482 · 36%
Direct639 · 47%
AI assistants59 · 4%
Links, social, other175 · 13%

Search engines: Google 441 · Yandex 24 · Bing 17

Visits from AI assistants 59
ChatGPT · 53Gemini · 4Perplexity · 1Doubao · 1
Google Search Console · last 3 months
378

clicks

13.7k

impressions

19.5

average position

Breakdown · Alexey Reinsberg

Why it happened

One in seven visits to the IWH site ends in a contact. The best converters are people sent by ChatGPT and pages built for a specific engine brand.

01

Goal rates by traffic source

Metrica recorded 53 visits from ChatGPT from 1 June to 29 October. Of these, 37.7% reached a goal. The archived source report covers 1,394 visits; its filters were not preserved, so this total is only used within that report. The sample is small.

SourceVisitsConversion
ChatGPT5337,7%
Links from other sites12025,0%
Google45019,1%
Direct visits65912,0%
Whole site1 39414,9%
02

A brand page is a ready enquiry

A buyer does not search for “spare parts” but for “MTU parts” or “Perkins generator parts”. Pages for a specific brand and engine type convert at 18–37%: Perkins 36.7%, MTU 26.6%, Caterpillar 25%. Generic pages such as “Brands” convert at 7.9%.

Landing pageVisitsConversion
/perkins-generator-engine-parts3036,7%
/request-a-quote9230,4%
/mtu-marine-engine-parts6426,6%
/caterpillar-marine-engine-parts2425,0%
/man-d2842-marine-parts1921,1%
/man-marine-engine-parts4418,2%
/brands637,9%
03

Buyers from 15+ countries

The site is in English, and Russia is only the second country by visits. Türkiye leads (195 visits, 18%), followed by the USA, Serbia, Germany and the UAE. The highest conversion comes from India (28.6%), Bangladesh (22%), Azerbaijan (21.6%) and the UAE (17%): markets with large fleets and generator parks that need parts.

CountryVisitsConversion
Türkiye19518,0%
Russia1306,9%
USA1147,9%
UAE5317,0%
Azerbaijan3721,6%
India3528,6%
What follows from this
  • A visit from ChatGPT converts 2.5 times better than average.
  • A “brand + engine type” page works as an enquiry landing page.
  • An English site gathers buyers from dozens of countries without ads.

Yandex Metrica data for 1 June — 29 October 2026. Conversion is the share of visits in which at least one goal fired: a call, a messenger click, a form or an email.

Log

DateWhat happened
October 2026413 visits for Oct 1–28; 59 AI referrals in total
August 2026350 visits; first referrals from chatgpt.com
July 2026Pages for engine brands and part types; 200 visits
June 2026English version launched, first 57 visits in Metrica

What worked

  • A clear offer above the fold: parts for marine, mining and power engines
  • Separate pages per engine brand — these are what ChatGPT cites
  • A detailed page about the company, delivery regions and genuine vs OEM

Where the site could improve

  • Built on Tilda: no llms.txt or FAQ markup yet — the nearest growth reserve
  • Average Google position is 19.5: most queries are still on page two
  • Yandex brings almost nothing (24 visits) — expected for an export site

How it is built

TildaOrganizationsitemap · 67 pages

Open the site and view the page source: page text is served in HTML, and the schema.org markup sits in the application/ld+json block.

Every language · first version in 7 days · free CRM

Want the same result?

Yandex Metrica and Google Search Console data as of 28 October 2026. We will build your site with the same method — in every language you need, in a week.

Plan and support prices are fixed in each currency. Examples use the Bank of Russia rate dated
Message on TelegramPricing