SEO and AI search visibility: measured case studies

B2B marketplace: fixing five indexing issues

An online marketplace with a catalog of 244 listings. The owner knew the listings were not in search and had been told by a previous contractor that indexation was fine.

What we found. We found five concurrent issues affecting discovery and indexing: a rule in robots.txt, a noindex tag in the page code, a canonical tag pointing every listing to the home page, the listings' absence from the sitemap, and no HTML links to them from the catalog — the catalog drew them with a script. The issues needed to be addressed together; a sitemap alone would not resolve an explicit noindex rule.

What we did. Removed all five in one change, with a rollback kept on the server; added eleven category pages built from the same template, with structured data for article, FAQ and breadcrumbs; installed the analytics counter that had been set up but never placed on the site (zero visits recorded before that).

What it gave. The eleven category pages entered the search index within one to three days of publication. The client came back with a second order. In AI answers the marketplace was not named before the work; the measurement is repeated monthly with the same prompt set.

Client not named; identifying details removed. Numbers from the search engine's webmaster console and our own crawl, kept in the client's report.

Medical equipment rental: improved search visibility

A rental service for newborn phototherapy lamps in two Russian regions, eleven years in business, all traffic from paid ads. In July the site was outside the top 30 in search for all five core queries and was not cited by any AI assistant.

What we did. Structured data for the organization and products, answer-first blocks for the questions parents ask, a verified medical reviewer block, corrected business listings on maps, and a fix for page rendering on mobile: with Google fonts blocked by some mobile carriers the page took up to 135 seconds to draw; moving the fonts to the site brought it to 0.3 seconds.

What it gave, three months later. Search impressions per week roughly tripled (199–312 in April–June, 642–773 in September). Query variants in the top 10: 25 of 50 in July, 125 of 159 in September. Average position for the main rental queries: 17.8 → 10.2. The search engine's AI assistant now names the service by name for the “where to measure bilirubin at home” question. Clicks grew modestly — positions 7–10 are clicked rarely — which is stated in the report as it is.

Client not named here; numbers from Yandex Webmaster and our own repeated measurements, raw files attached to the client's report. A longer write-up is on the Russian case page.

Entering the UAE market: validating search demand

A renovation company planning an office in the UAE came with one question: a keyword tool showed demand there a thousand times smaller than at home — is that true?

What we did. Before measuring the market we tested the instrument itself with control queries known to be live in that country. The tool was fine, so the zero was a fact: the channel “customer searches in a Russian search engine” does not exist for that market. Customers there search elsewhere — so that is where we measured: 12 queries in two languages, two runs each, results in the target city's locale, 17 competitors with prices and channels, 10 niches checked for occupancy.

What it gave. Two measurements four days apart showed that an “empty” niche in AI answers is not constant: on 25 August nobody was named for the main query, on the 29th a confirmed player was. Three conclusions about empty niches reversed after checking the answer text — the places were taken. We left them in the report together with the correction and the quotes that disproved the first version. A rejected measurement run (zero on everything, including controls) is attached and labelled as such, so the client sees not only the successful attempt.

Published with the client's consent; identifying details removed.