Keyword research for Uzbekistan: language hybrids and two scripts
Uzbekistan breaks the standard keyword research process: the same product is searched four ways, and volume tools see only part of it. Here is a working method.
Four forms of one query
A real Tashkent user looks for the same fridge as 'холодильник купить', 'muzlatgich narxi', 'музлатгич нархи' and the hybrid 'холодильник narxi'. All four return different results and face different competition.
Hybrids are not an anomaly but everyday speech: Russian nouns with Uzbek function words and vice versa. Volume tools represent them poorly because query normalisers try to force a phrase into one language.
Hence rule one: collect keywords in at least three passes — Russian, Uzbek Latin, Uzbek Cyrillic — then check hybrids manually through search-box suggestions.
Volume sources and their blind spots
Yandex Wordstat with a region operator returns Uzbekistan numbers but mostly covers Russian-language demand; Uzbek Latin appears only patchily. Wordstat counts all word forms, so use quotes with exclamation marks for an exact-match estimate.
Google Keyword Planner targeted at Uzbekistan shows ranges rather than exact values without an active campaign. A '100–1K' range is useless for prioritisation, so treat it as a demand-exists check, not a comparison tool.
The most reliable source is your own Search Console Performance report filtered to Uzbekistan and grouped by query. It surfaces Uzbek and hybrid phrases absent from every external tool, because these are real impressions of your own pages.
Suggestions and SERP highlighting
Google and Yandex autosuggest is a free source of live demand. Type a root word in both scripts and record the completions. Geo-bearing suggestions are especially useful: 'kir yuvish mashinasi toshkent', 'кондиционер ремонт чиланзар'.
The related searches block and People Also Ask surface informational phrasings that need their own pages. Harvest them manually across 20–30 root terms — automation here tends to hit captchas.
Watch the bolding in competitors' snippets: the engine bolds words it treats as synonyms of your query. That is a direct hint about which phrasings belong in one cluster and which must be separated.
Normalisation: collapsing four forms into one entity
After collection you will have thousands of rows where one entity is smeared across forms. You need a mapping dictionary: Latin ↔ Cyrillic ↔ Russian equivalent. Transliteration can be scripted from the Uzbek alphabet table but needs manual review: o', g', sh and ch combinations break automatic conversion.
You cannot simply sum volumes across forms — they are distinct queries with distinct SERPs. Store them as variants of one entity with separate numbers and decide per variant whether a dedicated landing page is warranted.
The signal that separate pages are needed: the two SERPs overlap by less than half. If the same sites and URLs occupy the top ten, one page carrying both variants in the copy and image alt text is enough.
Clustering against site structure
SERP-based clustering beats word-based clustering. The method: pull the top ten URLs for each query and merge queries into a cluster when URL overlap exceeds a threshold — usually three to five shared results.
For Uzbekistan the threshold must be looser than for larger markets: SERPs are less stable because few strong local sites exist in most niches and aggregators or marketplaces often occupy the top. Too strict a threshold produces hundreds of single-query micro-clusters.
The output is not a word list but a page map: which URL owns which cluster, which pages to create, which to merge. That document is what the content team works from.
Common mistakes
First: collecting only Russian keywords because 'everyone understands Russian'. They do, but they search differently — and you hand competitors the whole Uzbek-language mobile stream from the regions.
Second: padding the set with zero-volume regional queries for the sake of size. A page for a query with no demand is wasted content budget and an extra cannibalisation risk.
Third: ignoring seasonality. Air conditioners, school supplies and wedding categories have pronounced peaks in Uzbekistan. Wordstat provides month-by-month history — use it to schedule publication instead of shipping everything at once.