Engagement signals: keeping Uzbek users on the page
Yandex confirms it uses behaviour in ranking; Google does not. Here is what follows practically, and how to raise engagement without manipulation.
Confirmed versus assumed
Yandex states openly that user behaviour in search results feeds its machine-learning ranking. That is the company's official position, not an SEO observation.
Google consistently denies using Google Analytics metrics in ranking and speaks about search-result interaction very cautiously. Treating GA4's bounce metric as a Google ranking factor is simply wrong.
The practical conclusion is identical either way: a page that answers the query fast and completely retains users and earns repeat visits. Work on real interaction quality, not on a number in a report.
Measuring correctly
GA4 has no classic bounce rate from the old version. It has engagement rate — the share of engaged sessions, where a session qualifies at ten seconds, two or more views, or a conversion. Comparing it to Universal Analytics bounce rate is invalid.
For pages answering short questions, low session duration is normal: the user got the answer and left satisfied. Judge the share of returns to the SERP rather than duration in isolation.
Yandex.Metrica provides session replay and scroll maps. For diagnosing a specific page that beats any aggregated metric: you see exactly which screen stops people and where they close the tab.
The first screen decides
On mobile over 3G you have a few seconds. The first screen must contain the answer: a heading matching what the person searched, the key fact or price, and an obvious next action.
A common Uzbek site mistake is a full-height banner slider with not one line of text. Users cannot tell whether they landed in the right place and return to the SERP. Sliders also hurt layout shift metrics.
On commercial pages put the price in soum, availability, Tashkent delivery time and a contact button above the fold. Those are the four things people came for, and hiding them below the scroll costs conversions.
Speed on real networks
A significant share of regional traffic in Uzbekistan runs over unstable mobile networks. Test pages with throttled connections in DevTools — the Network tab, Slow 4G profile — not just on office Wi-Fi.
The heaviest elements are usually unoptimised images, multiple font weights, and third-party chat and analytics scripts. Each third-party script adds delay to interactivity, and three widgets in a row leave the page unresponsive for seconds.
Check field data in the Core Web Vitals report in Search Console alongside lab data from PageSpeed Insights. A lab test on a fast connection hides what real users experience.
Language and comprehension
If someone arrives on an Uzbek-language query and lands on a Russian page, behaviour will be poor regardless of content quality. Language mismatch is among the most common causes of high return rates on .uz sites.
Do not auto-redirect by IP or browser language. It breaks indexing (the crawler sees one version) and irritates users who deliberately chose another. The correct approach is hreflang plus a visible language switcher.
Write in the language the audience reads, not the one considered official. Mixed vocabulary is normal in speech and often reads more naturally than a sanitised literary register.
Why manipulation fails
Behaviour-manipulation services simulate SERP clicks followed by 'correct' behaviour. Both engines detect such patterns: uniform trajectories, identical devices, anomalous time-of-day distribution.
Yandex issues a corresponding sanction with a notice in Webmaster, and lifting it takes months. The risk is disproportionate to the effect, especially in a small market where natural query volume is low and a spike is statistically obvious.
Worse, manipulation masks the real problem. If a page does not retain people, the cause is content or speed, and simulated behaviour fixes neither.