ChatGPT and Google Cite Russian Sources for Your Query. What to Do About It

Almost one in five Google AI Overview answers to a Ukrainian-language query is built entirely on Russian domains. Another third mixes Ukrainian and Russian sources. These figures come from a Focus.ua publication (August 2026) based on a sample of about 5 million queries.

Below is the technical mechanics behind this, an expected breakdown of .ru share by niche, and a list of what a business can actually do. This is an analysis of causes and practical actions, not an opinion column: where the data does not support a conclusion, I say so.

What the data shows

Key figures from the Focus.ua publication (12 Aug 2026, author Kateryna Hordiienko), on ~2.5M Ukrainian and ~2.5M Russian queries:

The publication does not explicitly name the tool that collected the sample, so I cite the Focus.ua publication itself rather than a "study by company X." The full methodology and complete numbers from my own measurement are published separately — citation data for the Ukrainian SERP.

It is more useful to look at the spread by niche: the share of Russian sources is not uniform. Exact percentages will come from the measurement (to be filled from the dataset), but the direction is predictable. Below is an expected hypothesis, before measurement — not measured data: where the Ukrainian-language corpus has historically been thinner, the .ru share is higher.

Why this happens

The causes are technical and explainable. It matters to understand them, because some are things a business cannot influence, and some are things it can.

Volume of indexed content

For many topics there is simply more Russian-language text on the web, and it is older. The model works with what exists in its data, not with what should exist. Where the Ukrainian-language corpus is thin, source selection shifts automatically. This is a factor a single site can barely affect.

Language proximity

A Ukrainian query does not guarantee Ukrainian sources. Models transfer relevance between close languages, so texts in another language can end up in the answer. The Focus data confirms it: 90.8% of answers are in Ukrainian, yet the source mix differs. The answer language and the source language are two different things.

Content structure

Russian content farms and aggregators have optimized for extractability for years: lists, tables, direct answers up front. Ukrainian sites more often write in solid prose that is hard to pull a ready answer from. This is a factor you control fully — through the structure of your own pages.

Wikipedia as a crutch

5.01% of all citations of Ukrainian domains go to a single domain — Wikipedia. It means that, short of sources, the model retreats to the encyclopedia. For a topic without a strong Ukrainian-language corpus, Wikipedia often becomes the only "safe" source the model is willing to cite.

The gap between ranking and citation

Only 0.77% of Ukrainian URLs from the top-100 get a mention in AI Overview. So presence in organic results barely converts into citation. Ranking high in Google and getting into an AI answer are two separate tasks, and the second has to be solved on its own.

Breakdown: how it looks on a specific query

Schematically, the mechanics look like this: for a Ukrainian-language query, AI Overview forms a short answer and shows its sources alongside. Some of them are Russian domains and Wikipedia, while Ukrainian topical pages are absent or nearly absent from the sources.

What Russian pages have and Ukrainian equivalents lack on the same topic: a direct answer in the first lines, lists and tables versus solid prose without a clear answer. It is structure, not "the algorithm's attitude," that decides whom the model cites.

What a business can do

Actions are sorted by how realistic they are, with honest notes on timing and whether the result depends on you.

• Rewrite the structure of your answers — a direct answer in the first 40 words, self-contained paragraphs, facts with numbers and dates. (fast · depends on you) More — how to get into AI Overview.
• Create Ukrainian-language content with substance — not a translation, but primary material: your own figures, dates, data. Models prefer what can be attributed. (slow · depends on you)
• Take the cited sources — correct presence in Wikipedia, industry directories, associations and trade media. At a 5.01% share this is no longer cosmetic. (medium · partly depends)
• Build mentions on Ukrainian platforms — this is the job of GEO. (slow · partly depends) Service — GEO promotion.
• Measure visibility regularly — without measurement you don't know if anything changed. (fast · depends on you) How — how to check AI visibility.

What not to do. Complaining to Google about the sources — there is no mechanism for that. Mass-translating your site — it doesn't fix structure or substance. Blocking AI bots on principle — it only removes your chance of being cited; when and why this is actually worth doing is covered separately in whether to block AI bots.

Honestly about the limit of influence. Part of the problem cannot be solved by a single site. Where a topic has no Ukrainian-language corpus, one resource won't change the overall picture. But it can become the first quality source — and then the model starts citing it instead of a Russian one. That is a realistic goal, unlike "removing .ru from the results."

How to check your topic in 20 minutes

A quick self-check with no tools:

1. Take 10 real questions your clients ask, not "what is X."
2. Run each in ChatGPT, Perplexity and Google AI Overview in incognito mode.
3. Note the source domains and their zones (.ua / .com / .ru / .org).
4. Count the share of .ru sources among all links.
5. Compare your result with the figures in this article to see how "Russian-language" your topic is for AI.

If the .ru share in your niche is above average, that is not a reason for emotion but a list of topics where Ukrainian primary content is still open.

If you need a systematic measurement rather than a one-off check — I run an AI visibility audit on your niche's prompts, with a source breakdown by domain zone.