Hello! I’m Yana Lyashenko, a Google logistics specialist. Today I want to analyze with you a case that clearly demonstrates: the problem is not with the feed language. I constantly receive questions along the lines of “which feed version should I keep — Russian, Ukrainian, or both?” Some people are convinced that one of the versions is dragging the account down. Others want to reduce everything to a single language and not bother with it.
Let’s break down a typical mistake that pushes people to make wrong decisions. But first, I’ll say it straight: I recommend keeping both language versions.
Why keep two language versions of the feed?
There are two key reasons here:
How many calls and sales will I get by ordering contextual advertising from you?
Different language versions are essentially different ad formats. The system gets options to display.
Google itself determines which version to show to a specific user. It relies on the search query language, browser settings, linked Gmail account, and many other signals. You simply give the algorithm more room to maneuver.
Here’s what people encounter in practice. Suppose you open analytics for the last 14 days and see roughly this picture: the number of impressions for Russian and Ukrainian versions is almost the same. CTR is slightly higher for the Russian-language version.
Cost per click — practically identical, around 0.30 UAH. For Google, this is pocket change. The Ukrainian version has fewer clicks and lower spend. But also fewer conversions — say, by 10. Cost per conversion, accordingly, is noticeably higher. And the conversion rate differs by a whole percentage point.
Looking at the overall statistics, your hand reaches to disable the Ukrainian feed. Like, it’s ineffective and wasting budget. But wait.
There’s a category of specialists who assign one set of product IDs for the Russian-language version and another for the Ukrainian-language version. They think they’ve found a secret technique and “hacked the system.” But in reality, the problem is different.
Everyone is doing very superficial analysis. Very superficial. Glanced quickly at the numbers — well yes, the Russian version looks slightly better. So we keep it, remove the Ukrainian one. Logical? At first glance — maybe. But this is a trap.
Detailed product-level analysis: why two feeds are needed
Usually, analysis at the level of specific products is overlooked. Not a general summary for the entire feed, but a comparison of one product in two language versions. Let’s go through a few examples.
Example 1
Let’s take product ID 8088. What do we see for the analyzed period? The Ukrainian card received 439 impressions, cost per click — 3.50 UAH (quite high), conversion — 28 UAH. Conversion rate — 50%. The Russian-language version of the product — 73 impressions and two clicks. Zero conversions.
Here we only understand that one language version received more impressions. And here’s an important point: we don’t see associated conversions between versions. Those 439 impressions of the Ukrainian card could well have influenced the decision of a user who later converted through the Russian version. And vice versa. The system simply won’t show this connection.
Example 2
Product ID 8017. The situation is similar, but there are more conversions for the Ukrainian-language card — three in total. Remember, a minute ago we were going to disable the Ukrainian feed as “ineffective”? And here a specific product shows the opposite picture. And the Russian-language version of the product would also have given more conversions — if it had received clicks. Plus, again, associated conversions aren’t accounted for.
How many calls and sales will I get by ordering contextual advertising from you?
I found another pair with matching IDs. The picture here is telling. Russian card — 877 impressions, Ukrainian — 322. The Ukrainian one “underperformed” in impressions, cost per click dropped by a hryvnia. Same number of conversions — one.
This analytics shows the overall effectiveness of a specific product position. It doesn’t prove that the Russian card performed better than the Ukrainian one. If we could match geography — it still wouldn’t tell us anything. Just numbers for the selected period.
Example 4
Let’s take product ID 2621. For the Ukrainian-language card — two conversions, for the Russian-language — zero and one click.
Is the Ukrainian version more effective? Or did the Russian one just not get enough traffic? And in the previous example, by the way, the Russian-language card was more expensive than the Ukrainian one. See how easy it is to get confused when diving into product-level analysis? All products tell their own stories, and they may contradict previous ones.
Example 5
And finally — product ID 140. Here’s something even more interesting: the Russian-language card was shown, but the Ukrainian one — wasn’t.
Conclusion from general and product-level analysis results
If you look at the overall picture ignoring the breakdown by product IDs — it seems that Russian and Ukrainian versions receive roughly the same number of impressions. A difference of one or two thousand isn’t fundamental. Clicks differ due to better CTR on one side. And at first glance, the Ukrainian feed looks like the weak link.
But dig deeper into product-level analysis — and the preliminary conclusion about “ineffectiveness” crumbles. With a small number of conversions on individual products, you can’t make a statistically correct conclusion. Without a specific ID — the Ukrainian card seems like a failure. Add the ID and you see — for a number of positions it works great: both conversion rate is decent, and cost per conversion is acceptable.
The conclusion is quite simple — don’t get hung up on the “right” language version. And I don’t recommend duplicating or changing product IDs for different feed versions. Don’t think that by doing so you’ll radically “break” or “improve” the algorithm. There’s a tool, and it has clear instructions.
Why should the same product ID be specified in different language versions?
Google’s documentation states: when there are multiple language versions of a product card — upload them with the same ID.
This is done so the system gets maximum variety of ads. Google tries to pick the most relevant version for all users — taking into account their language preferences, search history, account settings.
Suppose I’m in Lviv region, but occasionally search in Russian or English. The system sees this and may serve me results in the language it deems appropriate. But if I set only Ukrainian in my settings and search using it — the probability of seeing a Russian-language card will be close to zero.
But even if I suddenly get a card in a language different from my query — it’s not a catastrophe. Yes, there’s no 100% guarantee that everything will work perfectly. I’ll go to the page, see the Russian-language version, make a purchase or not — and Performance Max will account for all of this. It tracks which language versions produce the best results and adjusts impressions. This kind of variety is part of the mechanism.
Conclusions
Don’t think you’ve found a secret report or clever technique that will overturn Performance Max’s logic. The tool includes a huge number of well-thought-out mechanisms and signals — most of them impossible to grasp simultaneously. Forget about “conspiracy theories” like “Russian-language version works better” or “keep only Ukrainian.” This is a flexible adaptive system that you just need to know how to use.
