My translated pages rank better than my English ones, in a race nobody entered

Katto is an AI video clipper. I build it on my own, in public. Over the summer I translated the whole marketing site into ten languages, on the theory that every one of them was a market nobody else was serving.

Here is what ninety two days of Search Console says about that, from 6 July to 5 October.

The number

Locale	Pages	In the top 10	Clicks	Impressions	Avg position
en	87	43	254	8,409	18.2
ja	18	13	7	318	7.8
pt	15	11	8	195	11.9
de	13	10	6	135	10.9
fr	19	11	13	102	15.0
es	16	9	1	101	12.4
ko	10	1	10	46	27.3
zh	14	11	3	43	14.5
pl	10	6	4	32	14.6
it	9	6	3	21	28.1
hi	5	3	1	6	27.7

The ten translated locales are 129 of my 216 indexed pages and 999 of my 9,408 impressions. Sixty percent of the surface, under eleven percent of the demand.

They did better on clicks: 56 of 310, about eighteen percent. Worth saying plainly, because the two numbers tell different stories and I nearly published only the one that flattered my argument.

The part I did not expect

Read the position column again.

My Japanese pages average 7.8. Thirteen of eighteen are in the top ten. They produced 318 impressions and seven clicks in three months.

My English pages average 18.2. Fewer than half are in the top ten. They produced 8,409 impressions.

The translations are not failing. They are winning, in a race nobody entered. A position is not an outcome, it is a share of a volume, and if the volume is zero then first place is worth exactly zero. I had been reading my German dashboard as a success story: ten of thirteen pages in the top ten. Ten of thirteen pages in the top ten, and 135 impressions.

Where the traffic actually comes from

This is the sentence that reorganised my thinking.

My number one country is India: 42 clicks, 702 impressions. My Hindi pages got six impressions. Not six clicks. Six impressions, in ninety two days.

Pakistan sent eleven clicks at an average position of 7.0. Algeria sent fourteen clicks at a CTR of 29.8 percent. The Philippines, Indonesia and Brazil are all in my top twelve.

None of those people searched in their own language. They searched in English, found an English page, and clicked. Meanwhile Germany, whose language I did translate, sent six clicks.

I had conflated the language of a country with the language of its searches, for a technical product, in a field whose vocabulary is English to begin with. In my data, almost nobody searches the translated vocabulary at all: the thing is called an AI video clipper nearly everywhere, German included. I can't tell you that's true of every market. I can tell you it was true of mine.

Two caveats, because this would be a dishonest post without them

My two Search Console tables disagree by a third, and I found out why. The daily chart totals 234 clicks and 8,052 impressions. The per-page table totals 310 and 9,408. Same export, same window, one filter: Search type Web.

My first guess was a different Search type on one of them. Wrong: it's a single export with a single filter. The answer was sitting in the same zip file. The Countries table totals 234 clicks and 8,052 impressions, matching the chart exactly. Only the Pages table is higher.

That's the documented behaviour, and once you see it stated that way it's obvious. A search happens in one country, so grouping by country preserves the property total. But a single search results page can show several of your pages at once, and grouping by page gives each of them its own impression, and potentially its own click. My per-page table is inflated by every query where more than one Katto page ranked, which for a site with eleven locales of near-identical pages is a lot of them.

So the two numbers aren't in conflict, they answer different questions. Every ratio in this post is computed inside the per-page table, never across the two, and that is now a choice rather than a workaround.

One consequence cuts against my own translated pages, so it belongs here. If a query surfaces both /pricing and /fr/pricing, both collect an impression, even when the searcher was only ever going to click the English one. Some of the 999 impressions my translations earned are ride-alongs on queries the English page won. Which means the honest version of my headline number is not "translations bring eleven percent of the demand" but "at most eleven percent, probably less".

The translated pages have had less runway than the English ones. They were only submitted in the sitemap across all locales from the end of September; for part of this window several of them were live but not announced. That shortens their fair share of the ninety two days. It does not come close to explaining a factor of eight, but it is real and I am not going to pretend otherwise.

What I am not doing

I am not deleting them.

They are built, they are indexed, and some of them rank well. Taking them down costs effort and buys nothing today.

The CTR column is the one argument I'd like to make for them, and it's the one I trust least. Korean converts impressions to clicks at 21.7 percent, Hindi at 16.7, Italian at 14.3, French at 12.7. English does 3.0. But Korean's 21.7 percent is ten clicks on forty-six impressions. A sample that small produces an exciting percentage and tells you almost nothing. I'm not going to claim I found a hidden multilingual conversion advantage on forty-six impressions.

What I am doing is re-measuring in three weeks and deciding per locale, from Search Console, not from a constant in my code. The decision to keep or drop a language belongs to the data, and I do not have enough of it yet.

What I would do differently

Before translating anything, I would have spent one afternoon checking whether the searches exist. Not whether the market exists, not whether the country is big. Whether people in that country type those words. For a product whose whole vocabulary is already English, that is a much narrower question than it sounds, and the answer would have changed the order I did things in.

Translating is cheap now. Being wrong about demand is still expensive, because it spends crawl budget, it spends your attention, and it gives you a dashboard full of green that means nothing.

How to check yours

• Group your pages by locale prefix and put impressions next to average position. If a locale ranks better than your main one and brings less, you do not have a ranking problem, you have a demand problem, and no amount of optimisation touches it.
• Compare your top countries by clicks against the locales you built. Where those two lists disagree, you translated the wrong thing.
• Add up your dimension tables and check they match your totals. If they do not, open the Countries table before blaming your export: it is the one that preserves the property total, and the gap with Pages tells you how often several of your own pages compete for the same query.