Five Languages Is Five Markets, Not One Campaign

Starting this month, the campaign-level language setting disappears from Google Ads Search campaigns. Google’s own documentation states it plainly: the setting is being removed, search ads will match automatically based on the language of your ads, and the system will prioritize the ad whose language matches the language of the search term (About language targeting, Google Ads Help). Performance Max keeps its language setting for YouTube, Display, Discover and Gmail, but the Search Network portion now behaves like a standard search campaign.

If you run paid media in more than one language, that sentence is the whole memo. The control you used to set in a dropdown is now carried entirely by the words in your ad and on your landing page. Which makes this a good moment to look at everything else that breaks when a program written in one language is pointed at several.

The setting is gone, so the copy is the setting

Under the old model you could run English creative and lean on the language setting to keep it in front of people who understood it. That safety rail is being removed. From here, comprehension is inferred from the creative, the landing page and the user’s own signals, and the ad that matches the searcher’s query language wins the prioritization.

Two operational consequences follow immediately. First, an account holding mixed-language assets inside a single campaign has no clean way to control which language shows where. Separate campaigns per language, with matching landing pages, is now the structure that gives you predictable serving rather than a stylistic preference. Second, your landing page language is a matching signal, not just a courtesy. Sending Spanish ads to an English page tells the system your Spanish audience is not comprehending the destination, and it undercuts the ad you just paid to write.

Google says no action is required from advertisers. That is true in the sense that nothing will break in the interface. It is not true in the sense that accounts built around the old setting will quietly start serving differently.

Translated ads lose to written ones, and the reasons are mechanical

The usual explanation is that translation lacks cultural nuance. That is real, but it is not what shows up in the account. What shows up is four specific failures, and all four are visible in a spreadsheet.

The first is the keyword-to-headline gap. A responsive search ad earns its click by containing the phrase the person typed. Translate an English headline and you get a grammatically correct sentence that happens not to contain the words locals actually search. The ad is fine. It just is not about their query.

The second is character limits. Google allows 30 characters per headline and 90 per description regardless of language, and languages are not equally dense. German compound nouns, Finnish agglutination and Russian case endings routinely exceed a limit that English fits comfortably, so translators truncate, and truncation lands on the verb or the offer.

The third is inflection breaking automation. Keyword insertion, countdown customizers and dynamic parameters assume a word can be dropped into a sentence unchanged. In languages that decline nouns, the inserted term arrives in the wrong case and the headline reads as broken to a native speaker, in a way no English-speaking reviewer will catch.

The fourth is register. Formality is a choice in German, Japanese, Korean and French, and a literal rendering of an English imperative picks one by accident. Half the time it picks the wrong one for the product and the buyer.

None of this argues for expensive translation. It argues for a different brief. Give a native writer the local keyword list and the offer, not the English ad, and let them write against the queries. That is a content production question rather than a translation purchase, and it changes who you hire.

Keyword research does not survive translation either

Search demand in a new language is a different distribution, not the same distribution relabeled. The direct translation of your best English term may carry almost no volume, while a borrowed English word carries most of it. Categories are frequently searched by the name of the market-leading brand rather than the generic term. Framing shifts too: a market may search for the problem where yours searches for the solution, which changes the entire page you should be ranking.

Intent moves as well. The same concept can sit at a different point in the funnel depending on how mature the category is locally, so a term that is commercial in one market is purely informational in another, and bidding it the same way wastes budget in one of the two.

The working method is to build each market’s keyword list natively, from local search data and from how local competitors describe the category, then map your English terms onto it afterwards. Doing it the other way round, translating the English list and looking up volumes, produces a list of terms that are correct and unused. Search work in a second language starts before anything is written, not after.

hreflang is a reciprocity protocol, not a tag you sprinkle

On the organic side, the mechanism that decides which language version of a page a searcher sees is hreflang, and most implementations fail on the same points.

The annotations are reciprocal. Every version of a page carries the same set of link elements listing every variant including itself, and the set is identical across versions. If page A claims page B as its Spanish alternate but B does not point back, Google may disregard the annotation, precisely so that a third party cannot declare itself an alternate of your page. Where a full matrix becomes unmanageable, Google’s guidance is to prioritize bidirectional links to your originating or dominant language rather than between new variants (Google Search Central, localized versions).

Add x-default for anyone whose language you have not listed, particularly on a homepage that auto-redirects or shows a country selector. Deliver the annotations in the page head, in HTTP headers where the file is not HTML, or in the XML sitemap, and pick one method rather than maintaining three that will eventually disagree.

The failure that undoes everything else is canonical. If every language version canonicalizes to the English URL, you have told Google the other versions are duplicates and should not be indexed, which cancels the hreflang set you just built. Self-referencing canonicals on each language version, every time.

The URL structure question, country domains against subdirectories against subdomains, matters less than the consistency of what sits on top of it. Any of the three can rank. None of them survives a broken reciprocity chain or a stray canonical.

The money usually leaks after the click

An ad written by a native speaker, matched to a native keyword, landing on a page that displays prices in the wrong currency and offers a payment method nobody in that market uses, will still fail. The friction on the post-click side is larger and less visible than anything in the ad account, and it deserves its own treatment, which is the accumulated cost of a store that was not built for the visitor.

For the acquisition side, one rule covers most of it. The language of the ad, the language of the landing page and the language of the checkout should be the same language, all the way through. Any break in that chain is both a conversion problem and, since September, a serving signal working against you.

Measure per market or you will optimize the wrong one

Blended reporting across languages is the most reliable way to make a bad decision. One market carrying an acceptable average conceals another burning budget, and an aggregate cost per acquisition describes no market you actually operate in.

Segment everything by language and by country, hold separate budgets, and set separate targets. A market you entered eight weeks ago should not be judged against one you have run for four years, and the temptation to shift its budget to the mature market after a slow month is how firms conclude that a language does not work when what did not work was six weeks of learning. Where the numbers matter is per market, which is also the level at which a paid program should be structured, staffed and reviewed.

Adding a language is not a feature you switch on. It is another market with its own demand curve, its own competitors, its own vocabulary and its own reasons for saying no. The firms that do it well treat the second language as a launch, and the ones that struggle treat it as a translation job with a deadline.

Frequently Asked Questions

Do we need separate campaigns for each language now?

In practice, yes, for search. With the campaign-level language setting removed, serving is decided by the language of your ads and landing pages alongside user signals, and mixed-language assets inside one campaign leave you without a clean way to control which creative appears where. Separate campaigns per language, each pointing at pages in that language, give you predictable serving, cleaner reporting and independent budgets. The structure also makes it obvious when one market is being subsidized by another.

Is machine translation ever acceptable for ad copy?

As a first draft for a native speaker to rewrite, yes. As published creative, no, and the reason is not quality snobbery. Machine output is fluent but written against your English sentence rather than against the queries people type, which is the specific job an ad headline has to do. It also has no awareness of character limits, formality registers or how an inserted keyword will decline. Use it to save time, not to skip the writer.

Should we use country domains, subdirectories or subdomains?

All three can work. Country-code domains give the strongest local signal and the highest maintenance cost, since each one builds authority separately. Subdirectories consolidate authority under one domain and are the least work to maintain. Subdomains sit in between. The decision matters far less than implementing hreflang reciprocally and keeping canonicals self-referencing, which is where most international sites actually lose their rankings.

What is the most common hreflang mistake?

Canonicalizing every language version to the original-language URL. It tells search engines the other versions are duplicates that should not be indexed, which quietly nullifies a correct hreflang set. The second most common is a broken return link, where one page claims another as an alternate without the reference being confirmed from the other side. Both are invisible in the browser and both are caught by a crawl.

How long before a new language market should be showing results?

Long enough that the answer should be set before launch rather than argued about afterwards. Paid campaigns produce data quickly but need a learning period plus enough conversion volume to optimize against, and both take longer in a small market. Organic visibility in a new language compounds over quarters. The practical approach is to agree the review point and the minimum volume in advance, then leave the budget alone until you reach it.