Google Deleted the Language Setting: What Language-Targeted AI Video Ads Must Decide Now

Google Ads language targeting change: what actually changed

Google Ads language targeting change landed in September 2026, and it removed a control media buyers relied on for years, which is why language-targeted AI video ads became a production question rather than an account setting. Google's Ads Help documentation states the campaign-level language targeting setting will be removed for Search campaigns, and that Search ads will automatically match based on the language of the ads.

For Performance Max campaigns, the campaign-level language setting no longer applies to ads shown on the Google Search Network, though the language settings an advertiser selects continue to apply to other channels inside the same Performance Max campaign, such as YouTube, the Display Network, Discover and Gmail. Shopping ads within Performance Max are not affected by language settings. Google states plainly that no action is required by advertisers for this change. That last sentence is where the damage hides: nothing breaks, no campaign is rejected, no warning fires. What changes is where the decision lives. The setting used to be a targeting input. It is now a creative input, and creative inputs are owned by whoever makes the asset.

Google describes the Search Network mechanism plainly: it uses search term language plus the user's language settings and preferences to understand which language the user knows, and attempts to serve the best ad in a language that user understands. AI-based ad group prioritization then ensures ad candidates are prioritized in the most relevant language. Google also notes that user language can be any language the user set, regularly interacted with, or searched in, which is broader than the old interface-preference basis.

There is a worked example in Google's documentation that is worth sitting with. Pat understands both English and Spanish, and while her mobile browser is set to a Spanish interface, her other activity on Google strongly suggests she understands English too, because many of her queries are searched in English. She may therefore be shown ads in either English or Spanish when keywords match, and she may enter a query in Spanish and be served an ad in English. Under the old model, a bilingual campaign could use the dropdown to hold those two cases apart. Under the new model, the asset has to hold them apart.

Practitioners who worked through the transition describe the same shift. One agency notes the setting never filtered by what people typed, only by interface preference, so single-language accounts were already relying on Google's matching rather than the dropdown. The accounts that need attention are the bilingual and multilingual ones, where the dropdown was doing structural work. That work has moved into the creative.

Why one language-neutral cut no longer works

For video teams the consequence is sharper than for text-only advertisers, because a cut carries language in four layers at once. Spoken dialogue carries it. On-screen text and supers carry it. Caption files carry it. And the landing page the click lands on carries it, whether or not the team thinks of that page as part of the ad.

A single cut with one spoken language and English on-screen text used to be safe because the dropdown was the backstop. With that lever gone, the same cut is a bet that the audience it draws is the audience it was made for, and Google resolves that bet using what it can read: the query, the ad, the page and the user's settings.

That is exactly the inversion described in {{link}}, where a removed manual control transfers the decision to the asset and changes how teams bid.

There is a measurement consequence that arrives before the reporting does. If a language-neutral cut is now served to bilingual users in whichever language Google infers, then search-term reports stop being a clean read on language. The absence of the dropdown also removes one of the few controls that could previously rule language out when diagnosing odd traffic.

The multiplication is documented in {{link}}, where the number of distinct assets a team owes becomes a function of its audience combinations rather than its production capacity.

That is exactly the inversion described in how exclusion became bid suppression once the platform automated it, where a removed manual control transfers the decision to the asset and changes how teams bid.

The multiplication is documented in the asset-level versioning cost that multiplies with every market, where the number of distinct assets a team owes becomes a function of its audience combinations rather than its production capacity.

Language-targeted AI video ads: the four-layer decision

The workable answer is to treat language as a specification with four named layers, each owned by a different part of the workflow, rather than as a checkbox in an ad platform. This is a decision framework, not a platform feature, and it holds regardless of which ad systems a team runs.

The order of operations matters as much as the layers. Decide language before the cut is generated, because regenerating to fix a language problem costs a full pass while deciding up front costs a line in the brief. Teams that already run structured prompt specification can carry the language decision into that same document rather than bolting on a localization step at the end.

The order of operations matters as much as the layers. Decide language before the cut is generated, not after it is rendered, because regenerating a cut to fix a language problem costs a full pass while deciding up front costs a line in the brief. Teams that already run structured prompt specification for their generations can carry the language decision into that same document rather than bolting on a separate localization step at the end.

Two guardrails keep the framework from becoming an excuse to ship unreviewed variants. Per-language review is not the same standard as per-language generation, and a language a team cannot review competently is a language it should not generate into a market. And if a market already requires a specific cut for unrelated reasons, that constraint stays in force and language is layered on top of it.

The same shift shows up in {{link}}, where teams built video that never reached a serving specification and had to close the gap between production and delivery.

The same shift shows up in the delivery specification and calendar work that catches assets nobody serves, where teams built video that never reached a serving specification and had to close the gap between production and delivery.

Search Network versus YouTube: the split most teams will miss

The Performance Max detail is the one most teams will read past. The removal is not uniform. The language setting no longer applies to ads on the Google Search Network, where those ads behave like standard Search campaigns. It still applies on YouTube, the Display Network, Discover and Gmail. One campaign, one language setting, two behaviours depending on inventory.

That split creates a specific and easily missed diagnostic problem. A Performance Max campaign with mixed-language assets in a single asset group will now have its mixture sorted out automatically on Search, and will continue to be filtered by the language setting on the other channels. Two inventories inside one campaign are therefore answering the same language question with two different rules, which means a performance gap between them is not necessarily a creative gap.

The structural response is to split asset groups by language before drawing conclusions, because an asset group that mixes languages cannot be read cleanly on the inventory where the setting no longer applies. That is a small structural change with a large effect on data readability, and it costs far less than reverse-engineering the mixing after spend has been allocated.

There is a second consequence for video specifically. YouTube inventory continues to honour the language setting, which makes it the one surface where an advertiser can still use a control rather than an asset to hold language apart. For teams running video, that asymmetry matters: the platform removed language targeting from the surface where text-heavy search ads live, and left it intact where video lives.

That is an argument for treating video inventory as the place where controlled language delivery still exists, and search inventory as the place where language is now earned by the asset. It also means a team that concluded language targeting was gone everywhere, and quietly stopped structuring its YouTube assets by language, has removed a control that was never taken from it.

What to audit this week

The audit is small and specific, and can be run before any budget changes. Start by listing which campaigns and assets currently carry a language setting, separating Search and AI Max for Search campaigns on one side and everything else on the other, since only the first group has had the setting's meaning changed.

Next, identify the bilingual and multilingual campaigns, because those are the only ones where the removed control was doing structural work. For each of those, check that each ad group is unambiguously one language from top to bottom: the creative, the captions, the supers, and the page the click lands on. Google reads all of it, and an ad group pointed at a bilingual page with thin copy in one language is now a coin toss rather than a controlled split.

Then check asset groups inside Performance Max campaigns for mixed-language assets, and split them by language so Search inventory becomes readable. Do this before attributing any performance difference to the algorithm, because the algorithm is now sorting something the structure left ambiguous.

None of this requires a platform migration, a rebuild or a new tool. It requires a decision that used to be made in a dropdown and now has to be made in a brief, by the people who own the asset, for every market the team sells into. That is a real transfer of work, and it is the most consequential part of a change that Google describes as requiring no action at all.

Teams already running structured review can reuse that habit, and {{link}} describes the quality bar those gates have to clear.

Teams already running structured review can reuse that habit, and the review gates that decide which cuts are shippable now describes the quality bar those gates have to clear.