Language targeting is the mechanism by which an advertising system decides that a given person is capable of reading a given advertisement, and either restricts delivery to that group or ranks candidates so the best-matched creative wins. It exists because advertising platforms sell audiences defined by geography, and geography is a poor proxy for language. A Spanish speaker in Chicago, a Turkish speaker in Berlin and a Cantonese speaker in Vancouver all sit inside markets whose dominant advertising language is not their own. Every major buying platform therefore carries a language control somewhere in its hierarchy, and every one of them draws on a slightly different set of signals to populate it.

The control is easily misread. It does not describe the language of the advertisement. It describes the language of the person the advertisement is served to, which is why a campaign can be set to Spanish while every headline in it remains in English. None of the large platforms translate creative on the advertiser's behalf.

The signals underneath

Three families of signal feed almost every implementation.

The first is declared preference. A browser transmits an Accept-Language header on every HTTP request, a mechanism defined as part of HTTP content negotiation in the 1990s, with quality values ranking each entry. Platform accounts hold a separate declared preference: the interface language on a Google account, the profile language on a Facebook account, the language chosen on a YouTube homepage.

The second is content. Ad servers infer the language of a page or app from its text, then treat the person consuming it as a probable reader of that language. Google Ad Manager describes its own system as determining the primary language of a page, then matching demand to it and to pages recently viewed rather than only the current one.

The third is behaviour. Search queries, past clicks and the language of previously served creative all become inputs. This layer has expanded fastest, and it turns language targeting from a filter into a ranking problem: a bilingual reader qualifies for two pools of demand rather than one, so the system must choose.

Where language sits in the bid request

In programmatic buying the signal is a field rather than a setting. OpenRTB, the real-time bidding specification maintained by the IAB Tech Lab, carries language in four places. The Device object holds a language attribute, documented in the 2.6 specification as browser language using ISO-639-1-alpha-2, the two-letter code set. The Content object holds its own language attribute, describing the content rather than the person. The BidRequest object holds wlang, an allowed list of creative languages imposed by the exchange or publisher. The Bid object in the response declares the language of the creative offered, with the non-standard code "xx" reserved for creative carrying no linguistic content at all, such as a logo-only banner.

OpenRTB 2.5, published in 2016, added wlang and the creative-level language field. OpenRTB 2.6, released in 2022, added the parallel langb and wlangb attributes, which carry IETF BCP 47 tags rather than bare two-letter codes and so allow a request to distinguish Brazilian from European Portuguese. Only one of the two forms should be present, leaving exchanges to pick a convention and bidders to handle both.

Demand-side platforms expose this as targeting. Display and Video 360 places language on the line item, where languages may be included or excluded but not both, and matches against the languages declared in the bid request by the publisher. Where a publisher declares nothing, Google's documentation states that the platform infers likely languages from content and user signals. The consequence is stated plainly on the same page: a Spanish-language line item can win an impression on an English-language website, because the target is the person, not the page. Campaign Manager 360 takes the stricter route, matching only against browser settings and marking ads ineligible when browser language information is unavailable.

The search platforms

Search buying inverts the emphasis. A query carries language directly, so the campaign setting has historically been a coarse eligibility gate laid over a much richer set of runtime signals.

Google Ads has carried language as a campaign-level setting alongside location for most of the platform's history, editable in bulk through Google Ads Editor. Its documented behaviour on Search is that the system reads search term language, account settings and preferences, then attempts to serve the best available advertisement in a language the person understands. Google's worked example describes a user whose mobile browser is set to a Spanish interface receiving English creative, because her query history suggests English comprehension.

Microsoft Advertising splits the control across three levels rather than one. When the platform added Thai, Malay, Indonesian, Filipino and Hindi on April 2, 2024, it documented an ad language, a campaign language available to Performance Max, and an ad group language. Feed-driven formats sidestep the setting entirely: Microsoft's Feed Labels, announced in March 2024, group products across feeds by shared attributes including language.

Publisher-side language support is a separate gate on the same axis. AdSense supported 43 languages when Telugu was added in 2018, and publishers writing in an unsupported language could not monetise at all. Feature rollouts remain language-bound: Related Search in Auto Ads reached French, Spanish and Japanese in September 2024, and Google's text guidelines beta gained full language support only in February 2026.

Limitations and disputes

The underlying signal is degrading by design. Chrome's Accept-Language reduction, opened as an origin trial with Chrome 109 in January 2023, sends a single most-preferred language instead of the full list, on the grounds that the header is a passive fingerprinting surface exposed on every request. Safari already behaves this way. Google's explainer cited a study finding only 7.2% of sites used the header for negotiation at all. A bilingual reader who once announced two languages to every server now announces one.

Comprehension is inferred rather than known, and inference needs volume. Sparse search histories and restricted networks generate fewer signals, a gap flagged in PPC Land's 2025 coverage of the Google Ads changes and not addressed since.

Exclusion is the most persistent misconception. Leaving a language unselected in Google Ads has never excluded it, a point Google restated when announcing the setting's removal, and Search Engine Roundtable logged that language exclusions remain unsupported.

Content language is also becoming unstable. YouTube automatic dubbing reached more than six million daily viewers across 27 languages by May 2026, which means a single video may carry audio in a language its publisher never produced. Measurement carries the same strain: Nielsen's revised estimate of Spanish-speaking households, effective September 26, 2026, changes how Spanish-language local television inventory is valued.

Adjacent terms

Location targeting selects by where a device is, not by what its owner reads; the two are frequently conflated because most markets have a dominant language. Meta's 2024 expansion of location targeting to people merely interested in an area widened that gap further, and PPC Land's explainer on the designated market area covers the geographic unit itself.

Ad language or creative language is the language the copy is written in, a production deliverable rather than a targeting setting.

Contextual targeting selects by page subject. Language detection is a component of it, but the two are separable: a Japanese-language cookery page and an English-language cookery page share a topic, not a language.

Locale bundles language with region, currency and formatting conventions, which is why a feed can be labelled by locale while a campaign is targeted by language alone.

Recent developments

Google announced on August 14, 2026 that the campaign-level language setting would be removed from Search and Performance Max campaigns starting in September, nine months after the end-of-2025 deadline set in its own documentation. Ginny Marvin, Ads Product Liaison at Google, wrote that the setting had become redundant because the system already handles language automatically, and added that "I know it sounds like a big shift, but it really isn't."

Scope is where the accounts differ. Marvin named Performance Max without qualification. Google's help documentation, as of September 2026, describes something narrower: the campaign-level setting stops applying to Performance Max advertisements shown on the Google Search Network, while continuing to apply to YouTube, Display, Discover and Gmail, with Shopping advertisements unaffected. PPC Land's weekly analysis recorded the same Search Network framing.

A second change ships alongside the removal and is not a deletion. Where a query carries a clear language, the system will prioritise the advertisement and landing page matching that language, a check Google marks as inactive before September and active after.

The removal lands in a crowded month. Automatically created assets and campaign-level broad match converted to AI Max on September 1, 2026, and a bidding change that lifts budget-capped campaigns toward their stated targets began rolling out on August 17. Attributing September variance to any one of them will be difficult, particularly given that Lunio reported 72% more invalid traffic on retail AI Max campaigns on August 12, 2026.

The direction is consistent. Manual boundaries have been converted into model inputs across the platform, from Lookalike segments in Demand Gen in February 2026 to the campaign creation flow itself in September 2025. Where controls return, they arrive as instructions rather than settings, as with AI Brief, or as reusable configuration, as with the targeting templates shipped alongside natural language search in Display and Video 360 in September 2025.

Timeline

  • January 1997 - HTTP/1.1 formalises content negotiation, including the Accept-Language request header with quality values
  • September 2009 - IETF publishes RFC 5646, the language tag standard referenced as BCP 47
  • 2016 - OpenRTB 2.5 adds the wlang allowed-language list to the bid request and a language attribute to the bid
  • July 2018 - Google AdSense adds Telugu, bringing supported publisher languages to 43
  • 2022 - OpenRTB 2.6 adds langb and wlangb, carrying BCP 47 tags alongside the existing ISO 639-1 fields
  • January 2023 - Chrome opens an origin trial reducing the Accept-Language header to a single preferred language
  • March 2024 - Microsoft Advertising announces Feed Labels, grouping products across feeds by attributes including language
  • April 2, 2024 - Microsoft Advertising adds Thai, Malay, Indonesian, Filipino and Hindi, documenting ad, campaign and ad group language levels
  • September 2024 - AdSense extends Related Search in Auto Ads to French, Spanish and Japanese
  • August 18, 2025 - Google documentation states the language targeting setting will leave Search campaigns by the end of 2025, with non-Search functionality unchanged
  • February 26, 2026 - Google text guidelines beta goes global with full language support
  • May 2026 - YouTube automatic dubbing passes six million daily viewers across 27 languages
  • August 14, 2026 - Google announces removal of the campaign-level language setting from Search and Performance Max, effective September 2026
  • September 2026 - Setting removed; prioritisation of creative matching search term language becomes active
  • September 26, 2026 - Nielsen's revised Spanish-speaking household estimates take effect for local television markets

Summary

Who: Buy-side advertisers and agencies configure it; demand-side platforms including Display and Video 360, Campaign Manager 360 and Amazon DSP expose it as targeting; search platforms including Google Ads and Microsoft Advertising expose it as a campaign or ad group setting; publishers and exchanges populate the underlying fields; the IAB Tech Lab maintains the OpenRTB attributes that carry it; browser vendors including Google and Apple control the declared preference signal.

What: A targeting dimension that selects or ranks advertisement delivery according to the language a person is judged to understand, drawn from declared browser and account preferences, the language of content consumed, and inferred behavioural signals. It governs who sees the advertisement, not what language the advertisement is written in.

When: The declared-preference layer dates to HTTP content negotiation in the 1990s. OpenRTB added an allowed-language list in 2.5 in 2016 and BCP 47 tags in 2.6 in 2022. Google announced removal of its campaign-level setting on August 14, 2026, effective September 2026.

Where: Present in every major buying interface, though at different levels of the hierarchy: campaign in Google Ads, campaign and ad group in Microsoft Advertising, line item in Display and Video 360, ad set in Meta Ads Manager, and bid request fields in OpenRTB transactions.

Why: Geography does not determine language, and advertising sold by market therefore mismatches multilingual populations. The dispute is over method: manual selection gives auditable control but rests on a signal browsers are deliberately reducing, while inference reaches bilingual audiences more widely at the cost of a diagnostic variable advertisers previously used to explain irrelevant impressions.