Multilingual contextual targeting: what the brief must specify
Part of Contextual targeting
The short answer
Multilingual contextual targeting places ads beside content in more than one language by reading each page in its own language. The brief must name the languages for each market, rewrite every topic natively for each language, assign a fluent reviewer per language, say how mixed language pages count, and decide whether each language gets its own site list.
Key takeaways
- A multilingual brief names languages per market, not one language list for the whole campaign.
- Topics written in one language and machine translated describe the wrong articles, so each language needs its own topic descriptions.
- Every language on the plan needs a named reviewer who reads it fluently; translation tools are for triage, not approval.
- Pages that mix languages need a written rule before review starts.
- A language earns its own site list when its creative, budget, reviewer or reporting differs from the others.
What is multilingual contextual targeting, and where do plans go wrong?
Definition
- Multilingual contextual targeting
- Selecting pages, sites or channels by their content when the plan covers two or more content languages, with each page judged in the language it is written in rather than through a translation.
A multilingual plan can fail at the brief. A planner writes the topics in English, adds a line saying the campaign also runs in Spanish and Korean, and assumes the matching will sort it out. The English sites end up judged against careful descriptions and the other languages against a literal translation, sometimes by nobody who reads them.
The method is the same as single language planning: describe the ideal content, read sample articles, keep the evidence. What changes is that each step repeats per language, with its own vocabulary, publishers and reviewer. Canadian French has its own planning questions, covered in our guide to French Canadian advertising.
What should a multilingual contextual brief specify?
Seven fields separate a multilingual brief from a translated one. Each answers a question that otherwise gets settled by default, which can favor the language the planners read.
| Field | What to write | Why it matters |
|---|---|---|
| Languages per market | Each market with its languages, such as Southern California: English, Spanish, Korean | One campaign wide language list overstates or understates a language |
| Language variant | A tag in the BCP 47 style, such as es-US or zh-Hant, when the variant matters to the message | US Spanish and Spain's Spanish use different words and institutions |
| Creative language | Which ad runs beside which content language | An English ad beside a Korean article should be a choice, not an accident |
| Topics per language | Ideal article descriptions written natively for each language | Literal translations describe articles native publishers do not write |
| Avoid topics per language | The same concern, phrased the way each language's press covers it | Sensitive local stories differ by community |
| Reviewer per language | A named person who reads the language fluently, with their sign off date | Approval by someone who cannot read the page is not approval |
| Mixed page rule | How pages with two languages count toward a site's language | Otherwise two reviewers classify one bilingual site differently |
The IETF's BCP 47 language tags (opens in a new tab) combine a language with optional script and region subtags; the specification's examples include es-419 for Latin American Spanish and zh-Hant for Chinese in Traditional script. Using that notation removes the ambiguity of a word like Chinese, which can mean two scripts and several spoken languages.
Should topics be translated or rewritten for each language?
Rewritten. A topic description portrays an article a real publisher would write, and publishers in each language write different articles about the same subject. A translated description keeps the English format and references, so it rewards sites that translate English content and misses the native publishers the language was added to reach.
Four steps to rewrite one topic for a second language
- Start from the reader's task, not the English sentence. Write down what the reader is trying to do, such as preparing a teenager for a first road test, before writing any description.
- Ask a fluent writer to describe the article that task produces in their language. Community papers, blogs and channels in that language may cover it in a different format.
- Name the local terms and institutions. Each language's press uses its own words for the product and refers to the offices, schools and services its readers deal with.
- Keep the two versions linked, not identical. Store both descriptions under one topic ID so reporting can roll up by topic while the wording stays native.
How do you review matches in a language the team does not read?
Machine tools can tell a planner what language a page is in and roughly what it says. They cannot judge whether the article is well written, whether its tone is respectful, or whether a headline joke would embarrass the brand. Those judgments need a fluent reader, named in the brief for every language.
A review routine for a language the planning team does not read
- Triage with translation, decide with a reader. A translation tool may sort obvious mismatches, but every site that stays is read by the named fluent reviewer.
- Give the reviewer the native topic descriptions. Pages are judged against the rewritten descriptions for that language, not the English originals.
- Ask for the evidence in English. For each kept site, the reviewer records matching article titles, a one line English summary of each and the reason it fits.
- Check the visible text, not the code. Confirm the language from the article body, because a declared language tag can be missing or wrong.
- Record the sign off. Note the reviewer, the date and the number of articles read per site.
Google Search Central's guide to multi-regional and multilingual sites (opens in a new tab) states that Google uses the visible content of a page to determine its language, not code-level information such as lang attributes or the URL. The W3C recommends declaring (opens in a new tab) a default language on the html element, so a correct tag signals a careful publisher, not what the article says.
How do multilingual models handle pages that mix languages?
Multilingual embedding models place many languages in one shared meaning space, so a Spanish article and an English description of the same subject land close together. The Language-agnostic BERT Sentence Embedding (opens in a new tab) paper reported 83.7 percent bi-text retrieval accuracy over 112 languages on Tatoeba, a test of finding a sentence's translation, not of judging a whole article.
Mixed language writing is harder. Code switching means moving between languages within a paragraph or sentence. The authors of the LinCE benchmark (opens in a new tab) built it because code switching methods were hardly generalizable across language pairs; Spanish and English is one of its four pairs. Expect a model's reading of such pages to be less certain, and send them to the fluent reviewer.
| Page pattern | What it looks like | Suggested treatment |
|---|---|---|
| Single language article | Body, headline and captions in one language | Count it for that language |
| Side by side translation | The same text in two languages on one page | Count it for both only if the reviewer confirms each version reads naturally; Google's same guide advises against side by side translations |
| Quotes in a second language | An English article quoting interviewees in Spanish | Count it for the article's main language |
| Code switched writing | Paragraphs that move between two languages | Send to the fluent reviewer; count it where the intended reader is |
| Translated frame only | Menus and buttons in one language, article body in another | Count it for the language of the article body |
When should each language get its own site list?
One combined list is simpler; separate lists per language are easier to control. The deciding question is whether anything downstream differs by language: creative, budget, reviewer, reporting or buying setup.
Buying platforms add another reason to split. In OpenRTB 2.6, the Content object (opens in a new tab) carries content language either as language, an ISO 639 two letter code, or as langb, a BCP 47 tag, and a seller may send either or neither. Google's language targeting (opens in a new tab) help, checked in October 2026, says that on the Display Network Google Ads may detect the language of pages or apps someone is visiting or has recently visited, and that on YouTube it relies on the languages the person understands. A campaign language setting describes people as well as pages, so a language specific site list is the clearest record of which pages each language's ads were meant for.
What does a multilingual brief look like in practice?
Fictional example
A driving school in Southern California, in three languages
A fictional driving school with branches in Los Angeles and Orange County wants display ads beside content for new drivers and their parents, in English, Spanish and Korean. The planner sets languages per market: all three in Los Angeles County, English and Spanish in Orange County, and Korean only near branches that teach in Korean.
The English topic describes a parent's guide to practicing for a first road test. Instead of translating it, the Spanish reviewer describes the first license explainers that Spanish language community outlets publish, and the Korean reviewer describes the practical guides Korean language community papers run for families and newcomers. Both sit under the same topic ID.
Code switched pages count for the language of the intended reader, as judged by the fluent reviewer. Because each language has its own creative and reporting, the plan uses three site lists, each signed off by its reviewer with matching article titles summarized in English.
How does PlacementGraph read sites in several languages?
What should be confirmed before a multilingual list is approved?
Multilingual list sign off
- Every market lists its own languages.
- Each language has topic and avoid descriptions written by a fluent writer, stored under shared topic IDs.
- Each kept site's language was confirmed from article text, not from its declared tag or its address.
- Every site in a language the planning team does not read was approved by a named fluent reviewer, with a date.
- Each content language is paired with the creative meant to run beside it.
- The one list or several lists decision is written down with its reason.
Frequently asked questions
Can one ad creative run across every language on a multilingual list?
It can, but make it a decision rather than a default. An English ad beside a Spanish article may still reach bilingual readers, yet it signals the brand did not plan for that reader. Pair each content language with a creative in the brief, and record any exception so reporting reads in that light.
Does a website's lang attribute prove which language it publishes in?
No. The W3C recommends declaring a page's default language, but a site can keep a template default. Google Search Central says Google uses a page's visible content to determine its language and ignores code-level hints. Treat a correct tag as a sign of care and confirm the language from the article body.
How should a brief specify Spanish for readers in the United States?
Name the market and the variant, for example es-US in BCP 47 style, and describe the articles US Spanish language publishers produce. That keeps the plan on publishers writing for local readers, not sites written for other countries that share the language.
Does a campaign language setting replace a language specific site list?
No. Google's language targeting help describes signals about the person, such as their settings and the pages they visit, as well as the page itself. A site list records which publishers each language's ads were planned for.
Sources
- About language targeting (opens in a new tab)Google Ads Help
- Managing multi-regional and multilingual sites (opens in a new tab)Google Search Central
- Declaring language in HTML (opens in a new tab)W3C Internationalization
- RFC 5646: Tags for Identifying Languages (BCP 47) (opens in a new tab)IETF, September 2009
- OpenRTB 2.6 specification (opens in a new tab)IAB Tech Lab
- Language-agnostic BERT Sentence Embedding (opens in a new tab)Feng, Yang, Cer, Arivazhagan and Wang, arXiv 2007.01852
- LinCE: A Centralized Benchmark for Linguistic Code-switching Evaluation (opens in a new tab)Aguilar, Kar and Solorio, LREC 2020, arXiv 2005.04322
Researched and written by PlacementGraph Research. Read how we research, source and update these pages. Corrections: diana@placementgraph.com.