Language settings, a voter-file model, or the content someone is already watching. Only two of them work, and they solve different problems.
By Bill Redding, CEO and Founder, WKQ Media
In August I sent an email to our data partner INTRVL about how we should deliver Spanish-language ads on two congressional programs we were about to launch. It kicked off a back-and-forth that laid out, more plainly than I have seen anywhere else, the different ways a media buyer can put a Spanish ad in front of a Spanish speaker. Here is what came out of it, along with what we have seen when we ran each one.
The default: device and in-app language settings
Most platforms’ language targeting works the same way. You pick “Spanish,” and the platform serves your ad to people whose device, browser, account or app is set to Spanish. Google’s Display & Video 360 documentation says as much — the targeting is based on the user’s language setting, which is why a Spanish speaker can see a Spanish ad on an English-language site.
The problem is who sets their device to Spanish. Pew Research Center’s work on language use among Latinos puts the picture in three rough buckets: about 36 percent bilingual, 25 percent mainly English, and 38 percent mainly Spanish. Roughly seven in ten U.S. Hispanics speak English fluently. Those voters bought their phones in English, set them up in English and left them that way. Device targeting finds the Spanish-dominant third and misses most of the rest, including bilingual voters who watch a great deal of Spanish-language television with a phone in their pocket that says otherwise. It also drifts: an account language changed after a trip, or a device synced from a relative, ends up assigning ads to the wrong person. It is a starting point, not a plan.
The model: language preference matched to the voter file
The second way is to stop asking the device and start asking the voter. On the New Jersey program we ran for Make the Road Action in 2025, INTRVL, working with DSPolitical, built a language model from thousands of digital surveys, then matched the modeled data to the voter file. The output was a named universe of voters who communicate in Spanish, scored individually and matched across connected TV, online video and social.
The model went one level further, into dialect. The word for the thing you drink through varies across the Spanish-speaking world — popote in Mexico, pajita across much of Latin America, sorbeto in Puerto Rico — and asking which word someone uses tells you which community they belong to, which tells you whose voice and accent should be in the ad. Creative was built for the community the model identified, not translated once and run everywhere.
Roughly a fifth of delivery ran in Spanish, with the English–Spanish split set by the model rather than fixed in advance. Measured against a 200,000-voter control group after the election, trust on immigration rose 8.1 points and trust on personal finances rose 8.9 among the voters we reached.
That is the most precise way to do this, and precision is worth paying for when the audience is small enough to saturate and the program will be measured. It carries a data fee on every impression, and a separate Spanish list is a smaller pool to bid into, which pushes CPMs up on its own. On one earlier set of districts a large share of the Spanish budget went to the data fee and increased CPM instead of more impressions.
The content: deliver in the language someone is already consuming
In Arizona this July we took a different approach from both. We worked with the Secretary of State’s advertising agency, RIESTER, to promote statewide BallotTrax signups. The program was nonpartisan, and had one job: get registered voters to sign up for ballot tracking. We used content and contextual flags on connected TV to serve the Spanish creative whenever a device was already viewing programming in Spanish — Telemundo’s World Cup coverage, Spanish-language streaming, Spanish-language news. The viewer told us their preference by pressing play. No data fee was paid to find them, so Spanish CPMs landed where the English CPMs landed.
Spanish creative featuring local soccer players and the Secretary’s own voice in Spanish ran from the first day of the flight and accounted for 13.9 percent of all delivery. Signups in the 13 counties the Secretary of State administers grew 84 percent across the flight window.
What content targeting cannot do is dialect. It knows the program is in Spanish; it does not know whether the household is Sonoran or Puerto Rican. In practice that matters less than it sounds. Producing creative for multiple dialects is time-consuming, and outside a handful of markets where several large Spanish-speaking communities sit side by side — New York, Miami, Los Angeles — one well-made Spanish spot serves the audience.
The same principle reaches languages no one has modeled. In a Michigan candidate program last cycle we needed Arabic and Hindi in the final week. Nobody builds a voter-file model for either, and almost no one in those communities has a device set to Arabic or Hindi. But YouTube lets you set the language of the creative itself, and it serves that creative to people consuming some of their content in that language. The campaign built Arabic and Hindi spots and we set each accordingly. People who watch part of their day in Arabic saw the ad in Arabic, even though their phone said English.
None of it works without the creative
Every one of these methods is a delivery mechanism, and a delivery mechanism cannot rescue a bad ad. Slapping an AI voice-over or a hurried dub in a second language onto the English spot does not move anyone. It reads as exactly what it is.
The creative has to be built from the ground up to speak to someone in English, Spanish, Arabic, Hindi or whatever they prefer — the messenger, the setting, the idiom, the accent. On the New Jersey program, where a message worked in both languages it was made twice, not translated. One spot ran only in Spanish and one only in English, because that is where each belonged. In Arizona the Spanish spots were their own spots, with separate video and VO of the Secretary explaining the importance of BallotTrax. That is what the delivery method is carrying, and it is the part that determines whether the number at the end moves.
Where this leaves a buyer
Device settings will find the Spanish-dominant voter and miss the bilingual one. A voter-file model will find both, at a cost, and can tell you which Spanish to speak. Content signals will find anyone consuming in a language, in any language a platform can flag, at no premium, without the dialect.
Most programs we run now use the last two together: the model where its precision is worth the fee, content and contextual delivery layered across the whole audience so that anyone watching in Spanish gets the Spanish ad whether or not the model flagged them. Either way, the goal is the same. Put the ad in the language a person actually prefers, even when their phone says English, and more of them watch it, click it, sign up, or come around.
If you are planning a program this fall with an audience that does not all speak English at home, I am at bill@wkq.media.
