In 2025, DSPolitical introduced a major advancement in civic engagement: the Language Preference Model. This data-driven targeting framework allows campaigns and public affairs initiatives to reach voters based on how they actually prefer to consume digital media. First launched in 2024 as the Spanish Language Preference Model, built in partnership with INTRVL, it expanded in 2025 to include five additional languages beyond Spanish in New York City — Korean, Mandarin, Cantonese, Russian, and Bengali — redefining what’s possible in multilingual outreach.

The model was built to solve a long-standing industry problem: outdated, imprecise methods for identifying language preference that didn’t just misclassify voters. They systematically hobbled efforts to reach linguistically diverse communities, leaving many residents overlooked, misunderstood, and effectively written out of engagement strategies.

The Challenge

For decades, campaigns relied on flawed and unreliable tools and proxies to infer language preferences, such as:

• Browser or device settings
• ZIP codes
• Surname or ethnicity lists
• Broad demographic assumptions

These methods often missed the mark. English and in-language ads went to voters who preferred content in different languages. The result was predictable: wasted budget, weak engagement, and uneven access to essential civic information.

Some linguistically diverse communities, such as Bengali speakers, weren’t targetable at all on major digital platforms, even when using flawed traditional methods, leaving organizations unable to invest in meaningful in-language outreach.

Our Approach

DSPolitical and INTRVL partnered up to tackle the problem as a data challenge, not a translation one.

The Language Preference Model uses digital survey responses to assess an individual’s preference for consuming media in a specific language, dialect, or combination of languages, placing each person on a sliding scale from English-dominant to bilingual to fully in-language. This preference-based method offers three key advantages:

• Accuracy: Directly measures preference instead of relying on guesswork.
• Nuance: Identifies bilingual and dialect-specific audiences that other models miss.
• Inclusion: Makes targeting possible for languages without existing digital identifiers.

This scalable model integrates with major digital ad platforms, allowing campaigns to deliver preference-aligned content across CTV, online video, social media, and display.

Hear directly from DSPolitical’s Vice President of Media, Jacob Garber, as he shares how the Language Preference Model is reshaping modern civic engagement at MediaPost’s Marketing: Politics Conference. Watch the full conversation below.

Implementation in New York City

Ahead of the 2025 elections, the New York City Campaign Finance Board needed a way to reach residents across dozens of linguistically diverse neighborhoods with clear, nonpartisan civic information. With a city mandate to ensure essential voting and election information was accessible in 13 languages, traditional outreach methods just couldn’t meet this need.

To close that gap, DSPolitical deployed its expanded language models in partnership with Fenton to support the NYC Campaign Finance Board’s 2025 voter-engagement, turnout, and education programs.

Although compelling creative assets were produced in several languages, the breakthrough came from the targeting infrastructure itself. For the first time, the campaign could reliably reach voters based on how they actually wanted to consume media, not on assumptions, proxies, or platform constraints. This shift ensured multilingual voter information reached communities that had long been underserved.

Turnout & Engagement

Use of the Spanish Language Preference Model in 2024 achieved statistically significant, promising results:

• 1.05% overall turnout lift
• 1.82% turnout lift among Spanish-preferring voters aged 18–29
• 1.44% turnout lift among Spanish-preferring male voters

Expanded Access to Underserved Communities

During the 2025 New York City program expansion, in-language impressions increased substantially:

• Bengali: +790%
• Korean: +1,600%
• Russian: +211%

These gains emphasize a fundamental shift: preference-based targeting didn’t just improve multilingual outreach. It gave underserved communities access to powerful civic information they would otherwise have missed.

Lessons for the Advocacy Field

1. Assumptions create inequities. Data corrects them.
   Proxy-based targeting systematically excluded linguistically diverse voters. Direct measurement restores accuracy and fairness.
2. Evidence-driven innovation delivers outsized results.
   First-party survey data produced measurable increases in turnout and more efficient program spending.
3. Inclusion must be built into the infrastructure.
   The Language Preference Model addresses structural gaps in digital platforms, enabling scalable access to historically underserved communities.
4. Multilingual audiences are not monolithic.
   Identifying where voters fall on a language continuum leads to more precise and more respectful outreach.

Conclusion

DSPolitical’s Language Preference Model represents a significant leap forward in multilingual civic communication and engagement. By shifting from flawed inference to defined preference, it delivers the accuracy, efficiency, and equity that modern advocacy demands, opening the door to more inclusive outreach at a scale the industry hasn’t seen before.