AAPI Victory Alliance cut multilingual campaign drafting from 45 to 15 minutes

Political campaigns serving Asian American and Pacific Islander voters faced a practical bottleneck: translators for AAPI languages were harder to find than Spanish translators, and many of the languages had little training data for large language models. Under-resourced campaigns needed a faster way to adapt messages without treating accuracy as automatic.

In December 2023, AAPI Victory Alliance used a $20,000 Higher Ground Labs grant to run a three-week test. The team built a multilingual chatbot with ChatGPT and Claude, translated high-performing Democratic campaign messages into Hindi, Tagalog, Chinese, Hmong, Korean, and Vietnamese, then asked the models to reformulate them for specific communities. Humans remained responsible for prompts, review, and the final copy.

The workflow reduced a translation-and-refinement task from about 45 minutes to 15. In final tests, AI translations performed roughly on par with human translations, winning 52% of the time; Hindi reached 58%. The decision produced cheaper, faster message capacity, not proof that AI could replace native speakers or deliver electoral wins.

AI cut translation and refinement time from 45 minutes to 15; in final tests, AI versions won 52% of comparisons and 58% for Hindi.

AAPI campaigns faced a shortage of translators for languages with relatively little data available to train conversational models.

A three-week grant-funded pilot limited the risk of adopting an unreliable tool across live campaign communications.

Starting from messages that had already performed well gave the team a concrete baseline for testing AI versions.

Native-speaker review kept model errors, cultural mismatches, and bias from moving straight into voter outreach.

Use AI to multiply scarce-language campaign capacity, but keep native speakers responsible for prompts, testing, and the final message.

The pilot showed parity with human translations but not a universal advantage: Hindi performed better than the six-language average, while experts warned that low-resource languages and biased prompts still required human control. AAPI Victory Alliance's result was therefore a workflow for extending scarce language capacity, with review built in, rather than an autonomous campaign system.