Viewers rated AI-translated marketing videos less natural than videos presented by human speakers in two experiments, although their stated willingness to engage with the content did not differ significantly.
The Journal of International Marketing study compared English-to-Indonesian and Indonesian-to-English video translations. Its results show that comprehension can depend on the direction of translation as well as the production method.
The researchers used one AI tool, HeyGen, and a video promoting a fictional mobile game company. The work is an early comparison, not a ranking of every translation product or language.
Comprehension differed by direction
In the first experiment, 140 participants in Indonesia evaluated a video in Indonesian. One version used a native Indonesian speaker; the other translated an English speaker’s video using AI. The AI version scored lower on perceived comprehension, naturalness and accent neutrality.
The second experiment involved 161 participants in the United States and United Kingdom. For an Indonesian-to-English translation, participants rated the AI version higher on perceived comprehension but lower on naturalness and accent neutrality.
In both experiments, the researchers found no statistically significant difference in reported intention to like, share or comment. Intention is not a count of actual engagement, and a nonsignificant difference cannot prove that the two versions would perform identically in a commercial campaign.
What the comparison can and cannot tell marketers
Risqo Wahid, Jiseon Han, Nizar Fauzan and Heikki Karjaluoto published the paper online in 2025; it appeared in the journal’s March 2026 issue. Participants saw different speakers in the human and AI conditions, and each condition used one video stimulus. The authors say speaker characteristics and the limited set of tools and languages constrain how widely the results can be applied.
For a business adapting a campaign, a review can separate three questions: whether the words are understood, whether the delivery sounds natural to the intended audience, and whether viewers take the desired action. A version that performs well on one measure may fare differently on another.
Local speakers can check product claims, idioms and cultural context before a campaign is released. Teams can then compare versions using a predefined outcome such as comprehension, qualified inquiries or sales, with appropriate consent and safeguards for the people whose voices or likenesses are used.
Our earlier coverage of AI video ads examined viewers’ emotional responses in a separate laboratory study. Neither experiment alone establishes how all audiences will respond to AI-produced marketing.
The study’s participants assessed a fictional game promotion under experimental conditions. A product demonstration, customer support video or sensitive financial message may create different expectations, so testing the actual audience remains more informative than assuming one method will always win.
