Human Narrator vs Digital Avatar: The Gap Sits in the Credibility Pathway, Not the Information-Quality Pathway
In an online experiment with 601 participants, a human narrator was perceived as more immediate and clearer than a digital avatar in short videos about intangible cultural heritage. Among the four structural links compared between the two conditions, only those in the credibility pathway differed significantly, and were stronger for the human narrator; those in the information-quality pathway did not differ significantly, and the link from inspiration to electronic word-of-mouth intention was not compared between conditions.
A Gap Concentrated in the Credibility Pathway, Not the Information-Quality Pathway
In an online experiment where 601 participants were randomly assigned to a short video about intangible cultural heritage (Hakka tulou), the human narrator was perceived as having higher nonverbal immediacy (M=3.745 vs 3.397, d=0.474) and higher clarity (M=3.725 vs 3.355, d=0.497) than the digital narrator, on a 1-to-5 scale, both differences being significant (p<0.001).
Across the full sample, five structural links were positive and significant: nonverbal immediacy to narrator credibility (β=0.249), perceived clarity to perceived information quality (β=0.264), credibility to customer inspiration (β=0.174), information quality to inspiration (β=0.154), and inspiration to electronic word-of-mouth intention (β=0.358). The multigroup analysis compared four of these links between the human and digital narrator conditions; the inspiration-to-word-of-mouth link was not compared between conditions. Among the four compared links, only the two in the credibility pathway (immediacy→credibility, credibility→inspiration) differed significantly by narrator type, and were stronger for the human narrator, while the two links in the information-quality pathway did not differ significantly between the human and digital conditions.
Nonverbal immediacy and perceived clarity were also linked to inspiration through a direct effect (β=0.195 and β=0.179) and a smaller indirect effect running through credibility (β=0.026) and perceived information quality (β=0.024) respectively; the authors describe this pattern as complementary mediation. The chain extending to electronic word-of-mouth intention remained significant for the credibility pathway (β=0.009, p=0.023) but not for the information-quality pathway (β=0.009, p=0.059).
Viewers' focused attention moderated both pathways to inspiration: at low attention, neither narrator credibility nor perceived information quality was significantly related to inspiration, whereas at high attention both links became significant and markedly stronger (credibility: β=0.303; information quality: β=0.276).
What the Measurement Actually Supports
Reliability and convergent validity of the scales used were generally satisfactory (Cronbach's alpha between 0.765 and 0.821, average variance extracted between 0.527 and 0.583, HTMT ratios all below 0.80), but only partial measurement invariance could be established between the human and digital conditions, and the model's out-of-sample predictive ability, although positive (Q²_predict > 0), did not consistently outperform a simpler linear benchmark model.
The manipulation check, which asked participants to identify the type of narrator they had just seen, was passed by only 53.33% of participants in the human condition and 53.49% in the digital condition, barely above chance, which cautions against assuming viewers consciously recognized and distinguished the two narrator types.
What a Single Pair of Videos Cannot Establish
The experiment compares only one human narrator and one digital narrator, each embodied in a single video, without independently manipulating cues such as voice, movement, or gaze, and without a formal stimulus pretest; the digital narrator further came from a specific production pipeline, in which GPT analyzed the original video and generated prompts, Seedance 2.0 produced the digital video, Suno AI reconstructed the voice, and Adobe Premiere Pro handled final editing. The results therefore reflect a comparison between two complete presentations and this particular production pipeline, not the isolated effect of a narrator cue or a general category of digital avatar.
Because mediators and outcomes were all self-reported in the same post-viewing questionnaire, the authors note that the observed relationships should be read as associations rather than cause-and-effect relationships, and that conclusions generalize cautiously beyond the Hakka tulou context, this specific narrator pair, and this convenience sample of 25-to-38-year-olds recruited in mainland China. R² values also remain modest (from 0.062 for credibility to 0.192 for customer inspiration), which the authors themselves note indicates that narrator-related cues explain only part of participants' responses.
A Lead for Choosing a Narrator, Not a Validated Recipe
According to the authors, the choice between a human and a digital narrator should align with the communication goal: a human narrator would be preferable when the goal is to strengthen perceived credibility and the inspiration that follows from it, whereas attention to clarity and perceived information quality would remain important regardless of narrator type when the goal is knowledge transmission, since this second pathway did not differ significantly between the two tested conditions.
The broader design ideas the authors mention, such as synchronizing voice, lip movement, gaze and gestures, or using highlighted subtitles and segmented knowledge units, draw on existing literature and were not directly tested in this study; they remain recommendations to validate empirically, not a finding from this experiment.
Research & evidence
The study analysed
Human and digital human narrators in intangible cultural heritage short videos: dual psychological evaluation pathways through customer inspiration to electronic word-of-mouth intention
Read the original study Full textContinue reading
All analyses →
