Truace
Subscribe
68/100

Moderate evidence

Cite
More
Export CSVExport JSONEmailLinkedInX

Fraud warnings tailored to personality by AI worked as text (296 people) but not as voice (191)

Study title: Large Language Models Can Assist in Generating Personality-Matched Anti-Fraud Voice Warnings in the Chinese Context

Evidence type
Controlled or validated comparison
Sample
n = 487
Sources
1 sourcePeer-reviewed
Published
2 min read
Truvace record
TRV-2026-2527cite or check it

Study title: Large Language Models Can Assist in Generating Personality-Matched Anti-Fraud Voice Warnings in the Chinese Context

Researchers tested whether the large language model DeepSeek can write anti-fraud warnings matched to a person's Big Five personality traits. In Study 1, with 296 participants, text-only matched warnings were rated as significantly more effective, and the effect was clearest for Openness.

Read the study abstract

With the rise of AI-generated content, personalized interventions have become increasingly feasible in high-risk decision-making contexts such as online fraud. This research examined whether large language models (LLMs), exemplified by DeepSeek, can generate personality-matched voice warnings to enhance anti-fraud persuasion in the Chinese context. Across two studies, warnings were tailored to Big Five personality traits by manipulating either text content only (Study 1, N = 296) or both text and vocal parameter…

57Gain side

How the two sides scoreClose call

59Problem side

The gain

In a Chinese study of 296 people, anti-fraud warnings that DeepSeek tailored to personality traits were rated significantly more effective than unmatched ones, most clearly for Openness.

The problem

In a second Chinese study of 191 people, personality-matched warnings showed no significant advantage when DeepSeek generated both the text and the vocal parameters.

A fixed formula sets both scores from the evidence and its sources. Reader votes don’t move them.

What the scores mean

G is the gain side, out of 100. It goes up when the gain is specific, when it measurably helped people and when the sources behind it are strong.

P is the problem side, out of 100. It goes up when the problem is specific, when it measurably hurt people and when the sources behind it are strong.

In brief

Study 2, with 191 participants, extended the approach by having the model also set vocal parameters for spoken warnings. It found no significant personality-matching effects, so the benefit seen in text did not carry over to voice.

The outcome measured was perceived warning effectiveness, not whether people actually avoided fraud. The authors conclude the models may support personality-based warning generation under controlled text-only conditions, while effectiveness remains limited for multidimensional voice-warning design.

Main points

  1. Study 1 enrolled 296 participants and varied only the text content of warnings by Big Five personality trait.
  2. Personality-matched text warnings significantly increased perceived warning effectiveness in Study 1.
  3. The clearest matching effect in Study 1 was for the Openness trait.
  4. Study 2 enrolled 191 participants and had DeepSeek generate both text and vocal parameters.
  5. Study 2 found no significant personality-matching effects for the combined text and voice warnings.
  6. The work was set in the Chinese context and used DeepSeek as the example model.

The full story

The research asks whether AI-generated content can make personalised interventions practical in high-risk decisions such as online fraud. DeepSeek was used to generate warnings matched to Big Five personality traits, tested in the Chinese context.

Study 1 manipulated only the text of the warnings among 296 participants. Matched warnings raised perceived effectiveness significantly, with Openness showing the clearest effect.

Study 2 added vocal parameters among 191 participants, with the model generating both text and voice settings. No significant matching effects appeared, which the authors read as limited effectiveness for multidimensional voice-warning design.

What this doesn’t fix

The studies measured perceived warning effectiveness, and the personality-matching benefit disappeared when the model also generated vocal parameters.

Sources

1 on file
  1. International Journal of Human–Computer Interactiondoi:10.1080/10447318.2026.2674290
    Peer-reviewed2026-05-28

Talk it over

Four questions to start onThis record found a gain and a problem in the same place. Which matters more to you, and who might answer differently?

Start with a question

Fraud warnings tailored to personality by AI worked as text (296 people) but not as voice (191)https://truvace.com/item/large-language-models-can-assist-in-generating-personality-matched-anti-fraud-vo

  1. This record found a gain and a problem in the same place. Which matters more to you, and who might answer differently?

    The gain In a Chinese study of 296 people, anti-fraud warnings that DeepSeek tailored to personality traits were rated significantly more effective than unmatched ones, most clearly for Openness.

    The problem In a second Chinese study of 191 people, personality-matched warnings showed no significant advantage when DeepSeek generated both the text and the vocal parameters.

    Answer in the thread

  2. How do we know? What can this kind of evidence show, and what can’t it?

    The evidence This rests on a controlled comparison with a sample of 487, from International Journal of Human–Computer Interaction.

    Answer in the thread

  3. What does it leave open? What else would you want to know before acting on it?

    The record’s own caveat The studies measured perceived warning effectiveness, and the personality-matching benefit disappeared when the model also generated vocal parameters.

    Answer in the thread

  4. Have you seen AI used this way where you study, work or live? Did it go the same way?

    Answer in the thread

The questions are the same on every record. The details under them are taken from this record as written.

What readers say

More from Crime