Truace
Subscribe
70/100

Moderate evidence

Cite
More
Export CSVExport JSONEmailLinkedInX

Realistic AI avatar matched a real person on dental aftercare intent; an animated one did worse

Study title: Effectiveness of Humanized AI Avatars and Messenger Gender for Dental Postprocedure Instructions: Two Randomized Experiments

Evidence type
Randomized controlled trial
Sample
n = 906
Sources
1 sourcePeer-reviewed
Published
3 min read
Truvace record
TRV-2026-2583cite or check it

Study title: Effectiveness of Humanized AI Avatars and Messenger Gender for Dental Postprocedure Instructions: Two Randomized Experiments

Researchers ran two preregistered online randomized experiments on whether AI-generated video avatars can deliver dental implant aftercare instructions as well as a real person. In the first, 650 US adults watched identical instructions from a real woman, a near-photorealistic AI avatar or a stylized animated AI avatar, with or without being told the video was AI-generated. The realistic avatar matched the real person on intention to follow instructions (6.53 vs 6.52 on a 7-point scale).

Read the study abstract

Background: Postcare instructions play a vital role in preventive health care. Traditionally, information on post-dental procedure care is provided through printed leaflets or brief verbal explanations, which often fail to engage patients or ensure adherence. In contrast, emerging artificial intelligence (AI) systems designed to emulate human interaction represent a promising yet understudied approach to improving patient communication and understanding. Objective: This study evaluated the effectiveness of AI-ge…

57Gain side

How the two sides scoreClose call

59Problem side

The gain

In an online randomized experiment of 650 US adults, a humanized AI avatar delivering dental implant aftercare instructions matched a real person on intention to comply (mean 6.53 vs 6.52 out of 7) and on comprehension.

The problem

In the same 650-person experiment, participants who watched an animated AI avatar reported lower intention to follow aftercare instructions (mean 6.27 vs 6.52) and to return to the same dentist (5.97 vs 6.48) than those who watched a real person.

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

The animated avatar lowered intention to comply (mean 6.27) and willingness to return to the same dentist (5.97 vs 6.48 for the real person). Telling participants the video was AI-generated did not change intention to comply but reduced willingness to return to the same dentist. Comprehension stayed high in every condition.

A second experiment with 256 adults tested avatar gender. Women reported higher intention to comply after watching the man avatar (mean 6.69) than the woman avatar (mean 6.37), while men showed no clear difference. The results measure stated intentions in a hypothetical scenario, not actual patient behaviour.

Main points

  1. Experiment 1 randomized 650 US adults (mean age 44.5; 50.6% women) to five videos: real person, humanized AI, animated AI, and disclosed versions of both AI types.
  2. Intention to comply was 6.52 for the real person and 6.53 for the humanized avatar, within the preregistered equivalence margin; the animated avatar scored 6.27 (Cohen d = 0.28).
  3. Willingness to return to the same dentist fell to 5.97 with the animated avatar (Cohen d = 0.53), and disclosing AI origin also lowered it versus the real person (Cohen d about 0.35 to 0.37).
  4. Comprehension did not differ across conditions; 73% (97/132) of real-person viewers answered all five quiz questions correctly, and 22.5% of participants rewatched the video.
  5. The humanized avatar was rated like the real person on clarity, usefulness and engagement, but disclosure reduced its engagement ratings.
  6. In experiment 2 (256 adults), women's intention to comply was 6.69 with the man avatar and 6.37 with the woman avatar (Cohen d = -0.35); men showed no sizable difference.
  7. The animated avatar had larger negative effects among participants with lower education, while the humanized avatar's effectiveness was not moderated by education.

The full story

Participants were told they were having dental implant surgery and watched a video of aftercare instructions made with the Synthesia avatar platform. All videos used the same script and featured a White woman, and more than 97% spent enough time on the page to watch the whole video. Intention to comply was rated on two 7-point scales, and understanding was tested with a five-item quiz developed with a dentist. Analyses were linear regressions with robust standard errors.

Disclosure had mixed effects. For the humanized avatar, it lowered engagement ratings without clear effects on clarity or usefulness. For the animated avatar, clarity and usefulness ratings were no longer lower than the real person's after disclosure, but engagement stayed lower. Appearance ratings were lower for both AI avatars than for the real person.

In experiment 2, 256 US adults (50% women) saw the same instructions from a woman-presenting or man-presenting humanized avatar. Overall intention to comply averaged 6.36 regardless of avatar gender. Women's intention fell 0.52 Likert points more than men's with the woman avatar (P=.04), driven by women responding more positively to the man avatar. Women rated both avatars similarly on clarity, usefulness and engagement.

Understanding was high in experiment 2: 76.6% (196/256) answered all five questions correctly, with women scoring higher (mean 4.80) than men (4.45). Avatar gender did not change comprehension. The authors say humanized avatars could supplement or even replace human communicators in certain postcare contexts and may reduce clinician workload, but long-term adherence was not measured.

What this doesn’t fix

The experiments covered one clinical scenario (dental implant aftercare) with English-speaking US participants who identified as men or women, measured only stated intentions, and experiment 2 had no human control group, so it cannot show whether women's preference for the man avatar is specific to AI.

Sources

1 on file
  1. JMIR AIdoi:10.2196/85621
    Peer-reviewed2026-07-09

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

Realistic AI avatar matched a real person on dental aftercare intent; an animated one did worsehttps://truvace.com/item/effectiveness-of-humanized-ai-avatars-and-messenger-gender-for-dental-postproced

  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 an online randomized experiment of 650 US adults, a humanized AI avatar delivering dental implant aftercare instructions matched a real person on intention to comply (mean 6.53 vs 6.52 out of 7) and on comprehension.

    The problem In the same 650-person experiment, participants who watched an animated AI avatar reported lower intention to follow aftercare instructions (mean 6.27 vs 6.52) and to return to the same dentist (5.97 vs 6.48) than those who watched a real person.

    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 randomized controlled trial with a sample of 906, from JMIR AI.

    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 experiments covered one clinical scenario (dental implant aftercare) with English-speaking US participants who identified as men or women, measured only stated intentions, and experiment 2 had no human control group, so it cannot show whether women's…

    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 Health