1
00:00:00,160 --> 00:00:03,840
So my final take is that in using AI avatars in your videos,

2
00:00:03,840 --> 00:00:07,000
you really need to know your audience and know your purpose of your video. So

3
00:00:07,000 --> 00:00:10,000
if you are creating a video that is long,

4
00:00:10,400 --> 00:00:14,240
instructional screen based, it's okay to have some flexibility to try

5
00:00:14,240 --> 00:00:17,280
and add an AI avatar. But if you're creating a personal

6
00:00:17,840 --> 00:00:21,480
or sensitive or small team recording, just an update,

7
00:00:21,480 --> 00:00:25,240
it's not appropriate to replace a human presenter with AI avatars. So make sure that

8
00:00:25,240 --> 00:00:28,880
you know your audience, know your message and act appropriately in your

9
00:00:28,880 --> 00:00:32,560
videos. Good morning, good evening, good

10
00:00:32,560 --> 00:00:36,280
afternoon, wherever you are and wherever you're watching. My name is Matt Pierce, Sosa Visual

11
00:00:36,280 --> 00:00:40,040
Lounge and we are back with more AI research. That's right. If you

12
00:00:40,120 --> 00:00:43,800
listen to or watch the last episode, we talked about AI voices,

13
00:00:43,880 --> 00:00:47,480
AI voice generation and some of the impact there, some what people thought about

14
00:00:47,480 --> 00:00:51,280
them, how it can affect learning. Today we're going the next step further. We're going

15
00:00:51,280 --> 00:00:55,080
to be talking about AI avatars. So you might be on the fence here. You

16
00:00:55,080 --> 00:00:58,670
might be like, I don't know, I don't know about AI avatars. You might be

17
00:00:58,670 --> 00:01:01,510
on the side of you love them because they make your work faster and easier

18
00:01:01,510 --> 00:01:04,870
and you can produce more content. Or you might be on the other side where

19
00:01:04,870 --> 00:01:08,030
you're like, no way, I will never use an AI avatar. I don't like them.

20
00:01:08,030 --> 00:01:11,470
Whatever it might be, there is information here that you might find valuable

21
00:01:11,710 --> 00:01:15,350
based on research. So let's go ahead and jump back in and introduce our

22
00:01:15,350 --> 00:01:18,750
guest today, Stephanie Warnhoff. She's a market researcher for

23
00:01:18,750 --> 00:01:22,280
TechSmith and she has done this great research study. So Stephanie, Stephanie,

24
00:01:22,280 --> 00:01:26,120
welcome back to the Visual Lounge. Oh, thank you. Thank you very much.

25
00:01:26,120 --> 00:01:29,920
Good afternoon, Matt. Yeah, good afternoon. Well, we're going to dive in because there's

26
00:01:29,920 --> 00:01:33,760
so much here I think that's worth covering and I think this is so

27
00:01:33,760 --> 00:01:37,320
everyone for context, this is part of our AI research. We'll post the link in

28
00:01:37,320 --> 00:01:40,960
the context below so you can find that blog post in the PDF. So

29
00:01:40,960 --> 00:01:44,640
Stephanie, as you started going part through this research and you started learning from what

30
00:01:44,640 --> 00:01:48,410
people were saying, particularly with AI avatars, we what surprised you

31
00:01:48,410 --> 00:01:52,250
most in particular about what you found. So

32
00:01:52,250 --> 00:01:56,010
we ran basically the same, similar type of study. We

33
00:01:56,010 --> 00:01:59,370
had five different videos that we showed each

34
00:01:59,370 --> 00:02:02,930
individual participant. One. So one was using a human

35
00:02:02,930 --> 00:02:06,650
avatar in picture, in picture, so small circle kind of taking up, you

36
00:02:06,650 --> 00:02:09,930
know, a quarter of the screen. Then we had the human full screen which was

37
00:02:09,930 --> 00:02:13,530
more like half and half. Then we had an AI avatar that was also

38
00:02:13,530 --> 00:02:17,170
picture in Picture an AI avatar that was full screen, which is about half the

39
00:02:17,170 --> 00:02:20,270
screen. And then we had an audio visualizer, which is a

40
00:02:20,670 --> 00:02:24,270
feature that you can include in your Camtasia videos. But it is basically a still

41
00:02:24,270 --> 00:02:28,030
image with a bubble around it, for lack

42
00:02:28,030 --> 00:02:31,150
of a better word, that shows kind of the waveforms when someone is speaking. So

43
00:02:31,150 --> 00:02:34,950
it's not a moving video, but it is movement and engaging for your

44
00:02:34,950 --> 00:02:38,670
eyes. So we had those five videos and I think the thing that

45
00:02:38,670 --> 00:02:42,430
surprised me the most was that on the AI avatar side

46
00:02:42,430 --> 00:02:46,190
that participants felt like the smaller picture in picture avatar was

47
00:02:46,190 --> 00:02:49,930
actually higher quality than the full screen AI avatar. I think

48
00:02:49,930 --> 00:02:53,730
I was thinking, okay, bigger is better. So the bigger the avatar is,

49
00:02:53,730 --> 00:02:56,370
the bigger the human is, the higher quality they're going to think it is. And

50
00:02:56,370 --> 00:03:00,050
that was just not correct. It was not correct in terms of what they

51
00:03:00,050 --> 00:03:03,610
felt like for high quality. And it wasn't, it wasn't

52
00:03:03,610 --> 00:03:07,290
the biggest. When we talk about learning retention, which we'll get to later on,

53
00:03:07,450 --> 00:03:11,090
but 31% more participants felt that the smaller AI

54
00:03:11,090 --> 00:03:14,690
avatar was either good or excellent quality versus that full screen

55
00:03:14,690 --> 00:03:18,300
AI avatar. And we did have an open ended comment section there as well that

56
00:03:18,300 --> 00:03:21,900
let us know that basically with that larger on screen avatar, the tiny

57
00:03:21,900 --> 00:03:25,580
motions that make AI avatars look, I mean, kind of fake

58
00:03:25,580 --> 00:03:28,940
were more noticeable. So like the facial expressions, the kind of

59
00:03:28,940 --> 00:03:32,660
imperfect eye contact, kind of robotic sounding voice was more

60
00:03:32,660 --> 00:03:36,100
noticeable because that avatar was taking up more of the screen real estate

61
00:03:36,260 --> 00:03:40,020
on the screen. So although it's the most surprising thing,

62
00:03:40,100 --> 00:03:43,220
once I looked at the data and read through those comments, I could completely understand

63
00:03:43,220 --> 00:03:46,420
why people focused on felt that way. Yeah, well, we actually

64
00:03:46,740 --> 00:03:50,500
have the, the snippets of those. I think we're gonna. Let's play those now

65
00:03:50,500 --> 00:03:53,740
since you've described them so we can see them now. If you're a podcast listener,

66
00:03:53,740 --> 00:03:57,420
you can't see obviously through the podcast. I recommend you go check out our YouTube

67
00:03:57,420 --> 00:04:00,860
channel or on the Visual Lounge. We've been starting to post the videos from YouTube

68
00:04:00,860 --> 00:04:04,140
so you can check it out there. Anything we should know before we watch this

69
00:04:04,140 --> 00:04:07,980
beside beyond what you've already said, Stephanie? I don't think so.

70
00:04:07,980 --> 00:04:11,540
But just remember that each participant only saw one of these videos. This is kind

71
00:04:11,540 --> 00:04:14,900
of a montage of all five smashed together. So you'll notice kind of the cuts

72
00:04:14,900 --> 00:04:17,960
of, you know, or so seconds. It'll flip to another one. So this is not

73
00:04:17,960 --> 00:04:21,800
exactly what they saw, but it's one part of what they saw. We

74
00:04:21,800 --> 00:04:25,080
get the benefit of seeing all of them. But yeah, participants only saw a month,

75
00:04:25,080 --> 00:04:28,840
so. Well, let's go ahead and watch that. Google search results can be

76
00:04:28,840 --> 00:04:32,080
a bit much sometimes. A simple search like what is the best

77
00:04:32,080 --> 00:04:35,360
sunscreen? Is full of ads and profit driven biases.

78
00:04:35,680 --> 00:04:39,440
Google search results can be a bit much sometimes. A simple search like

79
00:04:39,520 --> 00:04:43,000
what is the best sunscreen? Is full of ads and profit driven

80
00:04:43,000 --> 00:04:46,760
biases. Google search results can be a bit much sometimes.

81
00:04:46,760 --> 00:04:50,600
A simple search like what is the best sunscreen? Is full of ads and

82
00:04:50,600 --> 00:04:54,120
profit driven biases. Google search results can be a bit

83
00:04:54,120 --> 00:04:57,840
much. A simple question like what is the best sunscreen? Is

84
00:04:57,840 --> 00:05:01,560
full of ads and profit driven biases. Google search results can

85
00:05:01,560 --> 00:05:04,800
be a bit much. A simple question like what is the best

86
00:05:04,800 --> 00:05:08,000
sunscreen? Is full of ads and profit driven biases.

87
00:05:09,600 --> 00:05:13,400
Okay, so we've got a few, few different options there. Obviously there's a

88
00:05:13,400 --> 00:05:16,760
human in there. There's the audio

89
00:05:16,760 --> 00:05:20,520
visualizer with the rings around it of a still image. Got lots to look

90
00:05:20,520 --> 00:05:24,320
at. Now one thing, Stephanie, you've worked on some of our other research projects

91
00:05:24,320 --> 00:05:27,960
as well. And in past research we've seen like, you know,

92
00:05:27,960 --> 00:05:31,560
most people prefer a real human over some type of

93
00:05:31,560 --> 00:05:35,240
AI avatar. Yet when we actually did this test, you know,

94
00:05:35,240 --> 00:05:39,040
this experiment here, this research, it looks like learners actually

95
00:05:39,040 --> 00:05:42,640
rated AI avatars equally or as

96
00:05:42,800 --> 00:05:46,000
equal professional kind of level as those as the humans.

97
00:05:46,720 --> 00:05:50,140
So one thing that stands out to me is that I think it was like

98
00:05:50,140 --> 00:05:53,580
92% of viewers rated avatar videos as professional

99
00:05:53,740 --> 00:05:57,020
and they would watch another video from that creator.

100
00:05:57,580 --> 00:06:01,020
What's the explanation between kind of that gap of what we've seen in past research

101
00:06:01,900 --> 00:06:05,660
and now how people are kind of judging the quality of the overall

102
00:06:05,660 --> 00:06:09,300
videos that are with avatars in them? Any sense of what's

103
00:06:09,300 --> 00:06:13,100
changing out there for people? You know, we deal with this a lot

104
00:06:13,100 --> 00:06:16,820
in research. The difference between what people either say or what they

105
00:06:16,820 --> 00:06:20,640
say they're going to do versus what they either actually do or their actual behavior

106
00:06:20,640 --> 00:06:24,160
shows. So there's research that says that the best predictor of future

107
00:06:24,160 --> 00:06:27,960
behavior is actually your past behavior. Right? So if you say I'm going

108
00:06:27,960 --> 00:06:31,520
to go to the gym every day this year, but actually the better predictor would

109
00:06:31,520 --> 00:06:33,920
be how many days out of the week did you go to the gym last

110
00:06:33,920 --> 00:06:36,440
year is a better predictor of actually what you're going to do in the future.

111
00:06:36,760 --> 00:06:40,280
So we have trouble reconciling this often with research. You know, do we, do we

112
00:06:40,280 --> 00:06:44,080
listen to what they say or do we watch what they do? And for

113
00:06:44,080 --> 00:06:47,920
this example, I'M not saying that people were incorrect when they said what they

114
00:06:47,920 --> 00:06:51,580
prefer, but I do want to point out that although this video viewer research is

115
00:06:51,580 --> 00:06:55,380
not old by any stretch, there has been huge

116
00:06:55,380 --> 00:06:59,180
advancements in AI between. Even when we ran that study at the end of

117
00:06:59,180 --> 00:07:02,860
2024 and this study that we're sharing now at the beginning of 2026, I

118
00:07:02,860 --> 00:07:05,940
mean, 18 months, things have changed so much.

119
00:07:06,819 --> 00:07:10,660
And so saying that they did prefer a human visual presenter at

120
00:07:10,660 --> 00:07:14,220
that time versus actual higher quality AI avatar

121
00:07:14,220 --> 00:07:17,930
presenter during the study is a little bit off. So they're

122
00:07:17,930 --> 00:07:21,770
not exactly comparing the AI avatars versus each other in this new study.

123
00:07:21,770 --> 00:07:24,970
Right. I mentioned they only saw one of these videos. They're not comparing the human

124
00:07:24,970 --> 00:07:28,490
to the AI, they're seeing one in isolation and basically had to

125
00:07:28,490 --> 00:07:32,290
evaluate the professionalism without comparing to what everyone else was seeing. So when

126
00:07:32,290 --> 00:07:34,930
they actually view a video and are trying to learn from it, they look at

127
00:07:34,930 --> 00:07:38,450
the screen content, they look at the size of the avatar, the voice that was

128
00:07:38,450 --> 00:07:41,810
used, you know, the facial movements, the tone of voice, everything. So this

129
00:07:41,810 --> 00:07:45,330
entire package is what viewers are evaluating when they say a video is

130
00:07:45,330 --> 00:07:48,570
professional. And I will point out that actually in this study, almost

131
00:07:48,650 --> 00:07:52,330
50% of them did not know it was an AI avatar. So the

132
00:07:52,330 --> 00:07:55,610
knowledge of it being an AI avatar didn't really affect their

133
00:07:55,610 --> 00:07:59,210
perception certainly of the professionalism or their perception of the quality.

134
00:07:59,930 --> 00:08:03,690
Yeah, which is, which is super interesting that people couldn't tell for whatever, for

135
00:08:03,690 --> 00:08:06,490
whatever reason. But you're right, the gap between

136
00:08:06,890 --> 00:08:10,650
2024 and beginning here of 2026 is,

137
00:08:11,050 --> 00:08:14,780
it's huge for AI. And so I think, yeah, it does

138
00:08:14,780 --> 00:08:18,380
make sense that maybe people's perceptions are changing. I know when you look at the

139
00:08:18,380 --> 00:08:22,220
quality of avatars or the quality of the technology, what it's able to

140
00:08:22,220 --> 00:08:25,060
produce is just a different scale, Right?

141
00:08:25,780 --> 00:08:29,140
Absolutely. So in the research,

142
00:08:29,780 --> 00:08:33,300
one of the questions or I guess we alluded to or looked at was

143
00:08:33,460 --> 00:08:37,100
when to use an avatar and when not to use avatars. Because I think this

144
00:08:37,100 --> 00:08:40,380
is important because I think a lot of us who are in the learning and

145
00:08:40,380 --> 00:08:44,210
development space, not everything is as clear cut as like just use

146
00:08:44,210 --> 00:08:47,410
it, or maybe it's not use it. We're looking for that guidance. We're still trying

147
00:08:47,410 --> 00:08:50,850
to figure these norms out of what makes sense. Are there any

148
00:08:50,850 --> 00:08:54,410
highlights from the research about when we should use avatars, not use

149
00:08:54,410 --> 00:08:58,009
avatars, any impact on maybe or

150
00:08:58,009 --> 00:09:01,450
perceived impact on trust when we are or are not using them?

151
00:09:02,570 --> 00:09:06,370
Yeah, that's a great question. We pretty much specifically did address that exact

152
00:09:06,370 --> 00:09:09,290
fact in the survey. So there was a question that basically said,

153
00:09:10,430 --> 00:09:14,190
when is it acceptable? In what style of video is it acceptable to use an

154
00:09:14,190 --> 00:09:17,950
AI avatar to you? And we got pretty clear answers on that. So in

155
00:09:17,950 --> 00:09:21,710
our study, our viewers were most accepting of an AI avatar in

156
00:09:21,710 --> 00:09:25,390
an instructional or a video that heavily featured screen

157
00:09:25,390 --> 00:09:28,990
based content. They were least comfortable when a personal

158
00:09:29,070 --> 00:09:32,750
presence was needed, like a welcome video from a CEO

159
00:09:32,750 --> 00:09:36,510
or a team update video, for example. Now, we didn't specifically ask

160
00:09:36,510 --> 00:09:39,830
about trust, but you can kind of infer that if they found an AI

161
00:09:39,830 --> 00:09:43,190
avatar acceptable, they would be like more okay with that video

162
00:09:43,190 --> 00:09:46,830
overall. So if an AI avatar is used in a video that was

163
00:09:46,830 --> 00:09:50,630
meant to distribute maybe personal or sensitive information, it could

164
00:09:50,630 --> 00:09:54,070
really turn off your viewers. So our advice here at least is to be

165
00:09:54,070 --> 00:09:57,790
intentional about the message of the videos, know your audience, and

166
00:09:58,110 --> 00:10:01,310
kind of proceed, you know, with that information in mind.

167
00:10:02,220 --> 00:10:05,820
Well, if I could follow up a little bit on that. It also seems like

168
00:10:05,980 --> 00:10:09,340
the thing that you said earlier that 50% of people didn't know

169
00:10:09,980 --> 00:10:13,340
when something was an avatar also maybe plays into

170
00:10:13,740 --> 00:10:17,579
that kind of trust issue. If I don't know it's an avatar

171
00:10:17,579 --> 00:10:21,380
and all of a sudden it's delivered maybe a very serious message that's

172
00:10:21,380 --> 00:10:25,060
maybe inappropriate for an avatar that causes potentially real issues,

173
00:10:25,060 --> 00:10:28,540
right? Yeah, absolutely. I mean, there's the

174
00:10:28,540 --> 00:10:31,860
knowledge of whether it's an AI avatar to begin with and then there's the message

175
00:10:31,860 --> 00:10:35,600
or the content that's trying to be delivered. So that's two separate factors there. And

176
00:10:35,600 --> 00:10:39,160
you're right. If they don't necessarily know it's an avatar, it can

177
00:10:39,160 --> 00:10:43,000
definitely seem sneaky, for lack of a better word.

178
00:10:43,000 --> 00:10:46,800
If you're trying to, you know, use AI as a blanket for every human

179
00:10:46,800 --> 00:10:50,600
visual presenter and every style of video and your viewers. And

180
00:10:50,600 --> 00:10:53,840
especially if it's something like I mentioned, like a sensitive topic or a small team

181
00:10:53,840 --> 00:10:57,680
update that's really going to make people, you know, sour on

182
00:10:57,680 --> 00:11:01,440
your video because it's just not the appropriate metric. You should have spent, you

183
00:11:01,440 --> 00:11:04,870
know, maybe an extra several minutes or however long it would take to be a

184
00:11:04,870 --> 00:11:08,710
visual presenter in that video to kind of humanize that video quite a bit more.

185
00:11:09,590 --> 00:11:13,350
Yeah, I'm waiting for the news articles to talk about the organization.

186
00:11:13,670 --> 00:11:17,270
Hopefully never. But you know, use uses the AI to

187
00:11:17,350 --> 00:11:20,990
avatar to lay off the workers or major

188
00:11:20,990 --> 00:11:24,670
changes. We don't look forward to that day. But obviously there are some impacts

189
00:11:24,670 --> 00:11:28,430
here. One, one of the impacts I think my audience is really interested in

190
00:11:28,430 --> 00:11:32,230
is the learning aspect. Right. So You've got lots of

191
00:11:32,230 --> 00:11:35,910
ways you could present content for people to learn from, particularly

192
00:11:35,910 --> 00:11:39,390
from what you found in this research. Was there anything interesting about the type of

193
00:11:39,390 --> 00:11:43,110
presentation of humans versus avatars and kind of

194
00:11:43,110 --> 00:11:46,230
overall effectiveness in being able to perform a task?

195
00:11:47,350 --> 00:11:50,030
Yeah, I mean, I kind of hit on this a little bit earlier too. But

196
00:11:50,030 --> 00:11:53,710
in terms of the actual learning retention, the differences between four of the five

197
00:11:53,710 --> 00:11:57,290
video types were very slight. Between the human full screen, human

198
00:11:57,290 --> 00:12:00,690
picture in picture, the avatar full screen, and the audio visualizer.

199
00:12:01,250 --> 00:12:04,330
We basically asked them kind of a pop quiz to answer a question that they

200
00:12:04,330 --> 00:12:07,450
saw in the video. And between those four types of video, there was only really

201
00:12:07,450 --> 00:12:10,890
a 3% difference in the amount of people that got the correct answer. So the

202
00:12:10,890 --> 00:12:14,690
difference was very slight. But one of them really excelled above those, and that

203
00:12:14,690 --> 00:12:18,330
is the avatar picture in picture, which actually performed 13%

204
00:12:18,330 --> 00:12:21,890
better than any of those other four examples. So

205
00:12:22,210 --> 00:12:24,970
I talked about this a little bit at the beginning, but people really rated that

206
00:12:24,970 --> 00:12:28,800
video as higher quality. But alongside that, they got, you

207
00:12:28,800 --> 00:12:32,480
know, the information that they received from that video. They were able to, you know,

208
00:12:32,640 --> 00:12:36,360
internalize and answer a question later on in the survey more correctly with

209
00:12:36,360 --> 00:12:40,000
that style of video. So like I said, they also had that

210
00:12:40,000 --> 00:12:43,680
video also had the highest number of people that felt it was

211
00:12:43,680 --> 00:12:47,160
professional, and it also had a higher learning retention. That's kind of like two stars

212
00:12:47,160 --> 00:12:50,560
for that style of video, for sure. Yeah. And, you know, I

213
00:12:50,800 --> 00:12:54,560
can think about different research from a different era. Right. Not looking

214
00:12:54,560 --> 00:12:58,190
at avatars at all, but about on camera presence and, um,

215
00:12:58,190 --> 00:13:01,470
you know, I think one thing that's always been interesting to me that with on

216
00:13:01,470 --> 00:13:05,310
camera presence and again, not avatars, the research typically shows that

217
00:13:06,030 --> 00:13:09,870
performance doesn't change, but people have a preference for it. Right. They

218
00:13:09,870 --> 00:13:13,310
like having a person there or face there, but it's never.

219
00:13:13,870 --> 00:13:17,150
It's, you know, usually a picture in picture, not all the time on screen. So

220
00:13:17,150 --> 00:13:20,550
there are some elements that I think we could probably translate. But it is interesting

221
00:13:20,550 --> 00:13:24,350
that they actually still perform better on the task. It's one of those ones. I'm

222
00:13:24,350 --> 00:13:27,950
like, Stephanie, let's do more research on that. Let's find somebody to help us do

223
00:13:27,950 --> 00:13:31,790
that, because that's super interesting about why that might be. And I know

224
00:13:31,790 --> 00:13:35,470
we don't have answers for that, but something I'm definitely, definitely curious about.

225
00:13:36,350 --> 00:13:39,870
And it does lead to the question of, like, as people are

226
00:13:40,030 --> 00:13:42,510
leaning in here, maybe using more

227
00:13:43,390 --> 00:13:46,670
avatars or want to use more avatars in their work,

228
00:13:47,390 --> 00:13:51,070
is there advice that you can give us? Obviously, you're not an instructional

229
00:13:51,070 --> 00:13:54,430
designer, you're not creating training videos. But from the research that you're seeing,

230
00:13:55,580 --> 00:13:58,780
any advice that you could give people to help them use them maybe more strategically

231
00:13:58,860 --> 00:14:02,460
or more effectively, I would say. Just

232
00:14:02,460 --> 00:14:06,140
overall, and I think I said this with the AI voice research as well, is

233
00:14:06,140 --> 00:14:09,820
if you have the flexibility and the buy in, you should try

234
00:14:09,820 --> 00:14:13,460
it. It's not a you should never use this or you should always

235
00:14:13,460 --> 00:14:17,180
use this. But my advice is to give it a try, kind of see

236
00:14:17,180 --> 00:14:20,260
how it works out for you and your audience. And this study proves to me

237
00:14:20,260 --> 00:14:23,900
that viewers are willing to watch videos with AI avatars,

238
00:14:24,060 --> 00:14:27,880
accept the style, and really believe that the qual is sometimes good or even better

239
00:14:27,880 --> 00:14:31,160
than with a human presenter. So for most cases, using an ar,

240
00:14:31,560 --> 00:14:35,120
excuse me, AI avatar will not harm your video. It could even help

241
00:14:35,120 --> 00:14:38,880
increase the information that your viewers retain. Now, there is an exception to

242
00:14:38,880 --> 00:14:42,520
that which I touched on earlier, which is if your video is personal

243
00:14:42,760 --> 00:14:46,440
or sensitive or something that really does need a human touch, we would not

244
00:14:46,440 --> 00:14:50,240
recommend using an AI avatar for that situation because the viewers have told us

245
00:14:50,240 --> 00:14:53,870
that that is not an acceptable use of that style. So I

246
00:14:53,870 --> 00:14:57,710
certainly wouldn't say replace 100% of your human presenters with AI avatars.

247
00:14:57,710 --> 00:15:01,510
But like I said earlier, I'd say know your video message, know your audience,

248
00:15:01,510 --> 00:15:04,790
and basically proceed within reason. It's,

249
00:15:05,270 --> 00:15:08,510
there's a comedian out there. I won't go into the whole story, but I'll rephrase.

250
00:15:08,510 --> 00:15:11,110
One of the things that he would say and make it for this is like,

251
00:15:11,110 --> 00:15:14,550
don't go avatar ing where you don't need no avatar in. Right? Like,

252
00:15:15,270 --> 00:15:19,120
just don't, don't, don't go there. I,

253
00:15:19,120 --> 00:15:22,800
I think there's, there's, like I mentioned at the kind of the opening, there are

254
00:15:22,800 --> 00:15:26,440
people who are very skeptical and maybe hesitant

255
00:15:26,440 --> 00:15:29,280
about avatars. And, and I love the advice, like if you have the means, you

256
00:15:29,280 --> 00:15:32,760
have the kind of go ahead to try. But

257
00:15:32,760 --> 00:15:36,400
anything specific out there for those people who are maybe saying no, I,

258
00:15:36,560 --> 00:15:39,280
I'm negative towards these. I don't. Why would I want to use them

259
00:15:40,640 --> 00:15:44,240
that we might. I. Look, I'm not trying to shift anyone's opinion here. We're not

260
00:15:44,240 --> 00:15:47,990
trying to make giant waves, but I'm curious that it's a new technology, looks like

261
00:15:47,990 --> 00:15:51,230
it has some potential. So what would we say to those folks who are a

262
00:15:51,230 --> 00:15:54,950
little bit still on that, who are on that negative side? I

263
00:15:54,950 --> 00:15:58,630
would kind of think about why they're skeptical. I'd say is it because they have

264
00:15:58,630 --> 00:16:02,190
a mistrust of AI in general. There's kind of a sentiment towards that just in

265
00:16:02,190 --> 00:16:05,630
culture today. Or is it because they have seen

266
00:16:05,710 --> 00:16:09,510
videos where the AI avatar is terrible and it turned them off

267
00:16:09,510 --> 00:16:13,190
from a video? So I think it's important to stay up to date on

268
00:16:13,190 --> 00:16:16,950
what avatars look like and how far they have advanced to look human like,

269
00:16:16,950 --> 00:16:20,570
as well as understanding basically what AI avatars can and cannot do.

270
00:16:20,730 --> 00:16:24,370
As I mentioned with the particular types of videos earlier, you cannot replace

271
00:16:24,370 --> 00:16:28,130
or you should not replace an AI avatar for a sensitive

272
00:16:28,130 --> 00:16:31,930
or personal or, you know, small group video. But if

273
00:16:31,930 --> 00:16:35,330
you're making a, you know, an instructional video that

274
00:16:35,330 --> 00:16:39,090
primarily has a lot of screen content and you're looking

275
00:16:39,090 --> 00:16:42,850
for something to provide a little visual interest or help engage

276
00:16:42,850 --> 00:16:46,490
your viewers a little bit more, our research shows that viewers would be accepting of

277
00:16:46,490 --> 00:16:50,090
an AI avatar in that situation. So if that's the type of video you're creating,

278
00:16:50,090 --> 00:16:53,870
I think you have more leeway in terms of trying to include that in

279
00:16:53,870 --> 00:16:57,510
your videos, but not on the alternative. As I mentioned, personal,

280
00:16:57,590 --> 00:17:00,790
sensitive, you know, human touch. Don't do it.

281
00:17:01,190 --> 00:17:05,030
Yeah, absolutely, Absolutely. Okay, so we've covered a lot

282
00:17:05,030 --> 00:17:08,870
of research and again, it's out there on the TechSmith blog. I'm curious,

283
00:17:09,110 --> 00:17:12,870
any questions that let's say we're looking at doing this again?

284
00:17:12,870 --> 00:17:16,390
I hope we do. I hope we look at some other kind of related

285
00:17:16,550 --> 00:17:20,260
areas. But any questions you'd want to try to answer if

286
00:17:20,260 --> 00:17:24,020
and when you get a chance to do research again? Yeah, I

287
00:17:24,020 --> 00:17:27,780
think the first thing is I would like to try and kind of compare these

288
00:17:27,780 --> 00:17:30,940
videos to each other. Now I mentioned we did not do that in the study.

289
00:17:30,940 --> 00:17:34,100
We had them watch one video and answer the questions. But I'd like to kind

290
00:17:34,100 --> 00:17:37,940
of play around with trying to have them watch maybe a human

291
00:17:37,940 --> 00:17:41,500
full screen and a human pip video and, you know, use that high

292
00:17:41,500 --> 00:17:44,940
quality voice and kind of have them rate them versus each other

293
00:17:45,100 --> 00:17:48,620
and kind of look at different aspects of that, whether, you know, we already talked

294
00:17:48,620 --> 00:17:52,440
about learning, retention, maybe engagement, you know, which one do they feel

295
00:17:52,440 --> 00:17:55,800
is more professional so that we can kind of evaluate one versus the other.

296
00:17:56,360 --> 00:17:59,840
I'd also like to dig a little bit more deeper on engagement. We did not

297
00:17:59,840 --> 00:18:03,520
really talk very much about engagement in this survey at all. But that

298
00:18:03,520 --> 00:18:06,760
is primarily the focus of the video viewer study, which we did at the end

299
00:18:06,760 --> 00:18:10,520
of 2024. So probably what my dream state would be would to

300
00:18:10,520 --> 00:18:14,320
rerun this study but include a ton of metrics about engagement

301
00:18:14,320 --> 00:18:18,040
and have them, you know, do multiple Videos versus each other, human versus

302
00:18:18,040 --> 00:18:21,850
AI engagement, professionalism, quality, and probably be a

303
00:18:21,850 --> 00:18:24,690
mega study. I don't know if my stakeholders will go for that because that seems

304
00:18:24,690 --> 00:18:26,610
like a really big project. But I think it would be cool to kind of

305
00:18:26,610 --> 00:18:30,450
mix those two and make one big study. Well, you've got my

306
00:18:30,450 --> 00:18:33,810
support. Not that that means much, but I love the research

307
00:18:33,970 --> 00:18:37,730
that you've done. And Stephanie, I think that

308
00:18:37,730 --> 00:18:41,370
the thing that's really interesting to me about this is that one, we are in

309
00:18:41,370 --> 00:18:45,130
this new era of this new technology that it is still a very

310
00:18:45,130 --> 00:18:48,830
wide open, we don't know what we don't know kind of space.

311
00:18:49,150 --> 00:18:52,830
And I'm. And I think there's a lot to learn and lot to understand.

312
00:18:52,910 --> 00:18:56,510
So I'm grateful that you were willing to dive in. You and Troy

313
00:18:56,510 --> 00:19:00,150
Stein really spearheaded this. And I'm just

314
00:19:00,150 --> 00:19:03,870
blown away that it gives me at least something, at least for now,

315
00:19:04,030 --> 00:19:07,870
knowing that in all of its imperfections, all the questions it doesn't answer because

316
00:19:07,870 --> 00:19:11,350
that's good research. It never answers all the questions. It just creates more questions.

317
00:19:11,350 --> 00:19:14,900
Typically. I love that it gives me at least a little bit of

318
00:19:14,900 --> 00:19:18,580
guidance and direction because I'll be honest, I was a little

319
00:19:18,580 --> 00:19:22,220
skeptical of avatars. And this has given me a little bit of that

320
00:19:22,220 --> 00:19:24,900
impetus to feel like, yeah, I can, I can try doing that. I should try

321
00:19:25,140 --> 00:19:28,740
using those a little bit more to see what's going to be most effective, particularly

322
00:19:28,740 --> 00:19:32,580
on those repeatable things that change often because that's, you know,

323
00:19:33,060 --> 00:19:36,860
if you got me, I can't record the same video a year

324
00:19:36,860 --> 00:19:40,060
from now and have it look and feel the same. But an avatar, pretty sure

325
00:19:40,060 --> 00:19:43,690
I can get them to be the same. So. Well, Stephanie, before

326
00:19:43,690 --> 00:19:47,530
we go into our closing, anything else that we missed or didn't cover

327
00:19:47,530 --> 00:19:51,290
that we should talk about for avatars? Probably

328
00:19:51,290 --> 00:19:54,850
not the only my last kind of final two things I like to think about

329
00:19:55,170 --> 00:19:58,650
is if you're interested in using AI avatar in video, my

330
00:19:58,650 --> 00:20:02,490
advice would be to do some research and stay up to date on

331
00:20:02,490 --> 00:20:06,210
what you can and, you know, what's out there and what's available for you. Matt,

332
00:20:06,210 --> 00:20:09,930
as you mentioned, it's moving so fast that probably the research you've done

333
00:20:09,930 --> 00:20:13,310
on Avatars 2 months ago is now may of date.

334
00:20:13,710 --> 00:20:16,470
So try and immerse yourself if you want to use it. Make sure you can

335
00:20:16,470 --> 00:20:20,190
understand how high quality they can be or, you know, what you want to include

336
00:20:20,190 --> 00:20:23,710
in your video. And my other piece of advice, as we learned from the study,

337
00:20:23,710 --> 00:20:27,510
is they are not applicable to all videos so make sure that you

338
00:20:27,510 --> 00:20:31,350
know your message and you know your audience and you choose what is

339
00:20:31,350 --> 00:20:35,030
appropriate for that video versus either, you know, going all in. We

340
00:20:35,030 --> 00:20:38,510
would not necessarily recommend that. All right, well, thank you,

341
00:20:38,510 --> 00:20:42,150
Stephanie. And so if people want to get involved in TechSmith research, we gave this

342
00:20:42,150 --> 00:20:45,210
link last episode, but I think it's helpful to do it again. And where, where

343
00:20:45,210 --> 00:20:48,850
can people connect with you and TechSmith Research? Sure.

344
00:20:48,850 --> 00:20:52,490
So you can connect with me through my personal LinkedIn page, Stephanie Warnhoff.

345
00:20:52,570 --> 00:20:56,170
Or if you are interested in more research at TechSmith, you can

346
00:20:56,250 --> 00:20:59,250
send an email to our research email address, which is just

347
00:20:59,250 --> 00:21:02,650
researchexmith.com we can get you signed up for,

348
00:21:02,970 --> 00:21:06,650
you know, in depth interviews, for beta programs, for receiving some

349
00:21:06,650 --> 00:21:10,290
surveys like this in the future. We have a lot of different research opportunities, so

350
00:21:10,290 --> 00:21:13,140
emailing that email address and we'll basically get you on the list, so.

351
00:21:13,780 --> 00:21:17,620
Perfect. Well, as we like to end most shows, Stephanie, we'd love

352
00:21:17,620 --> 00:21:21,180
to hear from you on your final take. So Stephanie Warhol, what

353
00:21:21,180 --> 00:21:24,980
is your final take? So my final take is that

354
00:21:24,980 --> 00:21:28,660
in using AI avatars in your videos, you really need to know your audience

355
00:21:28,660 --> 00:21:32,420
and know your purpose of your video. So if you are creating a video that

356
00:21:32,420 --> 00:21:36,260
is long, instructional screen based, it's okay

357
00:21:36,260 --> 00:21:39,940
to have some flexibility to try and add an AI avatar. But if you're creating

358
00:21:39,940 --> 00:21:43,420
a personal or sensitive or small team

359
00:21:43,420 --> 00:21:47,180
recording, just an update, it's not appropriate to replace a human presenter with

360
00:21:47,180 --> 00:21:50,660
AI avatars. So make sure that you know your audience, know your message

361
00:21:50,740 --> 00:21:54,540
and act appropriately in your videos. Perfect. Well,

362
00:21:54,540 --> 00:21:58,220
thank you, Stephanie. Thanks for the great research. Awesome. Thank you very much,

363
00:21:58,220 --> 00:22:01,700
Matt. You bet. All right everybody, if you're looking for

364
00:22:01,700 --> 00:22:05,460
AI avatars, also just recommend go try TechSmith

365
00:22:05,460 --> 00:22:09,170
Audio Camtasia Audit. It's got so many great cool features with the app. Got the

366
00:22:09,170 --> 00:22:12,410
avatars, you can try those. You got the new 11 lab voices which sounds so

367
00:22:12,410 --> 00:22:16,130
good. Hard to believe that they are AI. I can see why people

368
00:22:16,130 --> 00:22:19,810
maybe said that's not AI, that's a real person. So go check those out. You

369
00:22:19,810 --> 00:22:23,570
can try it for free. Or if you are using Camtasia, there's a

370
00:22:23,570 --> 00:22:26,730
bunch of audio features that are available to you. The AI features are at the

371
00:22:26,730 --> 00:22:30,570
higher level though, of course. But with that said, you know, part of this, why

372
00:22:30,570 --> 00:22:34,050
we bring forth this research is to help you get better, make better decisions, think

373
00:22:34,050 --> 00:22:37,280
through creating critically about what's going to make for good instruction, what's going to make

374
00:22:37,280 --> 00:22:40,600
for good video. And of course in doing that process, you got to just keep

375
00:22:40,600 --> 00:22:44,360
working at it. Keep trying and keep exploring and keep trying to get better

376
00:22:44,360 --> 00:22:47,760
every single day. And with that said, I hope you take a little time

377
00:22:47,920 --> 00:22:50,720
to level up every single day. Thanks, everybody.