[00:00]
Everyone says AB testing your thumbnail
[00:02]
and titles is the smartest thing you can
[00:04]
do for your YouTube growth. YouTube
[00:05]
literally built the feature for you, so
[00:08]
of course you should use it, right?
[00:09]
Wrong. Because if AB testing is supposed
[00:11]
to help your video perform better, then
[00:14]
why did six out of seven of my videos
[00:16]
completely tank the moment I turned it
[00:19]
on? I was just like you. I saw the big
[00:21]
creators raving about it. Mr. Beast uses
[00:24]
it. So, I thought, "Okay, let's me go
[00:26]
all in." I ran AB testing on seven brand
[00:29]
new videos back-to-back. I trusted the
[00:32]
tool and one by one I watched those
[00:35]
videos die. But, what if AB testing
[00:37]
itself isn't broken? What if YouTube is
[00:40]
quietly punishing you every single day
[00:43]
the test is still running and you don't
[00:46]
even know about it? Here's what I found.
[00:48]
When YouTube runs your test, it splits
[00:50]
your traffic equally between all your
[00:51]
variations. So, if you're testing three
[00:54]
titles, 66% of your audience is seeing
[00:57]
an experimental title that might be
[00:59]
terrible. And YouTube is judging your
[01:02]
video the entire time this is happening.
[01:05]
So, in this video I'm going to show you
[01:06]
the actual math behind why this tool can
[01:09]
destroy your video's momentum and who
[01:12]
should and honestly should not be using
[01:15]
it. But first, let's me quickly break
[01:17]
down how AB testing actually works in
[01:19]
2026 because YouTube just made some big
[01:22]
changes to it. As of 2026, YouTube is no
[01:26]
longer just AB testing thumbnails. You
[01:28]
can now test both titles and thumbnails.
[01:31]
You can test up to three different
[01:33]
titles, three different thumbnails, or
[01:36]
three combinations of titles and
[01:38]
thumbnails. How cool is that? One thing
[01:40]
most people don't understand is YouTube
[01:42]
doesn't pick the winner based on the
[01:43]
click-through rate or CTR. Instead,
[01:46]
YouTube picks based on watch time and
[01:49]
average view duration. That's quite
[01:51]
interesting, isn't it? I mean, it is
[01:52]
called AB testing, so wouldn't they want
[01:55]
to track clicks instead? The truth is
[01:58]
YouTube is doing this because YouTube
[01:59]
doesn't want clickbait. For example,
[02:02]
take a look at these three thumbnails.
[02:04]
If version A gets 10% clicks, but
[02:06]
everyone leaves in 5 seconds because it
[02:08]
is clearly clickbait, and version B gets
[02:11]
2% clicks, but they watch for 50% of the
[02:14]
video, then version B wins even though
[02:17]
the CTR is significantly lower. And
[02:20]
here's what I discovered when running AB
[02:22]
testing on seven newly launched videos.
[02:24]
Based on my tests, all six out of my
[02:28]
seven videos of mine underperformed,
[02:30]
failing to gain 1,000 views. You see,
[02:33]
the biggest issue with YouTube AB
[02:35]
testing is if it can't find a clear
[02:37]
winner, it drags the test up to a
[02:40]
maximum of 14 days. That's actually a
[02:42]
trap because it damages your channel
[02:45]
each day it drags. Let me explain to you
[02:47]
why. Check out this test I did for one
[02:50]
of my videos. Nothing fancy, just a
[02:52]
title test. Based on the final result,
[02:55]
it is clear my top title is the winner
[02:58]
with over 46%
[03:00]
watch time share. But you see, if you
[03:01]
squeeze your eyes real good, you'll
[03:04]
notice that this test took over 11 days
[03:07]
to conclude. So, just by thinking of
[03:09]
this logically, the math doesn't add up.
[03:11]
You see, in order to conclude which
[03:13]
title is the best, YouTube needs to test
[03:15]
all three of them, meaning my traffic is
[03:18]
split 33 33 33. If I have one clear
[03:22]
winner, then that makes the other two
[03:24]
experimental titles. And those two
[03:26]
experimental titles are shown to 66%
[03:30]
of my audience. So, let's do a simple
[03:32]
logical walk-through here. YouTube
[03:34]
starts with the top one, found that it
[03:36]
works, has over 40% watch time. But,
[03:39]
YouTube doesn't know if this is the best
[03:41]
yet, so it pauses the top one, then
[03:44]
tests the second title. Oh, no, the
[03:46]
second one did not perform well after 3
[03:48]
days. So, let's move to the third one.
[03:51]
Again, oh, no, the third one failed,
[03:53]
too. So, that makes the first one the
[03:56]
clear winner. And the entire test took
[03:59]
11 days. So, tell me, if you knew the
[04:01]
first one is the winner, then you would
[04:04]
stop the test and just use the first
[04:06]
title, right? But, you can't do that if
[04:08]
you are AB testing and I think that's a
[04:11]
massive flaw in this feature. Moreover,
[04:13]
I have a very bad feeling that when
[04:15]
YouTube is testing my experimental
[04:17]
titles and both underperformed, YouTube
[04:20]
might have received the signal that the
[04:22]
video is not a good one. Therefore, when
[04:25]
it decides that the first one is the
[04:27]
winner, the damage has already been
[04:29]
done. It is already too late. YouTube
[04:31]
thinks this video is bad because of your
[04:33]
experimental titles. The video has
[04:35]
already lost its initial first 48 hours
[04:38]
boost, so YouTube decides to not give
[04:41]
any more impressions to it and the video
[04:44]
fades away. I definitely do not have
[04:45]
concrete proof of that, but that's what
[04:48]
I saw in my AB testing and I hope that
[04:51]
isn't true. That's not to say this is
[04:52]
entirely YouTube's fault. There's
[04:54]
certainly an angle where it doesn't make
[04:56]
sense to use AB testing, especially for
[04:59]
small YouTube channels. And when I say
[05:01]
small YouTube channels, I don't mean by
[05:03]
subscriber count. I mean by view count.
[05:06]
You have probably heard that AB testing
[05:09]
has been a game-changer, that it helped
[05:11]
many videos boost their CTR and
[05:14]
subsequently more views. What you might
[05:16]
be aware of as well is that those
[05:18]
sayings often come from big channels.
[05:21]
So, what this means is that you need a
[05:22]
big amount of views in order to conclude
[05:25]
your AB testing effectively because if
[05:28]
you don't have sufficient views, then
[05:29]
your test will not only drag on, but it
[05:32]
will also fail to come to an accurate
[05:35]
conclusion. From my own experience and
[05:37]
testing, I highly suggest using AB
[05:39]
testing only if you can average 1,000
[05:42]
views per newly launched video,
[05:45]
preferably about 1,000 views in the
[05:47]
first hour. The more, the merrier, of
[05:49]
course. The reason is because you want
[05:51]
to have enough views to really run the
[05:52]
AB test to the fullest and also to have
[05:55]
AB testing conclude as soon as possible,
[05:58]
preferably within the first day. Heck,
[06:00]
even 1,000 views is me being optimistic.
[06:03]
I'm actually thinking about 10,000 views
[06:05]
average. So, if you're getting less than
[06:07]
1,000 views per video, I suggest you
[06:09]
stop using AB testing and instead use
[06:12]
this strategy pivot. My first strategy
[06:15]
is if I choose to use AB testing, I will
[06:18]
let it run for the first 24 hours only.
[06:21]
Then, I will evaluate the test so far
[06:24]
and manually select the one that is the
[06:26]
clear winner. I will not let it drag on
[06:28]
for more days. My other strategy is to
[06:30]
avoid AB testing altogether and do
[06:32]
manual swaps. So, if I am testing the
[06:35]
first title at launch, I will check it
[06:37]
again after 24 hours. If it fails, I
[06:40]
will manually swap to the second title
[06:43]
and test it. After that, I will pick and
[06:45]
go with my gut feelings based on which
[06:47]
gets more views. The new video boost
[06:49]
usually lasts for the first 48 hours or
[06:52]
so, so this to me is quite a good test
[06:55]
provided you are disciplined enough to
[06:57]
do it manually. And also, if you don't
[06:59]
have the traffic to support your tests,
[07:01]
then avoid testing on brand new videos.
[07:04]
You have already put in a lot of effort
[07:06]
and time into it, so don't let the video
[07:08]
underperform because of YouTube's tool.
[07:10]
Another strategy I tried is to look out
[07:13]
for underdogs in your channel, such as
[07:15]
videos with high impressions, but low
[07:17]
CTR. Videos like this actually signal
[07:20]
that YouTube sees a market in it, but no
[07:22]
one is clicking because of a bad
[07:24]
thumbnail or title. And I think AB
[07:27]
testing does help when it comes to this.
[07:29]
In fact, I actually tested this on one
[07:30]
of my videos. It is rather old, a few
[07:32]
years old, so I just did some tests on
[07:34]
it just to see what happens. But so far,
[07:37]
it doesn't look like I managed to revive
[07:38]
it, although I haven't left it long
[07:40]
enough yet, so maybe YouTube still needs
[07:42]
time to follow this one up, so we'll
[07:44]
see. So, yes, I am very disappointed
[07:46]
with YouTube's AB testing. I really wish
[07:48]
it worked for everyone, big or small
[07:50]
channels, but it is very clear to me
[07:52]
that I should stop doing any more AB
[07:55]
testing for now and instead rely on my
[07:58]
instinct. Seven videos used to test AB
[08:00]
testing is a lot of videos, but hey, I
[08:03]
think I gained quite a lot from it and I
[08:06]
hope you do, too. Also, I run YouTube
[08:07]
channel audits as well for my viewers,
[08:09]
where I will audit and give my most
[08:11]
honest feedback to growing YouTube
[08:13]
channels, help give them a direction,
[08:15]
and help them grow their channel. If
[08:17]
you're interested, check out the link in
[08:19]
the description. So, this concludes my
[08:21]
AB test. The next test I will do is
[08:23]
going to be what happens if I were to
[08:25]
upload seven videos without AB testing.
[08:28]
Is my gut feeling more accurate than
[08:30]
some YouTube tool? That's what I'm going
[08:31]
to test next, so when the video is
[08:33]
ready, it will appear somewhere over
[08:35]
here. So, be sure to check it out. All
[08:38]
right, thanks for watching and I'll
[08:39]
catch you in the next one.