Does Where the Face Sits on the Thumbnail Get More Views?
A recent LinkedIn post from a well-known YouTube thumbnail designer confidently claimed that faces in thumbnails should always go on the left and that you’ll get more views if you follow the advice.
The reasoning appears sound at first: most US-based viewers read left-to-right, top-to-bottom and a face at the left would be seen first. This pairs with the idea that a face is necessary to get attention in a thumbnail. That latter claim, that thumbnails with faces perform better, did not hold up in our recent studies: our first study tested whether to use a face at all (within a channel, a face was tied to slightly fewer views, the opposite of the advice), and the second tested how big to make it (essentially no effect).
To round out the research on how faces on thumbnails correlate with views we’ll study if the placement of the face on the left, center, or right has any correlation with increased views.
We classified the face as left, center, or right of frame, then compared each channel’s videos against its own, holding age constant. For the left-vs-right placement the advice is about, placement does not move views by more than about 6%, and our best estimate is no effect at all.
Moving the face from one side to the other doesn't change views in any way we can measure.
Our best estimate is about 0% (−0.3%), and we're 95% sure the real number is between roughly −5% and +6% — small either way, with our best guess being none.
This is not an absence of evidence: with ~380 channels that vary which side they place the face and ~16,600 videos, the interval is tight. But “small” is not “zero.” If a 5% lift in views is real money at your scale, this study does not clear that bar for you. It rules out a large side effect; it cannot rule out a few-percent one. For most teams that makes side placement a non-decision.
What we did
Section titled “What we did”We measured the single face’s horizontal position, its center as a fraction of frame width, and split it into three zones: left, center, right. Then we asked how a channel’s own views move as it places the face on one side versus the other, holding video age constant. Comparing a channel to itself cancels out channel size, niche, and audience, so the result cannot be “big channels just frame faces differently.”
We read every contrast two ways and only claim a result when both agree: a pooled estimator (every video counts) and a one-channel-one-vote estimator (each channel weighs the same, so a few high-volume channels can’t carry it). The plan was written and publicly timestamped before any placement-vs-views number was queried, and the headline was read once off a sealed set of channels we never touched while building the analysis.
“Real” vs. “big enough to matter”
Section titled ““Real” vs. “big enough to matter””Before the finding, one idea that makes this study make sense: a result can be real and still be too small to care about.
With enough videos, almost anything you measure will be “statistically real” — that just means we’re confident it isn’t exactly zero. But “not zero” is a low bar. A change can be real and still be so tiny that no creator would ever notice it or change a thing because of it. Doctors hit this all the time: a drug can have a “real” effect that’s so small it doesn’t actually help the patient. Real on paper, useless in practice.
So we don’t ask “is there any effect?” We ask “is the effect big enough to act on?” To answer that, we drew a line before looking at any results: 10%. A change has to move views by more than 10% to count as worth a packaging rule. Ten percent is our honest call for the smallest change a team would actually do something about. Drawing the line in advance means we can’t slide it around later.
That’s the lens for the finding below: left-vs-right placement comes out at essentially zero, and even the edges of our confidence range stay in the low single digits — so this isn’t “we couldn’t tell,” it’s a confident “side placement doesn’t matter.”
Key findings
Section titled “Key findings”Left vs right: no more than a few percent, best estimate zero
Placing the face left versus right moves within-channel views by −0.3% (95% CI −5.4% to +5.2%) on the pooled estimator and +0.1% (−4.1% to +6.0%) on the one-channel-one-vote estimator. The interval rules out any effect bigger than about 6% and centers on zero. It does not rule out a few-percent effect.
The forks agree
Locate each video by its largest face instead of requiring a clean single face, and left-vs-right is still near zero (+1.7%). Raise the face-size floor from 2% to 4%, and the best estimate stays small (+3.4%, on a smaller, noisier sample). Side placement doesn’t move.
One loose thread: a center dip we don't claim
Comparing all three zones, the pooled estimator shows center-placed faces underperforming both edges (left vs center −15.7%). But the one-channel-one-vote estimator doesn’t see it (−3.7%, interval crosses zero). Under our two-estimator rule, that’s unresolved shape, not a finding.
Third near-zero in the face line
Presence is slightly negative, size is flat, and now side is near zero. The “faces sell, and here’s how to frame them” advice does not survive a within-channel test at scale.
Left or right? Doesn’t matter.
Section titled “Left or right? Doesn’t matter.”The headline and both robustness forks straddle zero, every point far inside the ±10% bar.
Whiskers are the 95% interval, resampled across whole channels. The dashed framing is the ±10% practical bar set in advance; every point estimate is well inside it. The strictest cut (4% floor) has a wider interval because it runs on the fewest channels (216), so its whisker reaches further, but its best estimate is still small.
Where the face sits, side by side
Section titled “Where the face sits, side by side”Here is each zone’s views against the typical placement on the same channel: left up about 4%, center down about 5%, right near the middle. A 4 to 5% swing in views is well below the 10% we set in advance as the line worth acting on. So even read at face value, this chart says placement barely matters.
Whiskers are the 95% interval over whole channels. Count each channel once instead of each video and the bars shrink further still (about +0.1% left, −0.9% center, +2.6% right), and the left-versus-right headline stays a flat zero either way.
Methodology
Section titled “Methodology”The data. Single-face videos from English-language business and creator channels: public, organic, long-form (90 seconds and up), each channel with enough qualifying videos. The headline is read off a held-out 20% of channels: 381 of them have at least five left and five right single-face videos (16,561 videos). Not a random slice of YouTube.
The comparison. Each channel against itself (channel fixed effects), holding video age constant, with views on a log scale, so the headline is one number: the percent view difference between placing the face on one side versus the other. Because the whole set was measured on one clock, a video’s age is effectively when we captured its views; controlling for age removes that timing artifact.
Two co-equal estimators. Pooled (video-count weighted) and median-of-per-channel (one channel, one vote). We claim a result only when both agree in sign and rough size. For the headline they agree (both confidently near zero); for the center contrasts they don’t (pooled real, one-vote not), so the center dip is reported as unresolved, not claimed. This rule was fixed in advance.
Clustered inference. Videos in a channel are not independent, so every confidence interval resamples whole channels (10,000 times), re-centering within each draw, never resampling individual videos.
The placement check. Face placement is the detector’s box center as a share of frame width. Unlike face size, a human can mark a face’s horizontal center reliably, so we ran the numeric check: on 150 hand-marked faces the detector’s center error does not drift with true position (slope −0.002, well inside the ±0.05 bound), the miss rate is even across zones, and the error does not widen at the edges. The placement validation passed before the headline was read.
The statistics. At this scale a p-value is near zero for any real effect, so we report the effect size and a channel-clustered interval against a ±10% practical bar set in advance. Everything here is a correlation; nothing was randomized.
Limitations
Section titled “Limitations”- It’s a correlation, not a cause. Where a face sits travels with framing, topic, and the rest of a packaging choice we can’t fully separate. “Tied to,” never “causes.”
- It’s a conditional question. We look only at single-face thumbnails, which is itself a packaging choice related to views. The dominant-face fork bounds the single-face conditioning; the faced conditioning is bounded small by the presence study. We don’t claim either is zero.
- These channels aren’t all of YouTube. Channels that vary which side they place a single face, from our English-language business and creator funnel.
- We measured views, not watch-time, captured once. Deleted and private videos are invisible.
- Short-form leakage is bounded, not eliminated. We drop everything under 90 seconds to trim padded vertical thumbnails, but duration alone doesn’t catch every short. Because the headline is near zero, a little such noise widens the interval rather than manufacturing an effect.
What this means for your team
Section titled “What this means for your team”For a brand or marketing team deciding where to place the face on a thumbnail:
- Don’t spend the meeting on left vs right. The data rules out any large side effect, and our best estimate is none, so for most teams this is one fewer packaging decision to argue about.
- The exception: if you chase single-digit lifts. This study can’t rule out a few-percent side effect. If a 5% view swing is real money at your scale, treat side placement as “not shown to matter,” not “proven not to,” and test it yourself.
- Center is the one open question, and it’s small. Faces at center may underperform the edges, but only in the volume-weighted view and not for the typical channel, so we don’t treat it as advice. It’s a reason to test, not a rule to follow.
- Test it on your own channel. The result is correlational, so the only way to settle it for your audience is to run your own A/B test.
Credits & disclosures
Section titled “Credits & disclosures”- Data: Hitfactor’s dataset of YouTube channels and videos. Face detection and placement by a standard open model.
- Conflict of interest: Hitfactor builds tools for video teams. We wrote and timestamped the analysis before seeing the result so the outcome couldn’t bend the method.
- Replicate it: the three pairwise zone contrasts behind the curve and the headline/fork table, plus a de-identified row dataset carrying each face’s zone and horizontal position, ship in the study repo.
- Cite as: Hitfactor (2026). The location of the face in a thumbnail (left/center/right) changes views by essentially zero: a within-channel study of face placement across 381 YouTube channels. hitfactorapp.com/labs/2026-thumbnail-face-location/