Do YouTube Thumbnails with Faces Get More Views?
“Put a face on the thumbnail” is conventional YouTube wisdom. It sounds reasonable: people look at other people, so a face should pull the eye, win the click, and earn the view.
We tested this on 1.25 million videos. Inside a single channel, the videos with a face on the thumbnail earned about 6% fewer views than that same channel’s videos without one.
The effect is small and it’s a correlation, not proof. But it runs the opposite direction of the advice, and it held on a separate set of channels we never looked at while building the analysis.
A face is tied to slightly fewer views, not more — the opposite of the advice. Real, but small.
We're 95% sure the real number is between −9.3% and −3.3%. It clears zero, so it's a real signal, but it's below the 10% line we set as worth changing your process over.
If a face were a reliable win, you would see it the moment you compared a channel against itself. You don’t.
What we did
Section titled “What we did”We ran every thumbnail through a face-detection model, then checked it by hand against 1,100 labeled thumbnails to confirm the detector was accurate before we used it. Then, within each channel, we compared the views of its thumbnails with a face to its thumbnails without one, at the same video age.
Comparing a channel against itself is the whole point. It holds the channel’s size, niche, and audience constant, so the result can’t be dismissed as “bigger channels just use faces more.” A face video and a no-face video from the same channel are competing on equal footing.
“Real” vs. “big enough to matter”
Section titled ““Real” vs. “big enough to matter””Before the findings, one idea that makes this whole study make sense: a result can be real and still be too small to care about.
When you have over a million 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?” — at this size, there almost always is. 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 rearranging your work over. 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 to make a result look better or worse than it is.
That’s the lens for the finding below: the face effect is real (it clears zero, and the wrong way), but at about −6% it lands under the 10% line — real, but not big enough to reorganize your thumbnails over.
Key findings
Section titled “Key findings”A face is tied to ~6% fewer views
Compare a channel to itself and its face-thumbnail videos earn a median 6% fewer views than its no-face videos of the same age. Real, but small.
The answer flips depending on how you compare
Compare videos across different channels and faces look slightly positive (+3%). Compare each channel against itself and that becomes −6%. The cross-channel number is confounded: bigger channels both use faces more and get more views, so the face gets credit for the channel’s size. The within-channel number is the honest one.
Adding more faces does nothing
Once a thumbnail has one face, a second or third doesn’t add views. The effect is flat, if anything slightly negative.
Independent of face size
Section titled “Independent of face size”We had to pick a rule for when a face is big enough to count as present. A fair worry is that the answer depends on that choice. It doesn’t. We re-ran the comparison requiring the face to fill 0.5%, 1%, 2%, then 4% of the thumbnail, and at every cutoff, face videos still trailed no-face videos. The gap is largest when we count even small faces and smallest when we demand prominent ones, but it never closes to zero.
Whiskers are the 95% interval, resampled across whole channels.
More faces don’t help
Section titled “More faces don’t help”If one face helps, more should help more. They don’t. Among thumbnails that already have a face, adding a second or third leaves views about where a single face puts them, no measurable gain. The 4+ bar ticks up (+3.6%), but it rests on just 309 videos, and the detector misses about half of crowded thumbnails, so treat it as noise, not a signal. The takeaway isn’t that more faces hurt; it’s that piling them on does nothing.
A curious split by length
Section titled “A curious split by length”One pattern caught our eye, though we wouldn’t read much into it yet. On videos longer than three minutes the association is about −5%. On videos three minutes or shorter (still regular uploads, not Shorts) it’s closer to −16%. The split held on both halves of the study, so it isn’t a fluke of one sample, but we can’t say why, and it could easily be something about short videos other than the face. We flag it as a question for a future study, not a reason to treat faces differently on short content today.
Methodology
Section titled “Methodology”The data. 1,253,887 public, organic, long-form videos from 16,737 channels, each with at least 20 videos. Two locked filters (a 10-view floor and a 90-day age minimum, so views had settled) leave 1,189,217 videos across 16,705 channels. That frame drives every number here. The channels are English-language business and creator channels from our own dataset, not a random slice of YouTube.
The comparison. We measured each channel against itself, holding video age constant, on the log of views. Videos from the same channel aren’t independent, so every confidence interval resamples whole channels rather than individual videos. Treating 1.25 million videos as independent would make the intervals several times tighter than they should be.
The face check. A face counts only if the detector is confident and the face fills at least 2% of the thumbnail, a size we fixed in advance. Against the 1,100 hand-labeled thumbnails the detector scored 0.97 precision and 0.99 recall. The most important check: its misses don’t correlate with view count. A detector that missed more often on low-view videos could invent this result out of nothing, and ours doesn’t. Note that “face” means whatever the detector recognizes, cartoons included, not strictly a photographed human.
The statistics. With more than a million videos, a p-value is near zero for any real effect, so it tells you nothing useful. We report effect sizes and channel-clustered confidence intervals instead, measured against a +10% threshold we set in advance as the bar for a meaningful effect. Every result here is a correlation. Nothing was randomized.
Limitations
Section titled “Limitations”- It’s a correlation, not a cause. A face comes packaged with a topic, a title, and a dozen other choices we can’t fully separate from it. We say “tied to,” never “causes.”
- These channels aren’t all of YouTube. English-language, business and creator, mid to large, drawn from our own funnel. Your niche may behave differently.
- We measured views, not watch-time, captured once. Deleted and private videos are invisible to us. The visible drop-off was even across face and no-face videos, so it doesn’t explain the result.
- We caught and corrected one measurement error. Because the whole dataset was measured on a single date, a naive views-per-day comparison let a video’s age leak into the result. The headline controls for age directly, which removes it.
Practical takeaways
Section titled “Practical takeaways”For brand and marketing teams deciding where to put thumbnail effort and budget:
- Stop treating a face as a guaranteed win. The data doesn’t support “always put a face on it.” At best a face is neutral within a channel; here it’s tied to a small decline.
- Don’t pay a premium for a face. A shoot, a model, or a designer’s hours spent forcing a face into frame is buying a result the data won’t back. Spend it on the topic and the title instead, the choices a face is usually bundled with.
- Adding more faces won’t help. A second or third face buys nothing. A crowded thumbnail is effort with no measured return.
- Test it on your own channel. The effect is small and correlational, so the only way to know for sure for your audience is to run your own A/B test. Don’t inherit a rule, measure it.
Credits & disclosures
Section titled “Credits & disclosures”- Data: Hitfactor’s dataset of YouTube channels and videos. Face detection 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.
- Cite as: Hitfactor (2026). Thumbnail faces are tied to 6% fewer views: a within-channel study of 1.25M YouTube videos across 16,737 channels. hitfactorapp.com/labs/2026-thumbnails-with-faces/