On this page
- Why gaze works: viewers look where the face looks
- The four gazes
- Head direction and eye direction are two controls
- When eye contact reads as a sales pitch
- Two faces: the eyes should meet, oppose, or both look at one thing
- Failure modes that make a gaze unreadable
- Getting the gaze you want
- The gaze check
- Common questions
- What to do next
Where the face on a thumbnail looks decides where the viewer looks next, and so it decides what the viewer reads second, after the face itself. Eyes on the viewer hold attention on the face and suit personal promises. Eyes on the object or the headline send the viewer there and suit promises about a thing. Eyes pointed out of the frame at nothing send the viewer out of the cover. A fourth option, eyes out of frame at something the headline names, can open a question, and it is the hardest to use well.
The expression guide already says which of the first three to pick, in a sentence each. This piece is about the mechanism underneath and the cases the one-line version misses: the fourth gaze, the point at which eye contact stops reading as honesty and starts reading as a sales pitch, and what two faces should do with their eyes. It sits inside the wider account of why people click one cover in a row rather than another; gaze is one of the tools that account leaves to the supporting pieces.
Why gaze works: viewers look where the face looks
People follow other people's eyes. Long before anyone reads, a viewer glancing at a face registers which way its eyes point and looks that way too. This is a fact about human attention rather than about YouTube, and we state it without a number because the numbers attached to it in thumbnail advice have no source we can find. What it means on a cover is that the face is not just the first thing seen; it is a pointer, and it points somewhere whether or not you chose where.
That gives gaze two jobs. The first is ordering: after the face, the viewer's eye goes where the face's eye goes, so the gaze decides whether the headline or the object is read second. The second is meaning: a face looking at a bill says "this is about that bill"; the same face looking at you says "this is about me and you". The expression tells the viewer how to feel; the gaze tells them about what.
In the rows we look at, the covers where the gaze and the layout disagree are the ones creators describe as "it just doesn't work and I can't say why". The face is on the right, the words are on the left, and the eyes look right, out of the frame. Every element is fine on its own and the eye path leaves the cover through the eyes.
The four gazes
At the viewer
Eye contact holds the viewer on the face. The viewer looks, is looked at, and stays for the expression. It suits promises where the person is the subject: I did this, I lost this, I was wrong, I am warning you. It also suits channels whose face is the reason people click, where recognition is the promise.
The cost is that nothing sends the viewer to the words or the object. The headline has to earn its own look through size and position, which is why eye-contact covers want the words large and close to the face. Eye contact with a small headline far from the face is a portrait with a caption.
At the object
The face looks at the thing, and the viewer follows. It suits reviews, reveals, comparisons, food, tech, anything where the object carries the promise and the face carries the verdict. The viewer reads face, object, headline, in that order, and the object arrives with the face's opinion of it already attached: a wince aimed at a laptop is a review in one glance.
It fails when the object is too small to be worth the trip, or when it sits behind the face's shoulder so the gaze has to travel backwards across the cover. Put the object where the eyes can reach it in one line, and give it the size the face-size floor gives the face, or close to it.
Take a laptop review cover, say. The laptop is on the left at half the height, lid open, screen dark. The reviewer is on the right at a third of the height, mouth pulled into a wince, eyes straight to the lens, "DON'T" in red top-left. The viewer reads wince, then eye contact, then a person warning me about something, and the laptop arrives last, as an afterthought. Turn the eyes to the laptop and nothing else on the file changes, but the reading does: the wince now has a visible cause, the cause is the product, and "DON'T" becomes a verdict about it rather than an address to the viewer. If the channel wants both, head to the laptop and eyes back to the lens, which is the third row of the table further down.
At the headline
The face looks at the words. This is the gaze the design pillar has in mind when it says the face should close the eye path, and it is the safest of the four for a face-right, words-left layout: face first, then the words the face is looking at, then the object if there is one. It turns the headline into the thing the face is reacting to, which is what a headline should be.
The failure is a face looking at words that do not deserve a reaction. A shocked face staring at "MY MORNING ROUTINE" has a visible cause and the cause is dull. Where the words sit relative to the face is its own decision; for gaze, the rule is that the words should be on the side the face is turned to, at a height the eyes can reach.
We would point the eyes at the headline on most face-right covers, even though eye contact is the pose a known face is recognised by fastest and a channel gives up some of that recognition every time the eyes leave the lens, because the headline is where the promise sits, and a viewer sent to the promise is worth more than a viewer held on the face.
Out of frame, at something the headline names
The face looks past the edge of the cover with a strong expression, and the headline names, or half-names, what is out there. "IT'S BACK" with a face staring off-frame in dread. "$0" with a face looking down and away. The viewer cannot see the cause and the expression insists there is one, which is a curiosity gap built out of gaze: the viewer knows exactly what is missing (the thing being looked at) and the video supplies it.
This is the one gaze the expression guide warns against, and the warning holds for the common case. Off-frame at nothing, with a mild expression and a headline that does not point, is a door the viewer walks out of. The device works only when three things are true: the expression is strong enough to imply a cause, the headline or the title names what is being looked at, and the direction of the look is towards the words rather than away from them, so the eye path passes the headline on the way out. Two out of three and it is the failure, not the device. The wider account of opening a question the video has to answer sets out what makes a gap hold; an off-frame gaze is a gap with the missing piece literally out of the picture.
Decide the gaze after you decide the second element. Write down what the viewer should look at after the face: the words, the object, or nothing but the face. Then point the eyes there. If the answer is "nothing but the face", use eye contact and make the words large. If you cannot answer, the cover has no second element and the gaze cannot fix that.
Head direction and eye direction are two controls
The head and the eyes can point in different directions, and the combinations mean different things at thumb size.
| Head | Eyes | Reads as | Use for |
|---|---|---|---|
| To the camera | To the camera | Direct address | Confessions, verdicts, warnings |
| Turned to the object | To the object | Absorbed in the thing | Reviews, reveals, builds |
| Turned to the object | To the camera | Caught in the act, complicit | Reactions, "look what I found", commentary |
| To the camera | To the side | Shifty, uncertain | Doubt covers, "should I?" |
| Turned away | Turned away | Disengaged | Almost nothing; the viewer leaves too |
The third row is the one worth learning. A head turned towards the object with the eyes flicked back to the lens pulls both ways: the head points the viewer at the thing, the eyes hold them on the face. It is the strongest single pose for a reaction or commentary cover, and it is also how most people look when they are about to tell you something. The fourth row is the doubt cover's natural pose and it suits finance and review rows, where the audience trusts a sceptic.
When eye contact reads as a sales pitch
Eye contact is honest by default and it stops being honest in one specific combination: direct gaze, a wide smile with teeth, and a product held towards the lens. That is the grammar of a commercial. Viewers have seen it on every screen they own and they read it before they read a word, and what they read is "this person wants something from me".
The way out is to change one of the three. Keep the eye contact and drop the smile to doubt, calm or concern, and the same pose becomes testimony. Keep the smile and move the eyes to the product, and it becomes delight at a thing rather than a pitch to a person. Keep both and remove the product from the frame, and it is a happy person, which is fine for a vlog and weak for a review.
Niche changes the threshold. Finance and business audiences read direct eye contact with confidence as a pitch faster than any other group, which is why finance covers work with doubt rather than a grin. Gaming and reaction audiences accept full eye contact with a big expression as the normal register. Education audiences accept eye contact with a calm face and reject it with a shocked one. The table in the expression guide covers the expression side; the gaze side is that the more the expression sells, the more the eyes should look at the thing rather than at the buyer.
We think the strongest eye-contact covers are the ones where the face looks slightly unhappy about what it is telling you. A doubtful or resigned face making eye contact reads as someone who would rather not be saying this, and that is the most believable register a cover has. A delighted face making eye contact reads as someone who is paid to be saying it.
Two faces: the eyes should meet, oppose, or both look at one thing
Podcast, interview, reaction and versus covers put two faces in a frame, and two gazes multiply the options. Three arrangements work and one does not.
- Both at the viewer. Two people addressing you. Reads as a joint statement and it is the podcast default; it works when the pairing is the promise and both faces are recognisable.
- At each other. Confrontation or conversation, depending on the expressions. The two gazes make a closed loop between the faces and the words go under or between them. Strongest for debates and versus formats.
- Both at one object. Two reactions to the same thing. The object becomes the subject and the two faces supply two verdicts, ideally different ones.
- In different directions at nothing. Two doors out of the cover. The eye follows one, then the other, and leaves.
Two faces at the same size with the same expression and the same gaze are a tie, and the podcast cover guide covers the size and layout answers to that; the gaze answer is that the two pairs of eyes should do different jobs or the same job on one object.
Failure modes that make a gaze unreadable
A gaze can only direct the viewer if the viewer can see it. Each of these removes the pointer.
- Eyes too small to read. Under the face-size floor, the eyes are two dots and their direction is a guess. Gaze is a reason to enlarge the face, not a substitute for it.
- Eyes hidden. Sunglasses, a cap brim, hair, a downward look that shows only the lids. A face with hidden eyes has an expression but no pointer, and a lowered gaze reads as reading or as sadness whichever you intended.
- Mid-blink or half-closed. The frame between expressions. It reads as tired at thumb size. Shoot several and pick the one with the eyes fully open.
- Looking at the corner the interface covers. A gaze aimed at the bottom-right sends the viewer to the duration badge.
- The flipped photo. Mirroring a shot to make the eyes point the right way is a legitimate fix, but check for text on clothing, an asymmetric haircut or a parting that the audience knows, and a watch on the wrong wrist. The flip is invisible until it is not.
- The gaze that contradicts the words. Eyes on the object, headline about the person, or the reverse. The viewer is sent to one thing and told about another.
If you find yourself adding an arrow to tell the viewer where to look, the gaze is pointing the wrong way; the piece on arrows, circles and badges covers when a device is a patch for exactly this.
Getting the gaze you want
Shooting for gaze means shooting each expression three ways: eyes to the lens, eyes to where the object will be, eyes to where the words will be. Mark the last two with tape on the wall beside the camera so the angle is the same every time. In the big face layout the eyes sit near the top of the frame and small angle changes read large, so shoot more variations there.
On the bench, Face Lab takes one selfie of a face looking at the camera, cuts it out, and renders eight expressions from it. When a saved face is placed into a render, the model paints it into the scene's lighting and angle, and the prompt can describe where the person is looking ("looking down at the bill", "eyes to the camera"). That gives you the object gaze and the eye-contact version of the same expression from one photo, which is the pair you want when a cover could go either way. Face Lab works only with faces of people who agreed to be on your thumbnails.
The gaze check
- Name what the viewer should look at second. Do the eyes point there?
- If the eyes leave the frame, is there a strong expression and a headline naming the cause?
- Is eye contact paired with an expression that reads as telling rather than selling?
- Head and eyes: do they agree, or disagree on purpose?
- Two faces: do the gazes meet, oppose, or share one object?
- At thumb width, can you tell which way the eyes point at all?
- Is the gaze aimed away from the bottom-right corner?
Common questions
Should you look at the camera in a thumbnail?
When the face is the subject and the promise is personal: a confession, a verdict, a warning, a story that happened to you. Eye contact holds the viewer on the face. When the promise is about a thing, look at the thing instead, so the viewer's eye is sent to it. Pick by what you want the viewer to look at second.
Why does my thumbnail look like an advert?
Usually because of the combination: direct eye contact, a wide open-mouth smile, and a product held up to the lens. That is the grammar of a commercial and viewers read it as one. Keep the eye contact and change the expression to doubt or calm, or keep the smile and turn the eyes to the product.
Is looking away from the camera bad in a thumbnail?
Looking at something in the frame is good; looking at nothing outside it is bad. A gaze that leaves the cover with no visible cause takes the viewer's eye with it, onto the next cover in the row. A gaze that leaves the frame with a strong expression and a headline naming what is out there can open a question, but it is the hardest gaze to get right.
What to do next
Look at your last ten covers and write down, for each, where the eyes point and what the viewer was supposed to read second. Where the two disagree, the fix for the next cover is a gaze change, not a redesign. If the eyes are too small to tell, the fix is size first, and if the expression is the problem rather than the direction, the expression guide linked above is the next read.
