Open YouTube Studio, go to any video's analytics, and click the little dropdown next to impressions click-through rate. Split it by device type. On most channels I look at, the phone number and the TV number are two to four points apart, and the desktop number sits somewhere in between pretending to be the average. That gap is the whole reason your thumbnail feels inconsistent. It is not inconsistent. It is being judged by three different juries.
I run a packaging studio, so I stare at this split more than is healthy. Once you start reading thumbnail performance by device instead of one blended CTR, a lot of confusing results stop being confusing. The thumbnail that "worked last time" did not stop working. Your traffic mix moved.
Why does CTR change so much between phone, desktop and TV?
Because the thumbnail is a different physical object on each one. Same file, three completely different viewing conditions.
On a phone, your thumbnail is roughly the size of a credit card, viewed in one hand, often at an angle, often outdoors with glare, and surrounded by a stack of competing rows. The viewer is scrolling with a thumb and deciding in well under a second. Small text is gone. Fine detail is gone. Low contrast is gone.
On desktop, the thumbnail is larger, the viewer is usually seated and slightly more deliberate, and the title gets more visual weight next to the image. Desktop is the only place where a second layer of detail actually gets read.
On TV, the thumbnail is enormous but the viewer is three metres away with a remote in hand. Compression artefacts show. Sharpened edges look crunchy. And critically, the TV surface is lean-back browsing, so the promise has to be instantly legible from across the room and the intent is closer to "entertain me" than "answer my question".
Three juries, three verdicts. One blended number hides all of it.
Where do you actually find thumbnail performance by device?
In YouTube Studio, open Analytics, then Advanced Mode, then the Device type tab, and add impressions plus impressions click-through rate as your columns. Set the date range to at least the last 90 days so you are not reading noise. YouTube's own help documentation for creators walks through the reporting surfaces if you want the official tour.
Three practical rules for reading it without fooling yourself:
- Never compare across traffic sources by accident. Browse features, suggested, search and external behave differently, and their device mixes differ too. If TV looks amazing, check whether TV traffic is mostly returning subscribers from the subscription feed. That is a loyalty number, not a packaging number.
- Ignore any device row with fewer than about 1,000 impressions. Small samples produce big fake swings.
- Compare a video against your own channel baseline per device, not against a global "good CTR" number. A 4 percent mobile CTR on a channel whose mobile baseline is 3.1 percent is a win. The same 4 percent on a channel that normally does 6 is a problem.
What does a mobile-first thumbnail actually change?
It changes four things: subject size, contrast, word count and edge safety. Everything else is decoration.
Subject size. The face or the object has to occupy enough of the frame that it survives being 320 pixels wide. If I have to zoom in to tell what I am looking at on my laptop, the phone viewer has already scrolled past.
Contrast. Not colour, contrast. Convert the thumbnail to greyscale. If the subject and the background collapse into similar mid greys, it will read as mush on a phone in daylight. This one test catches more bad thumbnails than any other.
Word count. Three words is the ceiling for mobile. Four if two of them are tiny. I have watched people fight this rule and lose to it every time. Words on a thumbnail are not there to describe the video. They are there to add the one piece of information the image cannot carry.
Edge safety. Timestamps sit bottom right. Progress bars sit on the bottom edge for anything the viewer partly watched. Overlay UI eats the corners on TV. Anything important within about 8 percent of any edge is at risk.
A thumbnail is not a poster. It is a signal that has to survive being shrunk, compressed, glared at, and scrolled past in 0.4 seconds. Design for the worst viewing condition and every other condition improves for free.
Should you optimise for the device that already sends the most traffic?
Mostly yes, but not blindly. Optimise for your dominant device, then check that you have not actively broken the others.
Here is the reasoning I use. If 70 percent of your impressions are mobile, a one point CTR gain on mobile is worth roughly three times a one point gain on desktop. So mobile wins the design argument. But if your TV share is growing quarter over quarter, which it is for a lot of long-form channels, then a design that is illegible on a big screen is a bet against your own future traffic mix.
The practical compromise is a hierarchy. One dominant subject that reads at every size, one supporting element that only desktop and TV viewers will notice, and text that is sized for the phone. That way the phone gets a clean signal and the bigger screens get a small reward for looking closer. Nobody gets a broken image.
This is also where the curiosity gap in thumbnails has to be handled carefully. A gap that depends on reading a small caption works on desktop and dies on mobile. The gap has to live in the image itself, in a face, a contrast, an unexpected object, something visual.
How do device splits change what you test?
They change what "a test" even means. A single A/B swap that moves blended CTR by 0.3 points tells you almost nothing. The same swap read per device often tells you exactly what happened.
The pattern I see most: a new thumbnail lifts desktop and drops mobile. That is almost always a detail problem. Something got more sophisticated and less legible. The desktop viewers could see the sophistication. The phone viewers just saw a busier rectangle.
The second most common pattern: mobile flat, TV up sharply. That usually means the subject got bigger and simpler, and the big screen rewarded the clarity.
When you set up a variant test, decide before you look which device you are trying to move and by how much. Otherwise you will find a win somewhere in the split every single time, because with three device rows and enough patience, something is always up. That discipline matters even more when you have to explain the result to someone else, which is a whole skill of its own. I wrote about how to report packaging results to a client without cherry picking.
Which niches feel this device gap most?
Finance, education and long-form commentary feel it hardest, because their audiences are split most evenly across devices.
A finance channel gets a big desktop share from people watching at a desk, a big mobile share from commute and lunch break viewing, and a growing TV share from people who put a 40 minute market breakdown on the living room screen. Charts and numbers on the thumbnail are the classic trap here. A line chart reads beautifully on desktop and turns into a scribble on a phone. If that is your world, the specifics of thumbnail design for finance creators are worth reading, because the tension between "look credible" and "be legible at 320 pixels" is the entire job.
Gaming and entertainment skew harder to mobile and TV, so they can push saturation and scale further. Tutorial and how-to channels skew to search and desktop, so they can carry one extra word.
There is no universal rule. There is your channel's actual device mix, which you can read in about four minutes.
Can you fix old videos with this, or only new ones?
Old videos are where this pays off fastest, and it is the single most underrated move in the whole channel.
Pull your top 20 videos by lifetime impressions. Sort them by mobile CTR against your mobile baseline. The ones that are underperforming on mobile but fine on desktop are almost always fixable with a thumbnail that is simpler and bigger. Those videos already have distribution. YouTube already knows who to show them to. You are not asking the algorithm to discover anything new, you are just giving it a better asset for the impressions it is already serving.
I would rather repackage eight existing videos than make a ninth new one for most channels under 200k subs. The maths is not close. A new video costs days of production and starts from zero. A repackaged video costs one thumbnail and inherits years of accumulated reach. If you want the full argument with numbers, I put it next to ad budgets in packaging costs versus ad spend.
How we handle it at ClickTheory
Every thumbnail we make gets checked at three sizes before it leaves the studio, and we design the title alongside the image rather than after it.
Our core package is $250 for 8 custom thumbnails, which works out to about $31 each, one per video. That includes unlimited revisions, a dedicated designer, and title consultation, because packaging is the thumbnail and the title and the first-frame promise together, not just the picture. Turnaround is 48 to 72 hours per thumbnail. If you want to run a proper device-split test, an A/B variant is +$15, a title pack is +$25, and rush is +$20.
Everything is custom per video. No templates, no reused layouts with your face dropped in. We test styles on our own channels first, which is how we know what survives a phone screen in daylight and what does not.
We are Dubai based and work worldwide and async, which for most clients means you send the brief at the end of your day and the work is waiting when you wake up. If you are weighing up whether this belongs with a specialist or with your editor, I laid out both sides in packaging agency versus video editor, and you can see the range of what we make in our work.
Start with one number
Open Studio, split your last 90 days by device, and write down three CTR numbers. That is the whole homework. Whatever the biggest gap is, that is the audience your current thumbnails are quietly losing.
If you want a second pair of eyes on that split and the packaging that caused it, email samy@pixelsamy.com and tell me your channel and your device mix.