A thumbnail gets noticed when it passes seven checks: one clear focal point, strong light-dark contrast, a readable face or object at 168×94, three to five words at most, a color that separates it from the feed, nothing important under the duration badge, and a title that adds information instead of repeating the image. Test each one at feed size, not at full size, and compare versions blind before publishing.
What click-through rate measures, and what it does not
Click-through rate is the share of thumbnail impressions that turn into views. It depends on the audience, the topic, the title and where the video is shown, so there is no universal “good” number; the same thumbnail can earn very different rates on the home feed and in search. What a thumbnail controls is the first half-second: whether the eye stops on it and whether the viewer understands what the video is about.
That is why this checklist is about visibility and clarity rather than tricks. A design that is easy to see and easy to understand at small sizes gives the title and the topic a fair chance. A cluttered, low-contrast design loses before the viewer has read a word. Platform tests also reward the long game: thumbnail experiments are judged on watch time share, not only on clicks, so a misleading image that wins clicks and loses viewers does not help.
“Great thumbnails serve an important purpose beyond getting viewers to click.”
Test & compare thumbnails — video platform help center
The 7 principles at a glance
Use the table as a pre-upload checklist. Each row has a test you can do in ThumbProof or in your editor, and a pass condition. If a version fails two or more rows, fix it before you spend time on an A/B test.
| # | Principle | Test | Pass condition |
|---|---|---|---|
| 1 | One focal point | Squint (blur) test | One shape still stands out |
| 2 | Strong contrast | Grayscale + audit | RMS contrast ≥ 0.20 |
| 3 | Readable subject | View at 168×94 | Face or object recognizable |
| 4 | Few words | Count words | 3–5 words, ≥ 90 px tall |
| 5 | Color separation | Dark and light feed | Stands out on both |
| 6 | Clear corner | Safe-zone overlay | Nothing under the badge |
| 7 | Title pairing | Read title + image | Title adds new information |
1. One focal point
The eye needs one place to land. Thumbnails with three faces, two objects and a headline split attention and read as noise at small sizes. Pick the single most important element—usually a face with a clear expression or the object the video is about—and make it big, ideally covering 25–45% of the frame.
The squint test makes this objective: blur the image until details disappear. If one shape still pops out, you have a focal point. If everything melts into a gray mass, simplify. Eye-tracking research on web pages shows that people scan in quick fixations and skip what does not stand out, and a busy feed is the extreme case of that.
2. Strong light-dark contrast
Contrast is what survives scaling. Color differences fade at small sizes and on cheap screens, but light against dark stays visible. ThumbProof’s audit measures RMS contrast on a luminance version of your image: below 0.12 the design looks flat, 0.12–0.20 is moderate, and 0.20 or more is strong.
For text on the image, borrow the accessibility rule for web text: aim for at least 4.5:1 between text and its background. A thick dark outline or a soft shadow behind light text is the quickest way to get there on a busy photo.
Run the checklist in five minutes
- 01
Load your versions
Drop up to four exports into the feed preview, each with its own title.
- 02
Squint and grayscale
Turn on the squint filter, then grayscale. Note which versions keep a clear focal point.
- 03
Switch to the sidebar view
Check that the subject and words are readable at 168×94.
- 04
Toggle dark and light
Make sure each version separates from the surrounding thumbnails in both themes.
- 05
Read the audit breakdown
Open the per-dimension breakdown and fix any red rows: contrast, sharpness, corner, text.
- 06
Blind-test the survivors
Run five blind rounds with the versions that pass to see which one you spot first.
3. A subject you can read at 168×94
Faces work because people recognize them quickly, but only if the face is large. A face that fills less than about 4% of the frame becomes a few pixels at sidebar size. Crop tight, show the expression clearly and make sure the eyes and mouth have contrast against the skin and background.
If the video has no face, the same rule applies to the product, the result or the before/after. Show one object, large, from an angle that makes its shape obvious.
4. Three to five words, set large
Text on the thumbnail should add something the image cannot say, such as a number, a result or a twist. More than five words usually means the text repeats the title or becomes too small to read. On a 1280×720 canvas, the smallest important word should be at least 90–110 pixels tall in a bold weight.
Avoid placing text where the interface draws on top of it, and avoid thin script fonts. If you cannot read the words at 168×94, viewers on phones in a sidebar list cannot either.
5–7. Color, corner and title
Color separation: preview the thumbnail in both a dark and a light feed. A design built on a near-black background can vanish in dark mode; a pale design can vanish in light mode. One saturated accent color that is rare in your niche often helps more than a rainbow.
Clear corner: the duration badge covers the bottom-right corner. Keep text, faces and logos out of the right 22% and bottom 22% of the frame. Title pairing: the title and thumbnail should work as a pair, each carrying different information. If the image shows the result, let the title name the method; if the title asks a question, let the image hint at the answer.
- Preview on dark and light feeds
- Keep the badge corner clear
- Title and image carry different information
How to score a version against the checklist
Give each principle one point when the version passes and zero when it fails. Versions with 6 or 7 points are ready for a blind test. Versions with 4 or 5 points usually have one fixable problem, most often small text or a busy corner. Versions with 3 points or fewer need a new concept rather than tweaks.
Keep the scores for your published thumbnails next to their real results. After ten or twenty videos you will see which principles matter most for your audience, and you can weight them accordingly. That turns a generic checklist into your own design system.
| Score | Meaning | Next step |
|---|---|---|
| 6–7 | Ready | Blind-test against your best alternative |
| 4–5 | One clear weakness | Fix the failed row, re-run the audit |
| 0–3 | Concept problem | Rethink the subject and layout |
Worked example: two versions of a cooking thumbnail
Version A shows a bowl of ramen from above with the words “10-MIN RAMEN THAT BEATS TAKEOUT” in thin white letters across the bottom. Version B shows the same bowl at an angle, steam visible, with “10 MIN” in large yellow letters top-left and a dark outline.
In the checklist, A fails three rows: the six words are too small at 168×94, the text sits partly under the badge, and the RMS contrast is 0.14. B passes all seven. In a five-round blind test the creator spotted B first in four rounds, at an average of 0.8 seconds versus 1.4 seconds for A.
| Check | Version A | Version B |
|---|---|---|
| Focal point (squint) | Pass | Pass |
| RMS contrast | 0.14 (moderate) | 0.23 (strong) |
| Words on image | 6, thin | 2, bold |
| Corner clear | No | Yes |
| Blind picks (of 5) | 1 | 4 |
Common mistakes that hurt visibility
Copying the title onto the image
Repeating the title wastes the space. Add new information instead.
Too many elements
Three faces, arrows and a logo compete. Keep one subject and at most one supporting element.
Judging on a big monitor
Your editor shows 1280 pixels; viewers see 168 to 360. Always test small.
Low-contrast color schemes
Two saturated colors with the same brightness blur together. Check in grayscale.
Misleading images
Clickbait that the video does not deliver loses watch time, which is what platform tests reward.
Changing everything at once
When comparing versions, change one variable so you learn what made the difference.
