ThumbnailScore
Everything the tool does, what each number means, and how to turn a score into a better thumbnail and title.
A full analysis takes about a minute. You do not need an account, and you do not need to pick a category — the AI works that out from what you upload.
Drag and drop the image, click to browse, or paste it straight from your clipboard with Ctrl+V. Upload the real thumbnail you plan to publish, not a frame from the video — the analysis reacts to text overlays, crops, and framing that only exist in the finished thumbnail.
Paste the exact title you intend to publish, including capitalisation and punctuation. The scoring reads tone and phrasing, so a draft written in shorthand will score differently from the version viewers will actually see.
Press Analyze. The AI reads the thumbnail and title together as a single impression — the same way a viewer meets them in their feed — and returns a score out of 100 with a full breakdown.
Inputs are validated before anything reaches the AI. Files are checked by their actual bytes, not their extension, and re-encoded server-side.
The global score is not an average of the twelve metrics. It models the two things YouTube actually measures in sequence: whether an impression converts to a click, and whether that click turns into watch time. A package has to succeed at both.
Everything visible before the click: how readable the thumbnail is at feed size, how strong the title is, and the psychological triggers that make scrolling past feel uncomfortable. Thumbnail readability and title strength carry the most weight here, because they are what the eye lands on first.
What happens after the click. Promise clarity dominates this layer: if a viewer cannot tell what they are going to get, they leave early and the video stops being promoted. Coherence and emotional resonance make up the rest.
The two layer scores are combined with a geometric mean rather than a simple average. This is deliberate: a great title attached to a misleading thumbnail cannot average its way to a good score. One weak layer drags the whole result down, which mirrors how the algorithm treats a package that wins clicks but loses viewers.
Clickbait risk is not a metric that gets averaged in — it is a penalty applied to the combined result. Mild hyperbole barely registers. Severe overpromising is heavily punished, because YouTube detects it through watch-time drop-off and suppresses the video regardless of how many clicks the thumbnail earned.
Every result carries both a letter grade and a finer-grained band label. The bands are more granular than the grades, so use them when you are comparing two versions that land in the same letter.
| Score | Band | Grade | What to do about it |
|---|---|---|---|
| 90–100 | Viral Potential | A | Publish it. Further tweaking is more likely to hurt than help. |
| 75–89 | Strong Appeal | A–B | Strong package. Worth one focused pass on the weakest single metric, then ship. |
| 60–74 | Good Baseline | B–C | Solid but unremarkable. This is where the alternative titles usually pay off most. |
| 45–59 | Room to Improve | C–D | Something specific is holding it back. Work the two lowest metrics before republishing. |
| 25–44 | Needs Work | D–F | Structural problems, not polish problems. Rethink the concept rather than editing details. |
| 0–24 | Low Performer | F | Not competitive in a feed. Start over with a different angle on the same video. |
Each metric is scored 0–100 on its own. The groups below show which layer a metric feeds and how much influence it carries within that layer — useful for deciding what is actually worth fixing.
Text visibility, contrast, clutter level, subject clarity on mobile.
Length, promise, precision, absence of filler words, audience fit.
Unresolved narrative tension that makes not clicking feel uncomfortable.
Narrative tension, open loops, before/after contrast, implicit question.
Incomplete information that creates a mental loop the brain must close.
Fear of missing out — social proof that others are already benefiting.
Time-bound or now-or-never framing that pressures immediate action.
Exclusivity and scarcity signals that make content feel privileged or secret.
Viewer immediately understands what they will get from the video.
Image and title complement each other without redundancy.
Facial expression, colour energy, tone aligned with the content.
Higher means more risk. Exaggerated claims reduce your overall score.
Lower is better — this one subtracts from your score.
Below the score you get four sections. The first three open automatically; the fourth is collapsed until you need it.
A one-sentence summary of what the AI believes your video is about, based only on the thumbnail and title. Read this first. If it does not match your actual video, your package is miscommunicating before any metric matters — and that mismatch is usually the real reason the score is low.
Specific, ordered changes tied to your weakest metrics. These are concrete instructions rather than general advice, so they can be applied directly and re-tested.
Rewritten titles, each scored and sorted best-first. Only alternatives that outscore your original are shown, so an empty or short list is itself a signal that your title is already strong. Click any one to copy it.
Short overlay phrases built to stay readable at feed size. These are meant to be added onto the image, not used as titles — keep them to a few words so they survive being shrunk on mobile.
The share buttons under the score post your grade and title to social platforms, or copy the summary to your clipboard. Nothing is published without you pressing one of them.
Sign in with Google or GitHub to keep a record of what you have tested. History is the only feature that requires an account.
The most useful habit: analyze your first idea, apply one recommendation, analyze again, and compare the two entries. The difference between the two scores tells you more than either score alone.
Full details are in the Privacy Policy.
The interface is available in 17 languages, and the analysis responds in the language you are browsing in. The AI also detects the language of your title itself, so a title written in one language is judged by that language's conventions rather than translated first.
Change one thing at a time. If you rewrite the title and swap the image together, the score moves but you learn nothing about which change caused it.
Check the AI Video Context before the numbers. A wrong summary means the package is unclear, and no amount of metric-tuning fixes that.
Treat the score as a comparison tool, not a verdict. What matters is whether version B beats version A, not whether either hits an arbitrary target.
Test the thumbnail at the size viewers see it. Shrink it on your screen — if the text stops being readable, the readability metric is telling you something real.
Do not chase a perfect clickbait risk of zero. Some tension is what earns clicks; the penalty only bites when the promise stops being honest.
Take the alternative titles as raw material, not finished copy. Copy the one closest to your voice and edit it — a title that does not sound like you will underperform with your own audience.
No account, no setup. Upload an image and a title, and you will have a full breakdown in seconds.