Advice about YouTube titles tends to pull in two directions: be clear, and be curious. Clickbait sits at the far end of the second and gets blamed for everything. We have real data on this, because the pre-registered test of our score scored the titles and thumbnails of 642 real videos and compared the ones that beat their channel's average with the ones that fell short. This article reports what that data does and does not say about curiosity, clarity and clickbait risk in titles.
Key takeaways
- Of the title-side factors, promise clarity separated better-performing videos from weaker ones the most (75.7 against 71.4 on a 0-100 scale), then curiosity, then overall title strength.
- Every gap was small, between 0.7 and 4.3 points. Titles are one input among several, and none of these factors predicts a single video.
- Clickbait risk scored low almost everywhere (median 6 of 100) in a sample of established channels, so this data cannot tell you whether clickbait costs clicks or retention.
- Giving each video its pair partner's title moved the score by 11.3 points on average and flipped the ranking, so the title carries a large share of what the score reads.
What data do we actually have on YouTube titles?
We have scores for the titles, and the thumbnails that go with them, of 642 real videos from 60 established channels across six categories: tech, education, cooking, gaming, fitness and lifestyle. They form 321 pairs. In each pair both videos come from the same channel, were published a median of six days apart, and one clearly beat that channel's own average while the other clearly fell short. Comparing inside a channel cancels audience size, budget and design skill, which would otherwise swamp the effect of the title.
Two things limit what this can say. The channels are established, so the sample says little about small or new channels. And the score is a model's reading of the thumbnail and title together, not a count of words or a measure of viewer behaviour.
How are curiosity, clarity and clickbait risk scored?
Each is one of the twelve factors the score rates from 0 to 100. Promise clarity asks how clearly the pair states what the viewer will get. Curiosity asks how much it makes the viewer want to find out. Clickbait risk asks how exaggerated the promise looks, so here a high number is the bad one. Title strength is the model's overall judgement of the title as a title. The guide to the twelve factors has the definitions.
Which title-side factors separated better videos from weaker ones?
Clarity first, then curiosity. The table compares the mean score of the better-performing video in each pair with the weaker one, with the gap shown as Cohen's d, a standardised effect size in which about 0.2 is small and 0.5 is medium.
| Factor | Better-performing videos (mean) | Weaker videos (mean) | Difference (d) |
|---|---|---|---|
| Promise clarity | 75.7 | 71.4 | 0.31 |
| Curiosity | 53.5 | 51.9 | 0.24 |
| Title strength | 73.8 | 71.8 | 0.19 |
| Clickbait risk | 8.9 | 8.2 | 0.13 |
Promise clarity had the largest gap of all thirteen scores in the study, the overall score included. That fits the practical advice that a title should say what the video delivers. But the gap is about four points, which is small, and curiosity trailed it. The data supports clarity as a useful signal and says curiosity helps a little, not that either wins clicks on its own.
Does clickbait risk cost a video clicks?
This data cannot say. Clickbait risk was low for nearly every video: the median was 6 out of 100, and no video in the sample reached 50. That is what you would expect from established channels, which have little reason to risk their audience's trust. The better-performing videos averaged 8.9 and the weaker ones 8.2, a difference too small to read as a penalty or a reward.
The usual argument against clickbait is about retention: viewers who feel misled leave, and that hurts a video later. We measured no retention data, so we cannot confirm or refute it. For the reasoning, see our article on why retention is the cost of clickbait. The honest position is that low clickbait risk is a reasonable default for a channel that wants repeat viewers, not a finding of this study.
What about title length and word count?
The study checked both as trivial baselines, to see whether the score was just a proxy for them. Ranking the pair by the longer title picked the better video 51.7% of the time, and ranking by the title with more words did so 46.7% of the time. Both are indistinguishable from the 50% of a coin flip. In this sample, length told you nothing, and we make no claim about an ideal length.
How much of the score comes from the title?
More than the thumbnail, on the evidence of one control. We re-scored every thumbnail with the title of the other video in its pair, 642 combinations in all. The score changed by 11.3 points on average, with a median change of 8, and only 3.9% of thumbnails scored the same with a different title. After the swap the score ranked the better video first only 42.7% of the time, which is below chance and means the ranking inverted instead of fading.
If the thumbnail alone drove the score, the swap would have left accuracy at 59.5%. If the title alone drove it, accuracy would have fallen to about 35%. The result sat much nearer the second. The control changed only the title, so it says nothing directly about the image, but it does show that the title is not a rounding error. The full analysis is in our article on which half matters more.
How can you apply this to your own titles?
Work through the title in order of what the data supports, and keep your expectations modest.
- State the promise in plain words: what will the viewer know or be able to do after watching?
- Read the title without the thumbnail and check that someone could still say what the video is about.
- Add curiosity by naming the problem or the stake, not by withholding the answer.
- Check for exaggeration: if the video cannot deliver what the title implies, tone the title down.
- Score the title and thumbnail together with the free thumbnail and title analyzer and revise the lowest factor.
Our guide to title formulas covers the common structures and when each one backfires. For how the evidence was gathered and where it stops, see the method and evidence page.
What are the limits of this data?
The differences are small, the sample is 60 established channels, and the score is a model's reading, not viewer behaviour. The results are observational: better-performing videos had slightly different titles, but we did not change a title and watch what happened. Categories were too small to compare with one another, and the result holds for the channels sampled, not for every channel. Across all 321 pairs, the overall score ranked the better video higher 59.5% of the time, with a 95% confidence interval of 53.9% to 64.9%. That is a real effect, and a modest one.
Correction, 6 October 2026. An earlier version of this article presented data that had not been collected. It has been rewritten from the published results of our 321-pair test. The list of corrections is on the method and evidence page.
Sources
- ThumbSignal. Method and evidence: how the ThumbSignal score was tested, for the design, the result, the controls and the limits.
- ThumbSignal. Pre-registration, pair scores and verification script, public since 7 October 2026. github.com/MonteroDev/thumbsignal-score-study
- ThumbSignal. We tested our thumbnail score on 321 pairs of real videos.
Frequently asked questions
Does this prove that clear titles get more views?
No. It shows that, among 321 pairs, the better-performing video scored slightly higher on promise clarity on average. We did not change any titles, so the data is observational and cannot prove cause.
Is a more curious title always better?
The data points the other way on certainty. Curiosity scored a little higher on the better-performing videos, but by a smaller margin than clarity, and curiosity scores were low across the board. A curious title that is unclear about the content gives up the factor that separated videos most.
What is the ideal length for a YouTube title?
We do not have a figure. In this test, ranking pairs by title length and by word count both performed at chance, so length did not separate better videos from weaker ones. Write the length the promise needs.
Does clickbait hurt a channel?
This study cannot answer that, because clickbait risk was low for almost every video and we measured no retention. The reasoning against it is about trust and viewers leaving early, and it is covered in the article linked above.
Where can I check the numbers?
Every figure here comes from the published results of the 321-pair test, described on the method and evidence page. It lists the controls, the limits and what you can and cannot verify from outside.