YouTube titles: what experiment data actually shows

Short answer: The best public evidence on headline wording comes from Upworthy's A/B tests (2013 to 2015), and it shows small, context-dependent effects and weak predictability. In one analysis, language alone predicted the winning headline only 54.42% of the time. So treat title formulas as hypotheses and use YouTube's own title A/B testing, rather than relying on rules of thumb.

The data source and its limits

The Upworthy Research Archive holds thousands of randomized headline tests from a news-style site, 2013 to 2015. It is the largest public randomized headline dataset we know of. It is not YouTube data, and the audience and format differ. Every number below carries that caveat.

What the studies found

Negative words and sadness. Robertson et al. found that each additional negative word in a headline was associated with about a 2.3% higher click rate, and each positive word with about a 1.0% lower rate. Sadness drove clicks; anger had no effect (Nature Human Behaviour, 2023, doi:10.1038/s41562-023-01538-4). Correction note: many creator guides say anger and fear lift clicks and sadness depresses them. That is the reverse of what this paper reports.

Pronouns and language features. Gligorić et al. analyzed linguistic features of the same archive. In their model, second-person "you" was not significant at the 0.01 level, first-person plural "we" reduced clicks (β = −0.15) and first-person singular "I" increased them (β = 0.24). Predicting the winner from language alone reached 54.42% accuracy, with a peak of 58.3% (PLOS ONE 18(3): e0281682, 2023, doi:10.1371/journal.pone.0281682). Reading: the effects are real but small, and the weak predictive accuracy means copying a formula is close to a coin flip.

Curiosity gap. Aubin Le Quéré & Matias studied curiosity-gap headlines using the same archive (Scientific Reports, 2025, doi:10.1038/s41598-024-81575-9). Read the paper for effect sizes; we do not restate numbers we have not verified in this article.

What this means for YouTube titles

  1. Concrete beats vague is a reasonable hypothesis but not a measured law here; test it.
  2. Do not copy formulas. With 54% predictive accuracy, your audience and topic matter more than the template.
  3. Do not exaggerate. YouTube advises against clickbait CTR tactics, and a title the video cannot deliver costs retention.
  4. Test, do not argue. Since December 2025, YouTube Studio supports A/B tests for titles and thumbnails (long-form, desktop, up to three variants, winner by watch-time share). Check Help for your account's availability.

A title test plan

  1. Choose a video with enough impressions to compare.
  2. Write three variants that differ on one dimension (specific number, named stake, first-person framing).
  3. Run the test and read watch-time share, not CTR alone.
  4. Record the winner and the dimension; reuse the lesson only after it replicates on another video.

Our experience: On Decoroo, the channel's 4.7% CTR over 232.7M impressions came from audience-specific, visual packaging for MENA interior design, tested continuously across a long run. We do not attribute it to any single title formula. See the Decoroo case study.

Evidence limits

  • Archive data is 2013 to 2015 news-style headlines.
  • Effects are modest; predictive accuracy is weak.
  • Some widely repeated numbers (for example "numbers in headlines lift CTR by 36%") have no primary source we can verify, so we exclude them.

Sources

FAQ

Should I use negative words? The data show a small average effect on news-style headlines; test on your audience, and weigh brand trust, especially in finance.

Is "you" in titles better? In this dataset it was not significant at α = 0.01.