This Advertising Age article by Garrett Sloane reports on a series of secret ad test run by Google Unskippable Labs on YouTube “…to help brands get a sense of what works in mobile video and what grabs people's attention for the most amount of time…”

Researchers used 16 different videos with variations in pacing, sound and other factors. Although the average completion rate for videos viewed on mobile devices, the best-performing video in the tests achieved 33% average view-through.

Results by variable:

• Pacing: Faster version had 35% view-through on mobile vs. 29% for slower version. On desktop, faster version achieved 33%, slower version 31%.
• Vertical/horizontal video (Vertical orientation is common within apps like Snapchat): Horizontal video had 29% view-through on mobile vs. 27% for vertical video. This conflicts with Snapchat’s own results, which find vertical ads to have higher view-through.
• Captions on/off: Ad versions with subtitles were tested because ads sometimes run without sound on some social-media platforms. The subtitled version performed slightly better: 23% view-through vs. 22% for captionless version.
• Super captions: This test compared subtitled version against version using supplemental text (not just text of dialog). Supplemental-text version performed better (26% view-through vs. 25%) for both mobile and desktop.

Please note that we haven’t been given any information about the numbers of ad impressions in this “study”. Also the differences between versions used in tests of each of the variables are very small. If results are statistically significant, I think Google would have said so. Without any more quantification than what’s contained in this report, I wouldn’t be inclined to invest a lot of money in redesigning video ads per these guidelines.