This field record evaluates visual memory decay for video ads served in live mobile auctions. The short name used in the repository directory is the 3-second visual hook retention study. The measurement question is whether a hook that can be read while the feed is silent is remembered a day later more reliably than a hook that depends on sound.

We evaluate n=4,200 video ad assets across active Meta Reels and TikTok auctions to calculate visual memory decay velocity. Empirical tracking reveals that motion-driven text overlay hooks achieve a 34% higher 24-hour message recall score compared to purely audio-driven hooks in silent-start feeds, directly contrasting broad sound-on optimization guidelines.

Sample

The asset set is dataset EMR-2026-07: 4,200 video ads observed in active Meta Reels and TikTok auctions. The recall contrast is estimated inside silent-start feeds, where playback begins before sound is available to the viewer.

Contrast

Two hook classes are compared. A motion-driven text overlay hook puts the message on screen as moving type. A purely audio-driven hook carries the message in the soundtrack and does not use that on-screen motion. The outcome is message recall scored at 24 hours. Decay velocity is the summary of how that recall falls off across the asset set.

Indexed to audio-driven hooks at 100, motion-driven text overlay hooks land at 134. That is the 34% higher recall score. It is a relative difference on the recall score, not a 34-point change on a 0–100 rate.

Platform guidance

The result sits against broad sound-on optimization guidance: recommendations that treat an audible hook as the default way to stop a scroll. In the silent-start condition measured here, the remembered message traveled with the visible overlay. The repository’s companion audit of Meta Advantage+ creative enhancements is the place those platform disclosures are indexed. This study does not restate those disclosures.

What this record does not claim

  • No creative-level file is released with EMR-2026-07.
  • No confidence interval or subgroup split is part of this release.
  • The comparison is a field contrast in live auctions, reported as observed. It is not presented as a laboratory randomization.