Meta Says 9:16 Video With Audio Cuts Cost Per Result 34.5%: Test It

What exactly does Meta claim?

Meta's Reels ads page says 9:16 video ads with audio and safe-zone messaging see 34.5% lower cost per result than image ads, and 2x higher delivery. It also says 79% of surveyed users report purchasing after viewing Reels. I read the page on 2026-10-03; the page gives these as Meta's figures and I have no way to check them from outside.

Treat them as a reason to run a test, not as a forecast for your account. The comparison is a specific one: vertical video with sound and safe-zone-aware text against an image.

How do I build a fair test?

Keep everything but the format the same: audience, budget, placement, objective and primary text. Make two ads, one image and one video, from the same product shot. Run them at the same time for the same period. If you change the offer along with the format you will not know which one moved the result.

How do I produce the video arm with Sume?

Cut a 9:16 clip with video trim at $0.02 per job; use precision: "exact" if you need frame-accurate in and out points. Add captions with POST /v1/video-captions at $0.20 per job, and pick a style such as tiktok-green or black-outline with the design overrides where the style allows them, to keep text inside the margins. Add a soundtrack from the Music endpoint ($0.125 per generation) and mix it with the timeline's soundtrack and duck_db.

The image arm: POST /v1/images at 9:16, or a frame pulled from the clip with the video frames endpoint, so both arms show the same product moment.

• Same prompt, same product, same copy.
• One variable: image versus sound-on video.
• Same placements, same dates.

How long should I run it?

Meta's page does not give a test duration, and neither do I. Decide the budget and the decision rule in advance: for example, the arm with the lower cost per result after a spend you set wins. Write it down before launch so you do not change the rule after you see the numbers.