Does Immersive Audio Change Watch Time? A YouTube Experiment Protocol
Spatial9 Team · · 8 min read
Key takeaways
- This is a protocol only: no results are published yet, and Spatial9 will report its own outcome, including null results, as a follow-up.
- YouTube's Test & compare A/B tool tests titles and thumbnails on watch time share; it cannot split a video's audio between viewers.
- Use matched pairs with coin-toss assignment, a before-and-after baseline or a listening panel, and change only the audio.
- Pick one primary metric, such as average percentage viewed at day 28, and record average view duration, intro retention and device type alongside it.
- Match loudness, packaging and publishing pattern, and expect diluted effects because immersive playback depends on compatible devices.
Creators ask us a reasonable question: if I put the work into an immersive mix, will people watch longer? Nobody can answer that for your channel without testing it, and testing audio on YouTube is harder than testing a thumbnail. Below is a method you can run yourself, written so that the result means something even if it turns out to be "no difference".
Step 1: Write the hypothesis before you start
A clear hypothesis stops you from reading meaning into noise later. Write it down, with a date, before you upload anything. For example:
- Hypothesis: an Eclipsa Audio upload of a video has a different average percentage viewed from a stereo upload of a comparable video.
- Primary metric: average percentage viewed after a fixed period, such as 28 days from publishing.
- Secondary metrics: average view duration, the intro figure in the audience retention report, and views by device type.
- Decision rule: what difference you will treat as meaningful, and what you will conclude if the result is unclear.
Pick one primary metric. If you look at ten metrics, one of them will move by chance.
Step 2: Understand what YouTube can and cannot test for you
YouTube Studio has a built-in A/B testing tool, called Test & compare in Studio and documented in YouTube Help as "A/B test titles and thumbnails". It lets eligible creators test up to three titles, thumbnails or title and thumbnail combinations. YouTube runs the options and judges them on watch time share, and the test should complete within two weeks.
It does not test the video file or its audio. There is no built-in way to serve an immersive soundtrack to one group of viewers and a stereo soundtrack to another group on the same video. That is the core difficulty.
Step 3: Know why audio A/B tests on YouTube are hard
- No split on one video. You cannot randomise the audio track per viewer, so every design compares different uploads or different time periods.
- Most viewers may hear stereo. YouTube has accepted Eclipsa Audio files since January 2025, according to Google's open source blog, but immersive playback depends on compatible devices, such as Samsung's 2025 TVs and soundbars. Other viewers receive a rendition for their device. If only a small share of your audience can hear the immersive version, any effect is diluted.
- You cannot see who heard what. YouTube Analytics does not report which viewers received Eclipsa playback. Device type is the closest proxy.
- Retention has many causes. Topic, title, thumbnail, length, pacing and traffic source all affect it, usually far more than any single audio decision.
None of this makes the test pointless. It means you should design for a small, diluted effect and be honest about uncertainty.
Step 4: Choose a design
Option A: matched pairs across videos
Publish a series of comparable videos and alternate the audio: immersive, stereo, immersive, stereo. Better still, decide the order by coin toss so you do not unconsciously give the immersive slot to stronger videos. Compare the two groups. This needs many videos but is the closest you can get to randomisation.
Option B: before and after on one channel
Compare a run of immersive uploads with the channel's recent stereo uploads of similar length and topic. YouTube Studio's audience retention report can show typical retention, which compares a video with your 10 latest videos of similar length. This is simple, but seasonality, algorithm changes and audience growth all sneak in.
Option C: a controlled listening panel
Share unlisted versions with a recruited group, half hearing each version, and record whether they finish the video. This controls audio well, but it measures behaviour in an artificial setting, not real YouTube discovery.
Whatever you choose, do not upload the same video twice and compare the two. The second upload competes with the first, and their audiences and traffic will differ.
Step 5: Choose videos carefully
- Choose videos where sound plausibly matters: music, nature, travel, ASMR, gaming or cinematic storytelling.
- Keep length, format and topic as similar as you can within each pair.
- Exclude Shorts, live streams and premieres, which behave differently.
- Avoid launches that coincide with collaborations, sponsorships or news that might bring unusual traffic.
Step 6: Control everything except the audio
- Same packaging style. Keep titles and thumbnails consistent in style. Do not mention "immersive" or "3D audio" in the immersive video's title unless you also test that wording, because it changes who clicks.
- Same publishing pattern. Same weekday, time and notification settings.
- Same loudness. Match integrated loudness between versions. A louder version can seem "better" for reasons that have nothing to do with space. The free checker at /try-your-files runs a BS.1770 loudness and true-peak check in your browser; it is a guide for QC, not a certified meter.
- Verify the file. Before upload, confirm the immersive file really contains IAMF. Our guide is this file really immersive? shows how.
Step 7: Record the right metrics
YouTube Studio reports these at video level:
| Metric | What YouTube says it means | How to use it |
|---|---|---|
| Average view duration | Average minutes watched among those who stayed to watch, calculated from engaged views and their watch time | Compare only across videos of similar length |
| Average percentage viewed | The share of the video watched, on average | Better than duration for comparing different lengths |
| Audience retention report | How well each moment held attention, with key moments such as intro, spikes and dips | Look for dips at the same points in both versions |
| Intro | The percentage of viewers still watching after the first 30 seconds | Sensitive to the opening, where audio first registers |
| Device type | Where views came from: TV, phone, computer and so on | A proxy for who could hear Eclipsa playback |
YouTube Help notes that audience retention data typically takes one to two days to process, so do not read results on day one.
Step 8: Think about sample size in plain language
Retention numbers bounce around from video to video even when nothing changes. To see a real difference, it has to be bigger than that natural bounce.
- Measure your noise first. Look at average percentage viewed across your last 10 or 20 comparable stereo videos. How far apart are they? An effect smaller than that spread will be hard to see from a handful of uploads.
- More videos beats more days. In designs A and B, each video is one data point. Ten pairs tell you more than two pairs watched for a long time.
- Fix the end date in advance. Stopping when the numbers look good is the easiest way to fool yourself.
- Accept "inconclusive". It is a valid result. YouTube's own thumbnail tests report outcomes as winner, performed the same or inconclusive.
If you want formal statistics, a simple comparison of two groups' averages is a reasonable start, but the design matters far more than the formula.
Step 9: Record the results in a template
Keep one row per video, filled in on a fixed day.
| Field | Example entry |
|---|---|
| Video ID and title | |
| Audio version | Stereo or Eclipsa Audio |
| How the version was assigned | Coin toss, alternating or baseline |
| Publish date and time | |
| Length | |
| Integrated loudness | |
| Views at day 28 | |
| Average view duration at day 28 | |
| Average percentage viewed at day 28 | |
| Intro percentage | |
| Share of views from TVs | |
| Notes on unusual traffic |
Share the template and your hypothesis together when you publish, so others can judge the result.
Making the immersive version
You need a real immersive mix for this to be a fair test. Spatializer for Music separates instruments, places them in 3D and exports Eclipsa Audio for YouTube. Spatializer for Video is in early access. If you already have a Dolby Atmos ADM master, the ADM to Eclipsa Audio converter is free. For upload details, see our free Eclipsa Audio conversion guide and Eclipsa Audio for YouTube creators.
Scope of this protocol
This is a method, not a finding. It does not claim that immersive audio increases, decreases or leaves watch time unchanged. It reflects YouTube Studio features and help documentation as of November 2026, which may change. Spatial9 will publish the results of running it as a separate article.
Frequently asked questions
Can I use Test & compare to test audio?
No. According to YouTube Help, it tests titles, thumbnails and their combinations. The video and its audio stay the same for every option.
Will an Eclipsa Audio upload hurt viewers without compatible devices?
They are served a rendition suited to their device rather than the immersive version. Check the upload on the devices your audience uses; our article on stereo vs immersive sound explains what changes between them.
How long should the experiment run?
Long enough to cover several comparable videos per group, with each one read at the same age, such as 28 days. Decide the duration before you begin.
When will Spatial9 publish results?
When the experiment has run its planned course. We will publish the hypothesis, the design, the data template and the outcome, whatever it is.
Sources
- A/B test titles and thumbnails, YouTube Help.
- Measure key moments for audience retention, YouTube Help.
- Introducing the open source DAW plugin for Eclipsa Audio, Google Open Source Blog.
- Immersive Audio Model and Formats (IAMF) specification, Alliance for Open Media.
- ITU-R BS.1770: Algorithms to measure audio programme loudness and true-peak audio level, International Telecommunication Union.
Dolby and Dolby Atmos are trademarks of Dolby Laboratories. Spatial9 is not affiliated with or endorsed by Dolby Laboratories.