CTV Ad Loudness: The Rule That Silences Spots

On 1 July 2026, California's SB 576 made it illegal for a streaming service operating in the state to run ad audio louder than the programme it interrupts. The law that had governed broadcast commercials since 2012 now reaches streaming, and most CTV creative pipelines were not built to prove compliance.

The consequence people expect is that a loud spot gets turned down. That is not the main risk. When a platform under a fresh legal mandate receives an asset it cannot verify, the safe response is not to normalise it. The safe response is to reject it.

Which means the failure mode is an ad that does not run at all, on inventory you already paid for, discovered after the flight has started. This is a specification problem wearing a compliance costume, and it is fixable in an afternoon if you know the numbers.

The numbers you need

Three figures govern this, and confusion between them causes most of the rejections.

−24 LKFS is the target loudness under ATSC A/85, the US broadcast recommendation that the CALM Act gave legal force in 2012. If your creative is destined for broadcast or for a linear-adjacent streaming environment in the US, this is the number.

−23 LUFS is the equivalent under EBU R128, the European standard, and it is what most streaming and server-side insertion environments normalise toward. LKFS and LUFS are the same measurement under two names, so the practical difference between the two standards is one decibel.

−2 dBTP is the true peak ceiling. This one is independent of the loudness target and it is where a surprising share of failures happen, because a spot can measure correctly on average and still clip on a transient.

Loudness is measured across the whole asset rather than at any instant. That is the entire point of the standard. A commercial with a quiet opening and a shouted close can hit its integrated target and still be the loudest thing a viewer hears that evening, which is why true peak is regulated separately.

Why one decibel matters more than it should

If your agency masters to −23 LUFS for streaming and a portion of your buy runs in a US broadcast-adjacent context expecting −24 LKFS, you are one decibel hot on that portion. One decibel does not sound like much and it is comfortably inside most platforms' tolerance.

It stops being comfortable when the tolerance narrows, which is what a new statutory deadline does to platform policy. Publishers who were previously normalising quietly now have a legal reason to enforce, and enforcement under uncertainty means tightening the band rather than widening it.

The practical instruction is to master to the tighter target and supply the metadata rather than to pick one standard and hope. Where your creative actually runs across a mixed buy is covered in CTV ad pod strategy, and the answer is usually more environments than the plan anticipated.

Why server-side insertion makes this harder

Here is the part specific to CTV, and the reason this is not simply broadcast compliance with a new name.

In traditional broadcast, one operator controls the whole chain. Programme audio and commercial audio pass through the same processing, and the operator can measure and correct with full knowledge of both.

With server-side ad insertion, the ad is stitched into the content stream at delivery time, frequently by a party that did not produce the content and has never heard the creative before. It has a few hundred milliseconds and whatever metadata came with the asset.

If the loudness metadata is present, the stitcher can normalise intelligently. It knows the programme level, it knows the ad level, it applies the difference.

If the metadata is absent, it has two options. Guess, which under a fresh legal mandate means accepting liability for a guess. Or decline the asset. Platforms will choose the second, and they will choose it silently, because a rejection at stitch time does not generate the same alert as a rejection at ingest.

This is the mechanism that turns a compliance question into a delivery question. Your spot is not quieter than you intended. It is absent, and your reporting shows the impression as unfilled rather than as rejected, so the diagnosis points at inventory availability rather than at your audio file.

Anyone reconciling a shortfall should rule this out before assuming a supply problem. The distinction between filled and actually rendered is the subject of fill rate versus rendered impressions, and loudness rejection lands in exactly that gap.

Where the liability actually sits

The regulatory framing and the commercial framing point at different parties, which is the source of most of the argument when something goes wrong.

Regulators look at the distributor. The CALM Act put the obligation on broadcasters and multichannel distributors, not on advertisers, because the distributor is who transmits. SB 576 follows the same logic for streaming services. The party facing enforcement is the one delivering the stream to the viewer.

Publishers therefore enforce upstream. A platform carrying statutory liability for what it transmits will not accept assets it cannot verify, and it has no incentive to be generous about it. This is why creative rejection rates for audio reasons rise sharply whenever a compliance deadline passes.

Advertisers absorb the cost without holding the liability. You are not the party the regulator will contact. You are the party whose flight underdelivers, whose creative sits in a rejection queue, and whose media plan has to be rebuilt with the clock running.

That asymmetry is worth internalising: the incentive to fix this sits with the party least exposed to the penalty, which is why it stays broken. Nobody gets fined for your rejected spot. You just do not get the impressions.

Why compliant-sounding spots fail

Almost nobody masters a commercial to be too loud on purpose. The failures come from production habits that are correct for one destination and wrong for this one.

The asset was mastered for social. Vertical video for feed environments is routinely mastered hot, because those platforms do not enforce a broadcast standard and loudness reads as production value on a phone speaker. A spot repurposed from a social cut carries that level with it, and the repurposing step rarely includes a remaster.

Heavy limiting flattened the dynamics. A track compressed to sit consistently above the noise floor of a commute has almost no dynamic range left. Its integrated loudness can still measure inside spec while its true peak sits at the ceiling, which fails the second test while passing the first.

The music bed was licensed already mastered. Production libraries deliver at their own levels, and a bed dropped under dialogue without gain staging raises the whole asset. This is the single most common cause we see described in trafficking disputes.

Transcoding changed the file. The measurement that matters is on the deliverable, after whatever encoding the platform or your own pipeline applies. Measuring the master and shipping the transcode means measuring a file nobody will ever play.

Nobody owned the check. The most frequent cause of all. Post-production assumes the agency validates, the agency assumes the platform normalises, the platform assumes the asset arrived verified. Loudness is a one-line addition to a trafficking checklist and it sits in the gap between three teams who each think it belongs to another. The full path an asset takes from delivery to playback is mapped in how CTV ad serving works end to end, and the audio check has no owner at any step of it.

Diagnosing it after the fact

Suppose a flight is underdelivering and you suspect audio. The signature is specific enough to confirm or rule out quickly.

Look at the shape of the shortfall. A genuine supply constraint suppresses every creative in the line item roughly equally, because the inventory simply is not there. A loudness rejection suppresses one asset across many publishers, or every asset on the subset of publishers running the tightest tolerance. If your 30-second cut is delivering and the 15 is not, that is not a supply problem.

Check delivery by publisher, not in aggregate. Tolerance varies, so rejection clusters. A creative delivering normally on six publishers and at zero on two is describing a specification difference rather than an availability difference. Segmenting delivery this way is standard practice in CTV reporting and analytics.

Ask for the rejection log. Platforms that validate at ingest usually keep one. Platforms that validate at stitch time frequently do not, which is itself informative: if nobody can tell you why an asset did not run, assume it was declined rather than unavailable.

Measure the file you actually shipped. Not the master, not the version in the shared drive. Pull the asset from the platform library if you can and measure that. The discrepancy between what was made and what was delivered is where this usually resolves.

Beyond California

Treating this as a single-state issue is a reasonable-sounding mistake with a predictable outcome.

Streaming services do not maintain separate delivery pipelines by viewer geography for something like audio processing. When a national platform faces a statutory requirement in its largest market, the operationally sensible response is to apply the strictest standard to all inventory rather than to build and maintain a geographic exception. That is what happened with privacy legislation and it is what is happening here.

The practical implication is that a spot running nowhere near California is now subject to a Californian threshold, because the platform carrying it decided once, globally, that it was easier. Buyers planning by geography, as covered in CTV geotargeting, should not assume a regional buy escapes a regional law.

Europe has run on EBU R128 for years and enforcement there is mature, which is worth knowing if a campaign crosses regions. The standards are close enough that mastering to the tighter of the two satisfies both, and far enough apart that assuming they are identical produces exactly the one-decibel problem described above.

The timing problem

Audio compliance failures surface at the worst point in the calendar, and for structural reasons rather than bad luck.

Creative is finalised late. Trafficking happens close to the flight date. Platform-side validation runs at ingest, and a rejection at that point leaves days rather than weeks to remaster, re-traffic, and re-approve across every platform in the buy.

We have written about the general version of this in how approval queues set your media plan. Audio is the sharpest instance because it is the one failure that can be detected mechanically, which means it is the one most likely to be enforced automatically, without a human deciding whether to be flexible.

The fix is unglamorous. Measure before you traffic. A loudness measurement on a finished asset takes seconds and can be automated. Almost nobody does it, and the ones who do stop having this problem entirely.

A pre-flight checklist

For advertisers and agencies:

1. Measure integrated loudness on every deliverable, not on the master. The file that goes to the platform is the file that gets measured, and transcoding changes levels.
2. Check true peak separately. A compliant integrated level with a −0.5 dBTP peak will still fail. These are two independent tests.
3. Master to the tighter target when a buy spans broadcast-adjacent and pure streaming inventory, rather than choosing one standard and discovering the other at ingest.
4. Ship the loudness metadata with the asset. This is the single highest-return item on the list, because it is what lets a stitcher normalise instead of reject.
5. Test every variant. A 15-second cutdown of a compliant 30 is not automatically compliant, because removing the quiet section raises the integrated level of what remains.
6. Keep the measurement record. When a platform rejects an asset, a timestamped measurement turns a dispute into a two-message conversation.
7. Ask what each platform enforces before the flight rather than after. Targets and tolerances differ, and the differences are not published anywhere central.

Two of these are worth expanding because they are the ones most often skipped.

Cutdowns are separate assets

Loudness is integrated over the duration of the file. Cut fifteen seconds out of a thirty-second spot and you have changed the denominator.

If the removed section was the quiet dialogue and what remains is the music-led close, the cutdown measures materially hotter than the original despite containing no new audio. Every duration in your rotation needs its own measurement, and the general specification requirements are set out in CTV creative best practices.

Metadata is the whole ballgame

A correctly mastered asset arriving without loudness metadata is, from the stitcher's point of view, indistinguishable from an unverified one. It cannot tell that you did the work.

Supplying the measurement alongside the asset converts a silent rejection into a normal ingest. It costs nothing, it is the item most frequently missing, and it is the difference between an ad that runs and an ad that does not. How assets and their metadata travel through the standard is covered in the VAST 4.2 guide.

What publishers should do

If you are the party transmitting, the liability is yours and the practical response has three parts.

Measure at ingest, not at stitch time. Validating an asset when it enters your library gives the advertiser time to fix it. Validating at delivery gives them nothing, produces a silent underdelivery, and generates a support conversation you will lose.

Reject loudly. A rejection that reaches a human with a stated reason and a measured value is a solved problem. A rejection that manifests as a quietly unfilled slot damages a commercial relationship and teaches the advertiser nothing. This is the same principle we apply to invalid traffic classification: the value of a filter is in the explanation it produces, not just the decision.

Publish your specification. Target loudness, true peak ceiling, tolerance, metadata requirement, and what happens when an asset fails. Most CTV publishers do not, and the resulting guesswork produces rejections that cost both sides money for no reason. Our own inventory expectations sit alongside the rest of the publisher tooling.

There is a version of this that goes further, and a small number of publishers already do it: validate the asset and return the measured value to the advertiser whether it passed or failed. A pass with a number attached tells a trafficking team the asset is safe on your inventory and probably safe elsewhere. It costs nothing beyond surfacing something you already computed, and it is the kind of thing that makes a platform easy to buy. The general argument for exposing your own measurements rather than only your conclusions runs through CTV viewability standards as well.

The wider point

Loudness is a small problem that behaves like a large one, because it sits at the junction of three things that do not normally meet: a broadcast engineering standard, a statutory obligation, and an automated delivery pipeline with no human in it.

CTV inherited the standard from broadcast and the automation from digital, and the handoff between them is where assets disappear. That inheritance is the recurring theme of the medium, set out in linear TV versus connected TV: two industries with incompatible assumptions, joined by a pipeline neither of them designed. As more states follow California's lead, the tolerance available to publishers narrows further, and narrower tolerance means more rejections rather than more normalisation.

The advertisers who stop having this problem are the ones who moved audio validation from post-production into trafficking, where it belongs. It is a checklist item, it takes seconds per asset, and it removes an entire class of underdelivery that currently gets misdiagnosed as a supply issue. The cost of not doing it is not paid in fines. It is paid in impressions you bought and did not receive.

If you want to see how creative validation and delivery reporting fit together, the ad server documentation covers the ingest path, and reporting covers how rejections surface rather than disappearing into fill rate. To talk through a specific buy, get in touch.

Frequently asked questions

What loudness should a CTV ad be mastered to?

Target −24 LKFS for US broadcast and broadcast-adjacent inventory under ATSC A/85, or −23 LUFS under EBU R128 for most streaming and server-side environments. LKFS and LUFS are the same unit under different names, so the two standards differ by one decibel. Keep true peak at or below −2 dBTP as a separate requirement, since a spot can hit its integrated target and still fail on peak.

Does the CALM Act apply to streaming services?

The CALM Act itself was written for broadcast and multichannel distributors. California's SB 576, effective 1 July 2026, extends equivalent requirements to streaming services operating in the state, making it unlawful to run ad audio louder than the surrounding programme. Other jurisdictions are expected to follow, and platforms operating nationally generally apply the strictest requirement everywhere rather than maintaining separate pipelines.

What happens if my CTV creative is too loud?

The expected outcome is normalisation, and the more common outcome is rejection. Server-side insertion has a few hundred milliseconds and whatever metadata arrived with the asset. If it cannot verify the level, the safe response under a legal mandate is to decline the asset rather than guess. That produces an unfilled slot rather than a quiet ad, and the shortfall usually gets misdiagnosed as an inventory problem.

Who is legally responsible for ad loudness, the advertiser or the publisher?

Regulators place the obligation on the party transmitting, meaning broadcasters, distributors, and now streaming services. Advertisers are not the party facing enforcement. The commercial consequence still lands on the advertiser, though, because a publisher carrying that liability enforces upstream by rejecting assets it cannot verify. The party with the incentive to fix it is the one without the penalty exposure.

Do I need to measure every version of a creative?

Yes. Integrated loudness is measured across the whole file, so a 15-second cutdown of a compliant 30-second spot has a different denominator and can measure materially hotter, particularly if the removed section was the quiet part. Every duration and every language variant in rotation needs its own measurement and its own metadata. Budget the extra validation into the flight schedule rather than treating it as a post-production detail, which is where it belongs in campaign planning.

Why does loudness metadata matter if the file is already compliant?

Because the stitcher cannot tell. A correctly mastered asset arriving without metadata looks identical to an unverified one, and the platform has to treat it as unverified. Supplying the measurement alongside the asset is what allows intelligent normalisation instead of a precautionary rejection. It is the most frequently missing item and the cheapest one to fix.

How can I tell if loudness rejection is causing my underdelivery?

Look for a gap between impressions recorded as filled and impressions recorded as rendered, concentrated on specific publishers or specific creative variants rather than spread evenly. A supply shortage affects all creative equally. A loudness rejection affects one asset across many publishers, or all assets on the publishers with the tightest tolerance. Asking the platform for its rejection log resolves it immediately when a log exists.