How to verify a podcast ad actually ran.

The question behind the question

When an advertiser asks whether their ad ran, they are rarely asking a technical question. They are asking whether they can trust the invoice. So the answer "yes, we checked" is worth almost nothing, because it asks them to trust the same team whose bill is in question. What settles it is evidence they can look at themselves.

That reframing matters because it changes what you build. A workflow designed to reassure your own team can be informal. A workflow designed to survive an advertiser dispute has to leave a trail.

Four facts have to link together

Proof of play is a chain. Break any link and it stops being evidence.

• Identity. Which advertiser, campaign and creative was supposed to run. Without this, you are proving that an ad ran, not that the ad ran.
• Placement. Which show and episode it was booked into, and which slot: pre-roll, mid-roll, post-roll, or a specific named segment.
• Presence. Evidence that the read exists in the published audio or video, rather than in the plan for the published audio.
• Position. The timestamp and duration where it appears, which is what distinguishes a delivered mid-roll from one that quietly ended up as a post-roll after nobody was listening.

Most reporting stops at the first two, because the first two live in systems you already own. The last two require someone or something to look at what actually published.

Why impression counts are not proof

Hosting and ad-serving platforms produce impression numbers, and those numbers are genuinely useful for pacing and billing volume. They are not proof of play, and treating them as proof is the single most common gap we find.

An impression records that a file was requested. It does not record what was inside the file. With dynamic insertion, the correct creative can fail to serve while the request is still counted. With a host-read endorsement, the read is baked into the episode and never touches an ad server at all, so the metric that would prove it does not exist. In both cases the number looks healthy and tells you nothing about whether the advertiser got what they bought.

Why spot checks stop working

Spot checking is a reasonable strategy at small volume. Someone listens to a few episodes, everything sounds right, and the risk is genuinely low because the number of things that can go wrong is small.

The method fails quietly as inventory grows, and it fails for a mathematical reason rather than a diligence reason. Every additional advertiser, creative, episode and flight date multiplies the combinations that must stay aligned. Sampling five percent of a hundred placements is meaningfully different from sampling five percent of two thousand, because in the second case the ninety-five percent you did not check now contains far more absolute errors. The team's effort has not dropped. The coverage has.

Worse, spot checks are usually aimed at the campaigns most likely to be questioned, which are the campaigns from attentive advertisers. The unattended ones drift.

Checking from both directions

There are two ways to look for a problem, and each one is blind to a different failure.

Booking outwards

Start with the contracted schedule and confirm each committed read appeared. This catches the missing ad, the late ad and the misplaced ad. It cannot catch anything that was never booked, because it never looks there.

Publication inwards

Start with everything that actually published, including bonus episodes and video that was never in the reporting scope, and identify the sponsor placements inside it. This catches the read that ran without a booking, the bonus episode nobody logged and the campaign that kept running after its flight ended.

Run only the first and you will believe your delivery is clean while giving away unbilled inventory. Run only the second and you will have a catalog of what aired with no idea what was owed. This is exactly why we run verification and show QC as one service rather than two: the value is in the disagreement between the two views.

What to record, and why each field earns its place

A verification record is only useful if it can answer a question months later, when nobody remembers the campaign. In practice that means capturing:

• Show, episode and video identity, with the platform identifiers rather than the human title, because titles get edited
• Advertiser, campaign and creative reference
• Expected air date and observed air date as separate fields, since the gap between them is the single most useful signal you have
• Slot, timestamp and duration of the detected read
• Delivery status: contracted, late, bonus, unexpected or missing
• Review state: passed, flagged, blocked or awaiting human judgment
• Who resolved an exception, when, and what was agreed

That last field is the one teams skip and later wish they had. A record of the decision is what stops the same exception from being re-litigated every quarter.

Timing changes the economics

Verification tied to the reporting cycle is verification that arrives too late to be useful. If you check at month end, a placement error has already run for up to four weeks, the inventory to fix it may be gone, and your only remaining option is a make-good, which is the expensive outcome.

Verification tied to the publishing cycle changes what the finding is worth. Catch a missing read within a day or two and the fix is often a scheduling conversation. The cost difference between those two timelines, repeated across a year, usually dwarfs the cost of the verification workflow itself.

Automate the volume, escalate the judgment

Scanning, matching and comparison are exactly what software should do. They are repeatable, high volume and unaffected by fatigue on the four hundredth episode.

What software should not do alone is decide. An ambiguous read, a schedule that changed mid-flight, two systems that disagree about which episode a placement belongs to: these need a person who can call production, understand the commercial agreement and make a defensible decision. Automating that step does not remove the ambiguity, it just removes the human who would have noticed it, and then delivers the wrong answer to the client with total confidence.

The workable split is simple. The system creates coverage and surfaces mismatches. A person resolves mismatches and owns the outcome.

When something is genuinely missing

Log the exception before you do anything else, because the moment you start making calls, memory becomes the record. Then work through it in order: did the episode simply publish late, did the read move to a different episode or slot, was the campaign amended in a system your reporting does not see? Only when those are ruled out does it become a delivery failure, and only then is it worth discussing whether the remedy is a correction, a bonus placement or a credit.

Handled that way, the advertiser conversation is about a documented event with a proposed remedy. Handled without a record, it is two parties arguing from memory, and the party holding the invoice generally loses that argument.

The short version

Verification is not listening to episodes. It is producing a record that links what was sold to what was published, close enough to publication that problems are still fixable, with a human accountable for everything the automation could not resolve. Agencies that build that record stop having stressful advertiser conversations. Things still go wrong; they just always find out first.