Is Your Podcast Monitoring Capturing What People Say About Your Brand?

A host spends eleven minutes on your product. A guest disagrees. Neither types a word about it. No caption names you, no title names you, no description names you. The episode goes out, and your weekly report shows a flat line in the podcast row.

The question that matters is not whether the conversation happened. It is whether your current stack has a path for it to show up.

The absence of a signal and the absence of an index are not the same thing. One is a quiet week. The other is a reporting gap you cannot see from the dashboard. The point of this post is to help you tell them apart on your own stack, with a short operational test any comms lead can run before the next renewal call.

Mentions of a podcast and mentions inside one are different datasets

An export labelled "podcast coverage" may contain posts about episodes, mentions inside episodes, or both. Establish which dataset your report includes.

A mention of a podcast is someone writing about an episode on social or in an article. It tells you how audiences are reacting. A mention inside an episode is a brand named in the spoken content, at a timestamp, by a specific speaker, in a specific tone. Both can matter in the same week, and they can diverge. Treating one as a proxy for the other is where the reporting gap starts.

Episode-level coverage exists. Dedicated podcast monitoring tools advertise transcript search, brand alerts, episode excerpts and timestamps. What varies between stacks is which shows, which episodes, which languages, which audio and video formats are actually processed, how often, and how the output is scored. A buyer has to establish those answers for their own stack rather than assume one way or the other.

For the broader pattern of how a text index behaves on video content, why social listening tools misread video is the general case. The podcast case has its own shape because the episode carries so little text to begin with.

Why a text index behaves differently on long-form content

Short-form video surrounds the clip with captions, on-screen text, hashtags and a comment thread, so a text-based tool catches a share of the signal even when it misses the frame. A ninety-minute episode has one title, one description and a thumbnail. The brand discussion at minute 34 may appear nowhere in the title or description. Unless your tool processes the episode or an available transcript, it can miss that discussion entirely.

That is why the audit matters. If your current tool is a text index, it is not failing on podcasts because something is broken. It is working exactly as designed. The question is what it was designed to catch.

How large is the surface

Two figures with their primary sources, scoped to what they actually say.

YouTube reported that users streamed over 700 million hours of podcasts on living-room devices in October 2025 alone. That is a platform figure for one surface: podcasts watched on TV screens via YouTube, not total podcast consumption.

The Sounds Profitable 2025 creator study found that 71 percent of surveyed podcast creators now include video in their shows. The figure describes a studied creator population in 2025, not a listener figure.

The shape these two support is the one you need for the audit: a sizeable surface, and a visible shift toward video in the format itself, on screens where captions and comments are not the primary record of what was said.

What a complete comms record of a podcast mention looks like

Transcription can identify spoken mentions and their context. Speaker attribution, delivery and visual analysis add information the transcript may not capture. A serious evaluation separates these capabilities so you know what you are buying.

• Transcription. The episode is processed into timestamped text. Detection of a brand name happens here, along with the surrounding sentence or paragraph that gives it meaning.
• Speaker attribution. Which person on the episode said it. A host endorsing you and a guest attacking you are not the same event, and your comms team needs to know which happened.
• Tone from delivery. Tone depends on wording, delivery and context. On long-form speech, the three can disagree, and a transcript alone will not resolve them.
• A stated counting rule. A brand named once in passing and a brand that held the floor for eleven minutes should not both land in the report as "one mention" without a rule that tells the reader which is which.
• For video podcasts, visual brand appearances. Logos on the set, products held on camera, backdrop placements. On a video podcast this is often where the brand actually shows up, and transcription will not touch it.

A tool with transcription alone will give you detections and surrounding text. It will not give you speaker attribution, delivery-based tone, or visual placements. For the shape of a record that carries all five, see what is actually said and shown inside a video.

[Image placeholder: A side-by-side of the same episode as two datasets. Left panel: what a text-based tool records, a single row holding the episode title, the description and four social posts referencing it. Right panel: what in-episode analysis records, a timeline of the ninety minutes with brand mentions marked at their timestamps, each tagged with speaker, tone, and (for video) visual appearances. Caption: both panels describe one episode.]

The coverage test: analysis, discovery, and latency are three separate questions

A thorough evaluation separates what the vendor can analyse from what the vendor will find, and both from how quickly they alert you.

Test 1. Analysis capability. Label the episodes internally before the evaluation, including expected mentions, timestamps, speakers and any relevant visual appearances. Include a few episodes with no relevant mentions as negative controls. Give the vendor the episodes only, and withhold your labels. Then compare the vendor’s results against the labels you kept back. Without that separation, "we will tell you what we missed" has no reliable reference point, and the false-positive count is not meaningful either.

What to measure on the analysis report:

• Correct detections, misses and false positives against your withheld labels. Count all three.
• Timestamp accuracy. Does the reported timestamp land where the labelled mention actually is in the episode.
• Speaker and context accuracy. When speaker attribution is reported, does it match the label. When a surrounding sentence or paragraph is quoted, does it match the audio.
• Visual placements (for video podcasts). Did the system detect labelled on-screen logos, product-in-hand, or set placements, and at what timestamps.
• Coverage exclusions. Which shows, episodes, languages, platforms or episode lengths are explicitly out of scope. Written down.

Test 2. Historical discovery. Analysis capability on episodes you handed over does not establish whether the vendor would have found the episode on its own. Ask for the ingestion log for the shows on your list over the last six months: which episodes were processed, which were not, and why. If a vendor cannot show the historical log, their discovery is a story, not a measurement.

Test 3. Alert latency, going forward. Monitor newly published episodes against the vendor for a defined window. Measure two separate numbers: time from publication to processing, and time from processing to alert. Keep your own publication log for the target shows and manually check new episodes for relevant mentions, so missed episodes and missing alerts remain visible. Measuring only the alerts that arrive conceals the ones that did not. A replay of a past crisis cannot establish either number; only episodes the vendor meets for the first time can.

The same discipline applies to any format where speech carries the signal, including the broader surface in media monitoring when the reporter is a creator.

What to do with the result

Pick the ten podcasts your category actually lives on. Have those covered properly for a quarter, with the five record criteria satisfied and the three tests above run on your current vendor, rather than buying a number that claims everything. Ten shows you can name and audit beats a dashboard figure nobody can trace.

Then take the three tests into your next vendor review. The ones that can produce analysis reports scored against withheld labels, historical ingestion logs, and forward latency measurements against your own publication log are the ones with a real episode pipeline. The ones that answer with a demo are the ones to probe harder.

Key takeaways

• An export labelled "podcast coverage" can contain posts about episodes, mentions inside episodes, or both. Which dataset your report includes is a question you have to answer for your own stack.
• A text index was not designed to pick up a brand discussion at minute 34 of a ninety-minute episode. Unless the tool processes the episode or an available transcript, that discussion can be missed entirely, and a flat line in your podcast row can mean a quiet week or a reporting gap. The two look identical from the dashboard.
• A complete comms record of a podcast mention needs transcription with context, speaker attribution, tone from delivery, a stated counting rule, and, on video podcasts, visual brand appearances.
• The evaluation is three tests, not one. Analysis capability against labels you keep withheld. Historical discovery via the ingestion log. Alert latency against your own publication log, measured over a defined window.

Frequently asked questions

Can social listening tools track brand mentions in podcasts?

Episode-level coverage varies between tools, platforms, languages and formats. Many general-purpose listening tools collect posts and articles about podcasts, which is a different dataset from mentions inside the episode. Dedicated podcast monitoring tools advertise transcript search and timestamps, with varying scope. The right question for your stack is which shows and episodes it actually processes.

What is the difference between a mention of a podcast and a mention in a podcast?

A mention of a podcast is someone writing about an episode on social or in an article. A mention in a podcast is a brand named in the spoken content. The first measures audience reaction. The second measures what was actually said about your brand. Both can matter, and they are different exports.

How do you test whether your current tool covers podcast episodes?

Run three separate tests. For analysis, label a short list of episodes internally with expected mentions, timestamps, speakers and visual appearances, plus a few negative controls, give the vendor the episodes only, withhold your labels, and score the vendor’s results against them. For discovery, ask for the ingestion log of your target shows over the last six months. For alert latency, monitor newly published episodes for a defined window against your own publication log, so missed episodes and missing alerts stay visible, and measure processing time and alert time separately.

‍