Why a single figure cannot sum up an audience
A podcast’s growth is often presented through a number of downloads or views. That figure is easy to communicate, but it does not answer every important question. How many distinct people actually started listening? What share went beyond the first few minutes? Do listeners come back for the next episode? Is the show growing with a regular audience, or simply benefiting from one unusually visible episode?
These questions matter to creators, editorial teams, education and training providers, institutions and organisations that want to understand the reach of audio content. They matter to listeners too, because audience statistics sometimes influence how a programme is funded, distributed or assessed.
Sound measurement should therefore distinguish at least four dimensions:
- reach, meaning the number of people or devices exposed to the content;
- consumption, which shows whether the episode was actually started or downloaded;
- engagement, measured by how long and how much of the episode was listened to;
- loyalty, which shows whether listeners return over time.
These indicators are not interchangeable. A podcast can have modest reach but a very loyal audience. Conversely, an episode can rack up a large number of downloads while being abandoned quickly.
Downloads, plays and listeners: three different things
In podcasting, a download generally corresponds to a request sent to the server hosting the audio file. That request can be triggered by an app, an embedded player, a caching system or a manual download. It does not always prove that a person listened to the episode.
A play, or start, is a signal closer to actual listening, but its definition depends on the service supplying the data. Some tools count a play as soon as the file is launched. Others apply a minimum threshold of time or progress. It is therefore worth checking the exact definition before comparing two dashboards.
The number of unique listeners seems more telling. It nonetheless remains an estimate. Services can identify a device, an account or a browser, but they do not always know whether several people share the same device. The same person may also listen to the same programme on several devices and be counted more than once.
The recommendations of the IAB Tech Lab, notably its Podcast Technical Measurement Guidelines, aim to harmonise filtering and counting methods. They do not remove every difference between platforms, but they provide a framework for excluding some automated requests, duplicates and incomplete downloads.
Good practice is to use precise vocabulary in reports:
- compliant downloads or filtered downloads, when the method follows a defined standard;
- starts, for plays recorded by a tool;
- estimated listeners, if the metric relies on imperfect deduplication;
- completed listens, only when the platform clearly defines what “complete” means.
Retention reveals what volume conceals
Retention measures the proportion of listeners who are still there at different points in an episode. It can be expressed as a percentage of the total running time or shown as a curve indicating how many listeners remain after one minute, fifteen minutes or a particular segment.
This metric is often more useful than the raw number of downloads for judging how well the title, the editorial promise and the content fit together. A sharp drop at the start may point to an overlong introduction, unclear expectations or a mismatch between the description and the episode. A one-off dip may coincide with an advert, a change of subject or a passage that is hard to follow.
A retention curve should not, however, be read as an absolute measure of attention. The data do not always reveal whether the file was played in the background, sped up, paused or resumed later. Moreover, some players do not pass on the same level of information.
To track this metric, a few reference points are useful:
- the share of starts that go beyond the first few minutes;
- the proportion reaching the halfway point;
- the share continuing to the final minutes;
- recurring drop-off points across several comparable episodes.
Comparisons should cover episodes of a similar format. Comparing an hour-long interview with a ten-minute segment can lead to the wrong conclusion. Total running time, type of content, the placement of the call to action and the presence of an advert all need to be taken into account.
Loyalty metrics are often the most revealing
Loyalty describes a podcast’s ability to bring its audience back. It can be observed through listening frequency, the share of listeners who have played several episodes and the change in plays between early releases and later ones.
A useful indicator is the ratio between the listeners of a recent episode and those who come back for the next one. It is not a universal rate, as platforms do not all identify people in the same way. It nevertheless raises a concrete question: is growth coming from new, occasional listeners or from a core audience that is widening?
Publishing regularity can also be examined. A weekly programme that keeps a stable audience from one episode to the next should not be read in the same way as a series released in seasons. In the latter case, the analysis needs to separate launch performance, listening during the season and the audience’s return for the following season.
Subscription or follow statistics are useful, but they should not be equated with listening. Someone can follow a programme without starting every episode. Conversely, someone can listen to several episodes without subscribing. A subscription is therefore a signal of intent, not proof of listening.
To make loyalty visible, a tracking table might include:
- listeners to previous episodes;
- estimated new listeners;
- the number of episodes played per person, where this data is available;
- how often listeners return over four to eight weeks;
- the difference between the launch audience and the settled audience.
Why downloads can give a misleading picture
Downloads remain a useful indicator of how widely a file is distributed, but several factors can inflate them or distort how they are read.
- Automatic downloads: some apps fetch new episodes as soon as they are published, before anyone listens.
- Caching: a file can be stored temporarily on a server or a device, then counted using a method that does not correspond to immediate listening.
- Repeat listens: the same listener may restart an episode to find a piece of information, without being a new person.
- Incomplete downloads: a dropped connection can generate a request with no meaningful listening.
- Changes of method: a platform may change its counting rules, creating an artificial break in a historical series.
- An old episode resurfacing: a recommendation or a news story can trigger a spike that does not reflect the show’s steady growth.
The technical filters recommended by the IAB Tech Lab reduce certain anomalies, but they do not turn downloads into a perfect measure of audience. For this reason, a report should state the period, the source, the definition used and any changes of method.
An isolated spike should therefore not be presented as a trend. Look at several episodes, compare equivalent periods and check whether the increase also shows up in retention, returning listeners and interactions.
Platform data is valuable, but rarely comparable as it stands
Distribution and hosting services offer detailed dashboards. They can provide information on plays, devices, geographical areas, discovery sources and progress through each episode. Yet two platforms may attribute different figures to the same content.
The reason lies in their collection rules. One may count a play after a few seconds, while another waits for a different threshold. Some data is based on logged-in accounts, other data on device identifiers. Update schedules and rounding may also vary.
RAJAR’s MIDAS survey of audio listening in the UK, along with Ofcom’s research into media use, shows the value of a complementary approach. Self-reported or panel data does not measure the same thing as technical download logs. It can nonetheless help to place a show’s statistics within wider listening habits.
When comparing sources, it is better not to add their figures together mechanically. Instead, it is worth:
- choosing one primary source to track the trend over time;
- using the other sources to understand listener profiles and discovery paths;
- keeping a record of definitions and extraction dates;
- flagging any break in method or scope;
- comparing consistent periods and episodes.
A guide to telling real signals from false friends
| Metric | What it can show | Risk of misreading | How to use it |
|---|---|---|---|
| Filtered downloads | The technical distribution of files | They do not prove attentive listening | Track trends and check the filtering method |
| Estimated listeners | An approximation of human reach | A device does not always correspond to one person | Compare periods using the same source |
| Start rate | The ability to trigger a listen | Definitions vary between tools | Document the threshold used |
| Retention | How listening progresses through the episode | Data may be missing depending on the player | Compare similar formats and running times |
| Subscriptions or follows | An intention to come back | A subscription does not guarantee any listening | Cross-check it with listening to later episodes |
| Shares and comments | Signs of recommendation or involvement | Low interaction does not mean a small audience | Use them as complementary qualitative indicators |
Building a genuinely useful dashboard
An effective dashboard does not need to pile up every available metric. It should link each measure to an editorial or organisational question. For example: is the audience finding the programme? Do listeners stay long enough to grasp the subject? Do they come back the following week? Are the episodes reaching the people they are meant for?
A monthly baseline might include five groups of data:
- Reach: estimated listeners, compliant downloads and discovery sources.
- Consumption: starts, average listening time or proportion listened to.
- Loyalty: subscriptions, returning listeners and number of episodes played.
- Editorial quality: retention, drop-off points and performance by format.
- Context: publication date, running time, topic, promotional campaign and technical changes.
It is better to keep the raw data before calculating rates. A percentage without a denominator can be misleading. A 50% rise may simply mean going from two listeners to three, while a 5% rise may represent a far larger number on an audience that is already sizeable.
Tracking should also distinguish the performance of an episode from that of the series. A high-performing episode may attract people who do not come back. Conversely, a less visible episode may strengthen the loyalty of existing listeners. The two outcomes carry different value depending on the objective.
Interpreting results without confusing correlation and causation
A rise in audience after a promotional campaign does not prove that every element of the campaign had the same effect. Several factors may be at work at once: a topical subject, a recommendation, visibility in a search engine, a change of title or the release of a companion episode.
To better understand the possible causes, compare periods and document events. A simple method is to keep a record sheet for each episode with the date, format, running time, subject, promotion source and main metrics after one week, one month and three months.
Seasonal variations must also be taken into account. Holidays, the start of the academic year, exam periods and public events can all change listening habits. A one-off dip is not necessarily an editorial problem. Likewise, a rise is not necessarily lasting.
Finally, quantitative data benefits from being complemented by qualitative feedback, where it is gathered within a clear framework. Messages received, recurring questions and requests for clarification can show whether a subject was understood, misunderstood or found useful. They do not replace statistics, but they help to explain what the figures do not say.
FAQ: the main metrics at a glance
Is the number of downloads useless?
No. It tells you about the distribution of the file, especially when it is filtered using a recognised method. It becomes misleading when it is presented as an exact number of people who listened to the whole episode. It should be paired with metrics for starts, retention and returning listeners.
Which metric should you choose to measure growth?
There is no single metric. For reach, estimated listeners can be useful. For genuine interest, retention and the proportion listened to provide more information. For the strength of the audience, track returning listeners over several episodes and several periods.
Can you compare figures from two platforms?
Only with caution. Definitions, counting thresholds, identifiers and update schedules may differ. It is better to choose one reference source for trends and to use the other data to complete the analysis, without automatically adding the results together.
Is a growing audience necessarily a sign of success?
Not necessarily. The rise may be a one-off, tied to a particular topic or accompanied by a sharp fall in retention. More moderate but steady growth, with listeners who come back and listen to a large share of each episode, can be more significant in the long run.