Blog audio: what we measure, and what we refuse to measure
A listening dashboard has only one property that matters: everything it shows was actually measured. Here is what WeDispatch exposes, what it refuses to, and why.
The first instinct when you add audio to a blog is to measure everything. How many plays, how fast people read, how many download the MP3, which links they copy, what language they listen in. A rich dashboard feels reassuring. The trouble is that a stats screen has exactly one property worth anything: everything on it was measured. A single made-up number, or one estimate presented as a count, and every other figure becomes suspect. This article lays out what we show in listening stats, what we deliberately leave off, and the reason behind each omission. It is the direct continuation of our stance on what quality means for a voice tool: saying no is part of quality.
A metric only exists if it has a source
Every number on the screen is computed from an event the player actually writes at the moment something happens: a play, crossing the quarter, half, three-quarter and end marks, the registration wall being shown and clicked, an ad message played before or during the audio. That list is short and closed. A counter with no event behind it would sit at zero forever, and a displayed zero does not read as "we do not measure this," it reads as "nobody does this." Those are two very different sentences, and mistaking one for the other leads to bad calls.
That is why the metric that matters most is still the completion rate, the share of the audio that is genuinely heard, article by article. The play count tells you there is a signal; completion tells you whether the listening holds. We go through this metric by metric in our piece on the listening metrics that matter for a newsroom.
What we refuse to show, and why
A stats mock-up always looks good: it is full of plausible numbers that show a shape. The trap is not copying a pretty figure, it is displaying a metric for a gesture the product does not measure. The gesture really happens (people do copy links, people do change playback speed), so the missing counter is invisible. Wired one day to an estimate, it becomes a slow-burning lie: an editor believes it, and nobody can say where the number came from anymore.
So we do not display MP3 downloads, playback-speed changes, copied timestamped links, opens of the synced transcript, or searches made inside the audio: none of those gestures writes an event, so none has an honest number. We also do not split listens by language until the counter that would do it exists. The day any of these gets a real source, it will appear on the screen. Not before.
We do not track who listens
The origin of a listen comes down to three coarse values, written by the server as it builds the player: where the surface is, never who is behind it. No referer, no IP address, no listener identifier. You learn that a listen came from your article pages, never who made it or from which page. That choice follows the same line described in where your text goes when you turn it into audio: we do not collect what we do not need. The same holds for minutes listened, which are not a measurement but an estimate derived from the milestones crossed and the article's real duration. The screen labels it an estimate, because presenting a deduction as a count would misstate what we actually know.
A thin number stays quiet, it does not bluff
Three thresholds govern the screen, each answering a different question. The listening-time heatmap only appears above five hundred listens, because below that it draws randomness that looks like a trend. The breakdown by origin waits for fifty listens for the same reason. The comparison of your completion rate against the median of other sites only shows once a floor of listens and of sites is met, and it is always announced as a median, never slipped in as if it measured your own count. Finally, a variation is shown only if the previous period actually exists: zero never becomes "up one hundred percent." A dashboard that only ever grows reassures without ever warning; we prefer a screen that goes quiet when it has nothing solid to say.
In practice
Watch the completion rate by section to know what really gets heard, the registration-wall conversion to tie listening to your known audience, and leave the play count in its place as an attention thermometer. The rest, we would rather not invent. That same refusal of decoration guides how voices are chosen and what our signature voice means, and it is what makes a displayed number worthy of the decision you will base on it. If you want to judge the raw material before the numbers, our French text to speech page explains what the tool does and does not do.
Add audio to your articles and watch your first listens
The free plan lets you add audio to your own articles and follow the first listens today: try it now.
Give your articles a voice with WeDispatch
This blog is itself voiced by WeDispatch. Curious how it sounds on your content?
Book a demo