How Lullable is made: where the facts come from, and how the voice is produced

The voice is synthetic, and that is the disclosure

It is the first thing anyone wants to know, so it may as well be the first thing said. The narration is text-to-speech. You can hear it in the first ten seconds if you are listening for it, and pretending otherwise would be both futile and beneath the material.

What is worth adding is that this was not a cost decision dressed up as a principle. A voice that works at night has to hold a narrow pitch range for forty minutes, never accelerate, never lift at the end of a sentence to promise that something is coming, and get quieter across the episode without anyone noticing the slope. Those are the properties that let a voice disappear, and they are the hardest things in the world for a trained reader to sustain, because the instinct to perform is the instinct the craft is made of. A synthesised voice is flat by default, and it is still flat at minute thirty-nine.

The markup it receives is deliberately impoverished. The script carries pauses and nothing else — no emphasis, no rate changes, no shaping of individual lines. Anything expressive we could add would be a thing for the listener to notice, and being noticed is the failure mode.

Where the facts come from

The stories are true, which means the sourcing has to be real rather than atmospheric. The same applies to these essays, and the rule here is enforced by the program that builds the site rather than by anyone's good intentions.

Any post that asserts a year, a percentage or a measurement must carry at least two source URLs in its frontmatter. If it does not, the build stops and nothing is written — not the post, not the index, not the sitemap. The rule is keyed on the claim rather than on the kind of post, which matters more than it sounds: a conceptual piece with nothing to check cites nothing at all, because a citation nobody needed looks like diligence and is worse than having none.

Facts are read at their source rather than from the coverage of the source. Where a news write-up of a study repeats a striking number that the study's own abstract does not contain — which happens more often than it should — the number does not appear here. The story is duller without it. That is usually a sign it is correct.

What the build refuses to print

There is a second gate, and it is the one that exists because nobody reads these pages before they go live.

Sleep is health-adjacent, and the category it sits in markets almost entirely in promised outcomes. So a page here is not permitted to promise a result. It may describe a mechanism and then stop. A list of banned phrases — the clinical ones, and the softer marketing ones that mean the same thing — is checked against every essay, every title, every description and every social card before anything is published. A match aborts the build. Lullable is not a treatment, it is not a medical device, and persistent insomnia is a matter for a clinician rather than for an audio app.

Why say any of this unprompted

Because the honest version is more useful than the silence, and because the interesting question is not whether the voice is produced. It plainly is. The question is whether the material underneath it is true, whether the boring history and quiet natural science you are being read actually happened that way, and whether anyone would tell you if it did not.

So: they did, we would, and the address for corrections is at the bottom of every page. The full catalogue is here, and what the app does is answered a question at a time, in the same register as this.