About the AI
AudaStories is made by machines, end to end. Here is exactly how, in plain English, with the rough edges left on.
Every story is AI-narrated
There is no microphone, no studio, no voice actor. The narrator, Graham, is a synthetic voice generated by Inworld TTS. We keep Graham locked to one British male timbre across every story so the voice becomes familiar, but no human ever reads the script aloud.
The script is written by a language model
For each story we fetch a specific revision of the relevant Wikipedia article and hand it to a large language model with a prompt tuned for the length you picked. The model selects, orders, and rephrases. No editor, researcher, or historian reviews the script before you hear it.
Do not use AudaStories as a source
Treat this as a tour, not a citation. If a detail matters, follow the source link on the story page back to the Wikipedia article it came from. That article is the thing to quote, verify, and argue about. We are a way in, not the truth.
AI can make mistakes
Language models sometimes misread, invent, or confidently shuffle a date. We have guards around the prompt and we do not ask the model for facts that are not in the source article, but errors happen. If you spot one, tap the flag button on the player or email [email protected] and we will fix it.
Every story credits its source
Under the title of every story you will see the line "AI-narrated · Wikipedia · CC BY-SA 4.0" with a live link to the Wikipedia article. At the bottom of the transcript we list the exact revision ID we used and the date we fetched it, so anyone can reproduce the source we saw. Our scripts are released under the same CC BY-SA 4.0 licence as Wikipedia itself.
The audio carries an inaudible watermark
Inworld embeds a machine-detectable watermark in every generated audio file. You cannot hear it, but automated tools can, and it survives compression and sharing. This is how AudaStories satisfies the AI-generated content marking duty under Article 50 of the EU AI Act. We never transcode or re-encode the audio, so the watermark is always intact.
The full catalogue is public
Every entity we cover, every Wikipedia article, and every Wikimedia Commons image is listed on the attribution page, with per-asset credit and licence. If you want to audit us, start there.