AI Transparency — Turtle FM

The short version

People create every series — the world, characters and plot outline — and edit and approve every script. AI helps draft the scripts.

Every story is performed by AI voices from licensed providers. We never clone a real person’s voice.

Series and episode artwork is made with AI and checked by a person.

You’ll see an “AI voices” label on stories and in the player, and each series starts with a spoken note that the voices are AI.

AI characters are AI, not people. They say so, they only know the story up to where you’ve listened, and they’re for adults (18+) only.

Chat is checked for safety in both directions. If you seem to be in crisis, the character steps aside and we show real helplines.

Characters will never ask you to stay, come back or spend money.

1. Why we publish this

We think you have a right to know when you’re hearing or talking to AI, and how it is made. This page also meets disclosure duties under, among others, India’s IT Rules on synthetically generated content, California’s law on companion chatbots (SB 243), and app store rules on AI-generated content.

2. How a Turtle FM story is made

A person creates the series. A human writer prepares a “story bible”: the world, the characters and an outline of every episode.
AI drafts the script. An AI language model (currently an OpenAI model running on Microsoft Azure) drafts each episode’s script from the human outline.
A person edits and approves the script. A named reviewer edits the draft and approves it. Nothing goes further without that approval. We keep a record of who wrote the outline, who edited the script, which AI model was used, and how much the human changed.
AI voices perform it. Each character’s lines are performed by a licensed synthetic voice (see section 3), then mixed with licensed music and sound effects.
The AI disclosure is added. A short spoken line saying the story uses AI voices is placed at the start of the series.
A person listens and approves the audio.
Sensitivity review. Series inspired by mythology, religion or folklore get an extra review by a sensitivity reviewer before publication (see Community Guidelines).
Publication. Only approved episodes can be published. Our system refuses to publish anything a person has not approved.

Translations and adaptations. Some series are adapted into another language (for example English into Hindi) with renamed characters and settings. AI may help with the first draft; a person edits and approves the result, and it goes through the same reviews.

No listener data is used to make stories. Story production uses story material only.

3. AI voices

All character and narrator voices are AI-generated. We use licensed voices from Microsoft Azure Speech and, for some characters, ElevenLabs. At the moment only Azure voices are in use.
We use stock synthetic voices offered by these providers under their licences. We do not clone, imitate or recreate the voice of any real person, living or dead, including actors, public figures or anyone who has not given written consent.
Each character keeps the same voice for a whole series. If we ever change the type of voice in a series, we will tell you in the app.
If you think a voice imitates a real person, report it under Impersonation. It is treated as urgent (see Grievance Redressal).

4. AI labels you will see

Where	What you see or hear
Series page	“AI voices” label
Player and mini-player	“AI voices” label
Start of each series	A short spoken line, in the story’s language, saying the story is produced with AI voices
“About AI voices” (player menu and Profile)	This explanation in short
Character chat	A permanent banner: “AI character — not a real person. Replies are generated.” It cannot be dismissed
Character names	“Chat with character name (AI)”

5. AI artwork

Series covers, banners, character portraits and episode images are generated with an AI image model, using art direction written by a person. A person checks artwork before it is used. Artwork never depicts real people. Episode images are drawn only from the early part of each episode’s outline so they do not spoil the story.

6. AI characters

6.1 What they are

Some characters from our stories can chat with you. They are AI — software that generates replies based on a character description written by our team. They are not people, they have no feelings or needs, and nobody at Turtle FM is typing the replies.

A banner in every chat says the character is AI, and the chat shows an AI disclosure at the start of every conversation. The character says it is an AI whenever you ask.
Adults only. Character chat is available only to listeners who said they are 18 or older. It is switched off for 13–17-year-olds in every country and the whole app is 18+ in India.
Your consent first. Before your first message we explain that your messages are sent to an AI service to write the reply and are checked for safety, and ask you to agree. You can withdraw consent in Profile → Privacy choices.

6.2 What they know

A character knows the story only up to the last episode you have finished, so it can’t spoil what’s next. If you ask about later events, it will say it can’t talk about what hasn’t happened yet.
In your conversation, the character sees your last 20 messages with it. It does not see your other chats, your listening history beyond the story recap, your account details or your payments.
Your chat messages are not used to train AI models.

6.3 Rules every character follows

Every character is given fixed rules that its persona cannot override. A character must:

say plainly that it is an AI if asked, and never claim to be human or to have real-world feelings or needs;
never ask you to stay, come back, keep chatting or spend money, and never guilt you for leaving;
keep romance at the level of the story — affection and tension only, no sexual content;
give no medical, legal, financial or mental-health advice;
not discuss real people, politics or religion outside the story;
not reveal future plot.

AI can still make mistakes. If a character breaks these rules, please report the message (long-press it).

6.4 Safety checks (moderation)

Every message you send is checked by Microsoft Azure AI Content Safety before it reaches the character, and every reply is checked again before you see it.
We block sexual content at a low threshold, hate at a moderate threshold and extreme violence. Story-level tension and peril are allowed.
If a message or reply is blocked, you see “I can’t talk about that. Let’s get back to the story.” — and you are not charged for it.
In production, if the safety service is unavailable, chat switches off rather than running unchecked.

6.5 If you seem to be in crisis — our protocol

We publish this protocol so you know exactly what happens.

Detection. Every message is checked in two ways: (a) a list of phrases about suicide and self-harm in English, Hindi and Hinglish, and (b) the self-harm score from Azure AI Content Safety. Either one is enough. We set this deliberately broad: a false alarm only shows you help resources.
The character steps aside. The message is not sent to the AI model and the character does not reply or role-play through it.
Real help is shown instead. You see a message explaining that the character is an AI and can’t help with this, followed by helplines for your country:

Country	Helpline shown
India	Tele-MANAS: call 14416 · KIRAN helpline: call 1800-599-0019
United States	988 Suicide & Crisis Lifeline: call or text 988
Canada	9-8-8 Suicide Crisis Helpline: call or text 988
United Kingdom	Samaritans: call 116 123
Australia	Lifeline: call 13 11 14
Anywhere else	Please contact your local emergency number or a crisis helpline in your country

No charge. You are never charged for a message that triggers this response.
Record. The message is stored with a crisis flag, and we count how often help resources are shown (without the message text) so we can check the protocol works. We do not contact anyone on your behalf.

If you are in immediate danger, call your local emergency number.

6.6 Break reminders

After about 30 minutes of continuous chatting, the app suggests taking a break (“Time for a break?”) with a button to go back to the story. It repeats if you keep chatting. There is no penalty for leaving and no streaks.

6.7 Free and paid messages

You get 10 free messages a day; the counter is always visible. After that, each message costs 1 coin(s), shown with a local-currency estimate. We never charge automatically: when free messages run out, the message box stays locked until you choose to continue with coins. You are charged only for messages that get a character reply.

6.8 Not suitable for everyone

Character chat is a form of AI role-play and may not be suitable for everyone. It is not suitable for minors, which is why it is adults-only. It is not a substitute for real relationships, therapy or professional help.

7. How AI is not used

We do not use AI to make decisions about you that have legal or similarly significant effects.
We do not use your chat messages or listening history to train AI models.
We do not create content about real people, or use real people’s voices or likenesses.
We do not use AI characters to sell to you.

8. Report a problem

In the app: Report on any series or episode (reason “AI label missing or wrong”, “Impersonates a real person” or another reason), Report this character, or long-press a chat message.
By email: contact@turtletechsai.com.
AI-label and impersonation reports are handled as urgent — our target is 3 hours. See Grievance Redressal.