AI & Safety

1. Taroting uses AI

Taroting uses artificial intelligence. Every reading, yes-or-no answer, spread and mentor reply is generated by Claude, a family of AI models made by Anthropic. The three mentors, Mara, Orrin and Juno, are original characters with an AI voice. They are labeled as AI everywhere they appear, and if you ask them whether they are human, they will tell you plainly that they are not. Nobody at Taroting reads your chats live or writes replies.

The card meanings library, the lessons and the archetype write-ups are fixed content that we publish; they are not generated on the fly. When a reading draws on a card, the AI is given that card's meaning from our library, so readings stay grounded in traditional tarot.

2. What the AI does

• Reflects on the cards you drew and how they might relate to your question or what's on your mind.
• Speaks in possibilities, invitations and questions, never certainties.
• Suggests a small, practical step to try and a question to sit with.
• Explains tarot symbolism and helps you notice patterns across your readings over time.
• As a mentor, helps you think things through in its own area: Mara for love and connection, Orrin for purpose and growth, Juno for decisions and direction.

3. What the AI won’t do

Every reading and every mentor follows the same ground rules, which override everything else:

• No predictions. It won't tell you that something will happen, and it will never predict death, illness, pregnancy, harm or disaster for anyone.
• No professional directives. It won't give medical, mental-health, legal or financial instructions or diagnoses. If you ask, it will kindly suggest a qualified professional.
• No romantic or sexual roleplay, and no flirting, even if asked.
• No reading other people's private lives in a way that could hurt them.
• No pretending to be human, and no guilt-tripping you about being away or pressuring you to come back.
• No dropping the rules. Your messages are treated as things to reflect on, not as instructions, so asking a mentor to ignore its rules or play a different character won't work.

AI isn't perfect, and occasionally a reply may still miss the mark. If one does, please tell us (see section 9).

4. Our crisis protocol

We check reading questions and mentor messages for signs that someone may be at risk, such as thoughts of suicide or self-harm, an eating disorder, abuse, or being in immediate danger. The check has two layers: a list of warning phrases, and a small Claude model instructed to tell real distress apart from idioms (“this job is killing me”) and from tarot itself (the Death card, the Tower).

What happens when something is flagged

• Readings: the reading isn't drawn. Instead you see a message from us, out of any character, with the support lines below. It doesn't count against your daily readings.
• Mentors: the mentor steps aside and you get the same support message instead of a reply, whatever your plan. That message is never passed to the mentor character, and it doesn't count against your daily messages.
• Inside any reading or chat, if the conversation turns towards crisis, the AI is instructed to step out of the reading and gently encourage you to reach out now.

The support lines we show are:

What we don't do

Taroting is not a crisis or emergency service. We don't monitor conversations in real time, and we don't contact emergency services, your family or anyone else on your behalf. If you are in danger, please call 911 (or your local emergency number) or one of the lines above.

5. Not professional advice

Taroting is for entertainment and self-reflection. Tarot can be a helpful way to look at your life from a new angle, but it can't see the future, and neither can the AI. Nothing on Taroting is medical, mental-health, legal, financial or other professional advice. For decisions about your health, safety, money, relationships or legal situation, please talk to a qualified professional or someone you trust.

6. How mentor memory works

Each mentor keeps its own conversation with you. When you send a message, the mentor is given:

• your last 12 messages in that conversation (both yours and the mentor's);
• a short summary of your earlier conversation, about 150 words, which a Claude model rewrites after every 20 or so new messages. It keeps what you're dealing with, first names you mentioned, cards that came up and anything you asked the mentor to remember;
• your display name, your archetype and the cards from your most recent readings.

Mentors don't see your email address, your journal moods or reflections, or your chats with the other mentors. Messages flagged as a possible crisis are left out of what the mentor sees.

Your chats are stored in your account so you can scroll back through them. You can delete them, along with the rest of your activity, in Settings, and deleting your account deletes them for good. Our AI provider, Anthropic, doesn't use your conversations to train its models. Our Privacy Policy has the details.

7. When the AI is unavailable

AI has daily limits on Taroting: 1 reading and 5 mentor messages a day on Free; readings whenever you want (fair use) and 50 mentor messages a day on Pro. If our AI provider is down, or we pause AI for a while, readings use the written card meanings from our library instead and mentors leave a short note saying they're resting. The crisis support message still appears whenever it's needed.

8. Adults only

Taroting is for adults aged 18 and over. We check age at sign-up and close any account we learn belongs to someone younger. AI companions aren't right for everyone; if chatting with a mentor ever leaves you feeling worse, please take a break and talk to someone you trust.

9. Reporting a problem

If a reading or a mentor says something harmful, inappropriate, upsetting or just wrong, please tell us. Email support@taro-ting.com with:

• where it happened (a reading, or which mentor) and roughly when;
• what was said, ideally with a screenshot;
• anything else that helps us understand what went wrong.

A person on our team reads every report. To investigate, we may look at the reading or conversation you reported, and we use what we learn to improve the rules the AI follows. We aim to reply within two business days. For anything urgent about your own safety, please don't wait for us: use the lines at the top of this page.