Skip to content

How our AI works

The short version

Chance picks your cards. An AI writes the interpretation. People write the rules the AI works under and check the pages you can read without drawing a card. Nobody, human or machine, knows your future. The rest of this page explains each of those sentences.

Who chooses the cards

Not the AI. When you start a reading, our server shuffles all 78 cards using a cryptographic random number generator, the same kind of randomness used to create encryption keys. Every card has the same chance, and a second random draw decides whether a card is upright or reversed. In most readings you then pick from the cards laid face down.

The AI sees the cards only after they have been drawn. It cannot swap a hard card for a pleasant one, and neither can we. The card of the day is drawn once for your day, so reloading the page does not change it.

What the AI writes

  • Interpretations of readings. We use Claude models made by Anthropic. The AI receives the cards, their positions and your question, and writes one connected text about your situation. It is always labelled as written by an AI.
  • Short explanations in the yes or no tool. The answer itself comes from a fixed table; the AI only explains what it depends on.
  • First drafts of the pages about card meanings, spreads and guides. A draft is not published until a person has edited it.

The AI has no memory of you between visits unless you have an account with memory switched on; then it receives a short list of your recent readings and notes, which you can switch off.

What a person checks

  • Every published page. A human editor reads and corrects each page before it appears, and the page shows the date of that review.
  • The yes or no table. It is reviewed line by line by a practising tarot reader.
  • The rules for the AI. People write the instructions that say how the AI may and may not answer, and test every change on a set of difficult sample questions before it goes live.

What a person does not check is your individual reading: it is written for you on the spot, and nobody reads it before you do. That is why the automatic checks described below exist.

Limits

  • The AI does not know you. It knows only the cards and the words of your question.
  • It can be wrong and still sound sure of itself. If a reading does not fit, trust your own judgement over the text.
  • It does not predict events, name dates or promise outcomes.
  • It cannot know what another person thinks, feels or will do. Asked about someone else, it writes about your side of the situation.
  • It is not a doctor, a lawyer or a financial adviser, and it does not replace one. The site is for adults.

Safety

Before any card is drawn, your question is checked twice: by a small AI model and by a list of key phrases. If it touches self-harm, the fate of a missing or seriously ill person, war or similar matters, you do not get an ordinary reading. You see a calm page with help contacts for your country, or a gentle reading about getting through the waiting. These answers are free, and nothing is sold next to them.

After the AI has written a text, it is scanned for things we do not allow: promises, frightening claims, superstitious threats, medical statements. A text that fails is rewritten; if it fails again, you get a plain summary of the cards’ own meanings.

Your question is treated as text to interpret and never as an instruction to the AI. Daily limits protect the service from abuse.

If something looks wrong

If a reading frightened you, misled you or broke any rule on this page, write to hello@taromee.com and say which reading it was. We read these messages and use them to correct the rules.

Add Taromee to your home screen