Guide · 5 min

Reading tarot with ChatGPT

ChatGPT gives a warm, fluent reading the moment you ask. Below are three prompts worth keeping — paste them straight in. After that, the honest part: what none of them can do, and what to use when you want the draw itself to be real.

Copy and paste

Three prompts worth keeping

One card

Draw one tarot card for this question: [your question] Commit to the card and its orientation before you explain anything — name it in your first line, upright or reversed. Then read it in three short paragraphs: what the card shows, how it answers the question, and what it asks of me. Don't soften it, and don't add a second card.

Naming the card first keeps it from drifting into a horoscope shaped around whatever you just said.

Three cards

Lay a three-card tarot spread for this question: [your question] The positions are past, present, and what is forming. Name all three cards with their orientations first, in one line, before you interpret anything. Then take each position in turn in a short paragraph, and close with one sentence on what the three say together. No extra cards, no hedging.

Asking for all three names up front stops it inventing a fourth card mid-reading to tidy up the story.

When you don't want to be flattered

Read these tarot cards for me, and read them honestly: [your cards] You don't know what I want to hear, and you shouldn't guess. If a card is difficult, say so plainly. If two cards contradict each other, leave the contradiction standing rather than resolving it. Speak plainly, and stop when you have said the true thing.

A model trained to be agreeable will read for your mood unless you tell it not to.

  1. 01

    What the three have in common

    Each one makes ChatGPT commit before it explains — the card named in the first line, the orientation stated, no second card slipped in later. That single constraint is most of the difference between a reading and a horoscope, because it stops the model from choosing a card to fit an interpretation it has already started writing.

  2. 02

    The one thing a prompt can't do

    There's no shuffle behind the words. ChatGPT has no deck state, no real randomness, no reversal odds — ask twice and you may get the same “random” card, or a card chosen to match what you seem to want. A prompt can shape the reading; it can't make the draw real.

  3. 03

    Give ChatGPT a real deck

    In ChatGPT's developer mode you can add the Familiar connector — the same one Claude uses. Then “draw three cards on the move” pulls from an actual shuffle of seventy-eight, upright or reversed by real odds, with the card art to match.

Improvised vs. drawn for real

“One card on whether to take the offer.”

Two of Wands — wands suit, minor arcana. Planning, decisions, future vision.

With the connector, ChatGPT reads what the deck actually turned — say the Two of Wands, upright: a real choice between the safe harbour and the wider plan, named because the card came up, not because it fit the mood of the chat.

Add the deck to ChatGPTFree · same connector as Claude

Prefer Claude? The same deck works there too — see reading tarot with Claude. The connection guide covers the developer-mode steps.