Can an AI really read tarot? A skeptic's guide for skeptics

If you came here suspicious, you are our audience. We are going to do something most companies will not, which is to argue against ourselves first, and then explain why we still think the product has a use.

Here are five claims a careful skeptic might raise, what we think is right about each, and what we think is missing.

"Tarot is just confirmation bias."

Largely true. The reader pulls a card, reads a few words of generic-enough text, and projects their situation onto the symbol. The same trick is used by horoscopes, MBTI, and the back-of-magazine quiz. There is robust psychological evidence for the Barnum effect, and tarot lives inside it.

What is missing: the projection itself can be useful. The deck is essentially a randomized inkblot test with a thousand years of literary scaffolding underneath. Used badly, projection is self-deception. Used well, projection is the thing that gets you to look at something you would otherwise refuse to. The trick is whether the user is interested in seeing themselves, or in being told what they want.

"AI cannot really know anything. It is just predicting the next token."

Also largely true. A language model is not channeling cosmic insight; it is producing language conditioned on a prompt. There is no extra knowledge of your future hidden in the weights.

What is missing: producing language is exactly what tarot interpretation is. A reader does not need supernatural access to your future; she needs the ability to articulate what is on the table in a way you can hear. If a language model can do that — at the right register, with appropriate restraint — the function is performed. The product does not require AI to be magical. It requires AI to be a good interpreter.

"Card meanings are arbitrary."

Half true. Some tarot meanings are arbitrary, codified at specific moments in the deck's history (the Rider-Waite-Smith Three of Swords, the Marseille reading of the Hanged Man). But "arbitrary" is doing more work than it should. The meanings are arbitrary in origin and dense in convention; they have been used by millions of people for centuries, which makes them less arbitrary than the word implies, in roughly the way a language is.

A reading is not undermined by the conventional origin of meanings any more than a poem is undermined by the conventional origin of words.

"Selling spirituality is exploitative."

Often true. The wellness and divination industries are full of dark patterns — fake urgency, expensive "premium" readings that claim to reveal the "true" future, upsells targeting emotional moments. Skepticism about the category is warranted.

What is missing: this is a critique of business practices, not of tarot. A tarot product can be built without manipulation. The test is whether the free tier is honest, whether the paid tier offers more of the same thing rather than artificially gated insight, and whether the product survives a user who decides to read once and never come back. We think the answer should be yes to all three.

"People will use this instead of therapy."

A serious concern. Anything that performs the function of "someone reflecting your situation back to you" risks being substituted for real psychological care. This is not a problem unique to tarot — it applies to journaling, podcasts, group chats, and chatbots in general.

What is missing: we are explicit, inside the readings, about what tarot is not. We redirect health and crisis content. We do not pretend to do what a clinician does. We do not believe a reflection tool is a substitute for therapy, and we will not let our product imply it is. We do believe that for the vast majority of moments people use a tool like this — a small decision, a late-night question, a moment of doubt — the alternative is not therapy. It is endless scrolling. We would rather they read a card.

Where we end up

Most of what the skeptic is suspicious of, we are also suspicious of. We think the version of AI tarot that survives the critique looks like this: an honest reflection tool that does not predict, does not run dark patterns, does not pretend to do what therapy does, and respects the user enough to tell them when the cards are saying nothing they didn't already know.

If you came here a skeptic, we hope you leave a slightly more useful kind of one. We are not asking you to believe in tarot. We are asking you to consider that the version of the practice we are building is small, honest, and easier to live with than the one you were skeptical of.