We logged 1,370 AI tarot readings between January and May 2026. Anonymized. No fake stats, no editorial cherry-picking — just what people actually drew when they sat down with a question.
Most of those readings (76%) came from anonymous guests — no account, no email, just someone visiting the site and asking. The remaining 24% came from 69 registered users who came back enough times to be worth tracking individually. Roughly 750 unique participants in total, by IP fingerprint and account combined.
The most drawn card across the entire dataset?
Knight of Wands. 78 appearances.
Not The Lovers. Not The Sun. Not any of the cards Pinterest infographics keep promising. A minor arcana court card you've probably never thought twice about.
Here's what we found, why it matters less than you'd think, and what 1,370 draws actually tell us about how AI tarot works.
The top 15 cards
Across 1,370 readings, here's what the deck handed people most often:
| Rank | Card | Draws | Type |
|---|---|---|---|
| 1 | Knight of Wands | 78 | Minor (Wands) |
| 2 | The Hanged Man | 76 | Major |
| 3 | Queen of Pentacles | 69 | Minor (Pentacles) |
| 3 | Five of Wands | 69 | Minor (Wands) |
| 3 | The Tower | 69 | Major |
| 6 | Ten of Swords | 68 | Minor (Swords) |
| 6 | Six of Swords | 68 | Minor (Swords) |
| 6 | Three of Wands | 68 | Minor (Wands) |
| 6 | Four of Cups | 68 | Minor (Cups) |
| 10 | The Hierophant | 67 | Major |
| 10 | Eight of Wands | 67 | Minor (Wands) |
| 12 | The Sun | 66 | Major |
| 12 | King of Pentacles | 66 | Minor (Pentacles) |
| 14 | The Chariot | 65 | Major |
| 16 | The Moon | 64 | Major |
Most tarot blogs publishing "most drawn cards" lists either don't have data, or pick the cards that look good on a Pinterest pin. The Lovers. The Sun. The Wheel of Fortune. Cards that confirm what readers want to feel.
Our data says different. Knight of Wands #1. A card about restless energy, action without much patience, the urge to start something before knowing where it goes.
The Hanged Man — top Major Arcana with 76 draws — is even more telling. People came to a divination tool 76 times and the cards handed them suspended judgment, surrender, the moment before clarity.
That's not what people pin on aesthetic boards. That's what they ask about when nobody's watching.
What the math says
Major:Minor ratio across 1,370 readings: 28.4% / 71.6%.
A Rider-Waite deck has 22 Major Arcana cards out of 78 total. Pure math says 28.2% / 71.8%.
The deviation between observed and expected: 0.2 percentage points.
That's exactly what statistical randomness looks like. No mystical patterns. No "the universe is communicating." The card frequency we're seeing is what you'd predict if you had a python script flipping a fair 78-sided die.
For per-card significance — the kind of statistical claim where you'd say "Knight of Wands appears MORE than chance would predict" — we'd need somewhere around 6,000 readings, not 1,370. Right now, we're halfway there. Knight of Wands ranking #1 is interesting. It's not yet statistically significant.
This matters because a lot of tarot writing assumes patterns where there are none. If a card "comes up a lot for you," that's mostly your memory selecting confirming instances and ignoring the others. Confirmation bias is older than tarot. Tarot just provides better-looking hooks for it.
Why this might still be useful (the honest version)
If the cards are random, why log 1,370 readings?
Because the questions aren't random.
Across the dataset, 354 questions out of 1,261 with text content (28.1%) were future-oriented — variations of "will this happen," "when will it happen," "is this the right time." Another 169 (13.4%) were about love. 129 (10.2%) asked about work or career.
That's a useful signal. Not about whether tarot works. About what people use tarot for.
People reach for tarot when they're sitting with uncertainty about a specific area of their life. The cards don't predict the answer. The cards force the question to take a concrete shape — three cards, three positions, suddenly you have to articulate what you actually wanted to know.
The retention pattern backs this up. 69 registered users, averaging 4.9 readings each across the registered subset. Plus ~680 unique guest IPs, of whom most did 1-3 readings — but a long tail returned 5, 10, even 30 times without ever creating an account. That's not a lottery ticket pattern. That's a journaling habit dressed up.
The Tuesday effect
Here's the strangest finding in the dataset: Tuesday spikes 37% above average reading volume.
| Day | Readings | Share |
|---|---|---|
| Monday | 198 | 14.5% |
| Tuesday | 271 | 19.8% |
| Wednesday | 182 | 13.3% |
| Thursday | 210 | 15.3% |
| Friday | 185 | 13.5% |
| Saturday | 167 | 12.2% |
| Sunday | 157 | 11.5% |
A perfectly even distribution would put each day at 14.3%. Tuesday gets 19.8%. Sunday drops to 11.5%.
Why Tuesday? Honestly: we don't know yet. Hypotheses we can't rule out:
- The work-week reset. Monday is too chaotic. Tuesday is when the week's actual problems become visible.
- Sample bias. Maybe one user with high reading frequency happens to do most of theirs on Tuesdays.
- Western media patterns — many self-help newsletters drop Tuesday mornings.
To know which, we'd need user-segmented day-of-week analysis. That's our Q3 deliverable.
What this dataset is not
Some honesty about the limits.
We don't know if AI tarot is "accurate." Nobody does, because "accurate" isn't well-defined for tarot. The closest measurable thing is whether users come back — and 69 of ~750 participants registered to track their reading history (~9% conversion to account creation), with another long tail of guests returning by IP without registering. That's decent engagement for a free tool, but doesn't prove anything about the readings themselves.
We don't measure outcomes. We don't know what happened to the person asking "will my relationship work" after they got The Tower. We don't know if the reading helped, hurt, or did nothing.
Our sample is small. 1,370 readings across ~750 unique participants (69 registered + ~680 anonymous by IP). Our flagship insight (Knight of Wands #1) won't reach statistical significance until ~6,000 readings. We'll publish updates quarterly.
This is observational data, not an experiment. No control group. No randomization beyond what the AI's RNG provides. We can describe what people drew. We can't draw causal conclusions.
What we can say
Three things, with confidence:
-
The card distribution looks random. Major:Minor ratio matches mathematical expectation within 0.2 percentage points. If the AI is biased toward any particular card, we don't have the sample size to detect it.
-
People mostly use tarot for uncertainty. 28% of all questions are about the future. Another 24% are about love or work — areas where outcomes feel out of personal control.
-
Returning users exist. 69 registered, averaging 4.9 readings each. Plus a long tail of unregistered returning visitors by IP. That's a habit, not a one-off curiosity. The product makes some users return. We don't claim to know why yet.
Cite this dataset
If you're writing about AI tarot, divination markets, or applied skepticism, you can reference this work:
aimag.me Tarot Reading Dataset (n=1,370). Collected 2026-01-01 to 2026-05-02. Anonymized, open methodology at aimag.me/research/methodology.
We'll update quarterly. Next snapshot: 2026-08-01.
Where this goes next
We're publishing this as research, not as a sales funnel. If you're curious enough to try a reading yourself — and contribute one anonymous data point to the next snapshot — that's how this dataset grows.
If you'd rather just download the raw data and run your own analysis, we'll have that available at /research/dataset starting Q3 2026 once anonymization audit completes.
The math says random. The retention says something else is going on. Both can be true.