About
How GuessMTG works
The Oracle isn’t AI. It can only reason with the stats defined on the card and in Scryfall. GuessMTG combines that structured card data with question interpretation and a deduction system to track exactly what you’ve established about today’s card.
This page is the deep dive: how the Oracle reads questions, how deductions work, when information is revealed, and how to ask better questions. For the one-minute version, see How to Play.
How the Oracle understands questions
You ask in your own words:
Is it red?
Does it make tokens?
Can it kill creatures?
Is it expensive?
The Oracle maps each question to one known characteristic or gameplay concept: a color, a card type, “mana value is 5 or more”, “removal”, and so on. The game then answers that check from the card’s real data, which comes from Scryfall. The Oracle understands the question; the card data decides the answer. The same question about the same card always gets the same answer, for every player.
There’s no AI or language model behind the answers. The Oracle reads questions with fixed rules and can only reason with the stats defined on the card and in Scryfall, so it can’t answer about anything those don’t record.
What you can ask about:
- Colors
- Single colors, colorless, multicolored, mono-colored, several at once (“Is it blue and green?”, “Is it only blue and green?”), guilds, shards and wedges (“Is it Izzet?”), primary colors (red, blue or green), and color identity. “Is one of its colors green?” works once you know it’s multicolored.
- Types
- Card types, “Is it a permanent?”, “Is it legendary?”, creature types and other subtypes, “Is it a split card?”, and “Can it be your commander?”.
- Mana
- Mana value (“Is it 3 mana?”, “Is the mana value even?”), the exact mana cost in any order (“Is it 2UB?”, “Does it cost {1}{U}{U}?” — written in capitals), colored symbols (“Is it double blue?”, “Does it have UU in its cost?”), generic mana (“Does it contain a generic mana symbol?”), and hybrid (“split”) or Phyrexian symbols.
- Stats
- Power, toughness and loyalty, including whether it has them at all (“Does it have toughness?”), even or odd (“Is its power odd?”), and power against toughness (“Is its power greater than its toughness?”).
- Abilities
- Activated, triggered and static abilities (“Does it have an activated ability?”), how many it has (“Does it have more than one ability?” — each keyword counts as one, a modal spell as one), ETB effects, tap abilities (“Is its tap ability free?”), whether any ability costs mana to activate (“Does it cost mana to activate an ability?” — Equip and Cycling count, “{T}: Add {G}” doesn’t), whether it can be tapped at all (“Can it be tapped?” — a creature, a Vehicle or a {T} ability), and equip costs, which aren’t the mana cost (“Is the equip cost 1?”, “Does it cost 2 to equip?”).
- Keywords
- Flying, trample, haste, deathtouch and every other keyword. What a card has is different from what it gives: “Does it give haste?” or “Does the equipped creature have haste?” asks about Equipment like Lightning Greaves, which gives haste without having it.
- Printing
- The set it was first printed in (“Is it from Dominaria?”, “Is it an MH2 card?”), a plane or group of sets (“Is it from Ravnica?”, “…Eldraine?”, “…the Kamigawa block?”, “…a core set?”, “…an Un-set?”), Universes Beyond (“Is it Universes Beyond?”), rarity and year, all from the card’s first printing (“Was it printed in the 90s?”). It also knows whether a card has been reprinted (“Has it been reprinted?”, “Is it in more than one set?”), but not where.
- Formats
- Legality in Standard, Pioneer, Modern, Legacy, Vintage, Commander, Pauper and more (“Is it legal in Commander?”, “Is it banned in Modern?”). Restricted counts as legal. Commander’s Game Changers list too (“Is it a Game Changer?”).
- Name
- First letter or a range of letters (“Does it start with A–F?”), vowel or consonant (“Does it start with a vowel?”; Y counts as a consonant), the last letter (“Does it end with N?”), how many letters (“Is the name longer than 12 letters?”; spaces and punctuation don’t count), how many words, a word in the name (“Does the name have Signet?”), and punctuation (“Does the name have punctuation?”).
- What it does
- Gameplay concepts like removal, ramp and card draw (see Semantic MTG concepts), what it targets (“Does it target a creature?”), and rules text (“Does it mention the graveyard?”, “Does it enter tapped?”, “Does it have reminder text?”). Scryfall-style
o:unlessworks as “Does its text say unless?”.
One question rule
Each message is one logical question. Compound questions are cut down to the first one:
Is it red and does it cost 4? → answered as Is it red?
The rest is ignored and the message still uses one question, and the Oracle tells you it trimmed it. This keeps the game fair: nobody gets two answers for the price of one. Phrases that are really a single question stay whole, like “3 or less”, “between D and M”, “power and toughness”, and several colors: “Is it blue and green?” is one question about its colors. A No to that only tells you it isn’t both, not which one is missing.
How deduction works
Every answer is stored as a fact, not just a yes or no. Together those facts form your knowledge state, which you can see under What you know. After each answer, the deduction engine checks whether your facts now logically prove anything new.
Is it white? No
Is it blue? No
Is it black? No
Is it green? No
Is it colorless? No
✨ Deduced: it’s red. Free
You aren’t given another free question. The game simply recognizes what your existing questions have already proven, and tells you. Four colors ruled out isn’t enough by itself, because the card could still be colorless. Ruling that out too leaves only red.
More examples of what the engine can prove:
| Your facts | Deduced |
|---|---|
| Not an instant, not a sorcery | It’s a permanent |
| A permanent that isn’t an artifact, creature, land, planeswalker or battle | It’s an enchantment |
| Power less than 6, power at least 5 | Its power is 5 |
| Not common, uncommon or rare | It was first printed at mythic rare |
| It makes Treasure | It makes tokens, ramps, and produces and fixes mana |
| It isn’t removal | It doesn’t destroy, exile or bounce anything, and isn’t a board wipe |
| It’s an Equipment | It’s an artifact |
Deductions use only the rules of Magic, never the list of possible answers. They can’t reveal which cards are in the pool. New positive facts are announced (“It’s red”); facts that rule things out quietly update What you know and your filters.
Numeric reasoning
Numbers are stored as ranges, not just “yes” or “no”:
Is its mana value 6? No, less than 6.
The game stores that as
Mana value < 6
rather than simply
Mana value ≠ 6
Each answer narrows the range, and when only one value is left, it’s announced for free. This works for mana value, power, toughness, loyalty, first print year, tap-ability cost, equip cost, colored mana symbols and generic mana. Comparisons work too: “Does it cost less than 3?” → “No, 3 or more”.
Name letters are the exception. “Does it start with D–M?” and “Does it start with a vowel?” get a plain Yes or No, because a letter range already narrows things down a lot. A No rules those letters out without saying which side the real one is on.
Semantic MTG concepts
Some questions aren’t simple card fields. Scryfall has no “removal” property, and players don’t always agree what counts. So GuessMTG gives each concept one written definition and applies it the same way to every card, reading the rules text without reminder text. If two players ask the same question about the same card, they get the same answer.
For example, “Is it removal?” is Yes for effects that destroy, exile, bounce or force the sacrifice of an opposing permanent, board wipes, and damage that can hit a creature. Pure combat buffs, pure life loss and effects that only touch your own permanents don’t count.
Interaction
Removal, destroy, exile, bounce, forced sacrifice, board wipes, counterspells, graveyard hate, discard, mill, shrinking stats (−N/−N).
Advantage
Card draw (not “whenever an opponent draws” triggers), card selection, tutors, and graveyard recursion of spells, creatures or lands.
Mana
Ramp (mana rocks and dorks, extra lands, Treasure; not lands or one-shot rituals), mana production, fixing, cost reduction.
Board
Tokens, creature tokens, +1/+1 counters, Equipment, Auras, putting lands onto the battlefield, evasion (flying, menace, can’t be blocked…), boosting stats (+N/+N), giving keywords to other creatures.
Protection
Hexproof, shroud, indestructible, protection and ward (if the card has or grants them), blink, phasing.
Other
Burn, damage, life gain, life loss, theft, clones, copying spells, sacrifice, sacrifice outlets, extra turns, alternative casting costs (Overload, Flashback, Evoke, “without paying its mana cost”…).
When does the game reveal information?
Direct facts
What you ask about. Is it red? No.
Derived facts
What your answers logically prove, announced for free as a deduction (see How deduction works). If your facts already settle a question you ask, the Oracle just tells you, and it doesn’t use a question.
Numeric facts
You narrow a range until the exact value is known, and then it’s announced.
Oracle hints
Hints are deliberate rewards, not logic. Miss enough times in one category and the Oracle gives that answer away. A miss is a question in that category answered No: “Is it 4 mana?” → “No, less than 4” is a mana value miss.
| Category | After | Revealed |
|---|---|---|
| Colors | 2 misses | A color: the whole answer for a single-colored or colorless card, or one of a multicolored card’s colors |
| Colors | 3 misses | All of a multicolored card’s colors |
| Mana value | 3 misses | The exact mana value |
| Card type | 3 misses | The full type line |
| Power | 3 misses | The exact power, or that it has none |
| Set | 3 misses | The set it was first printed in |
Each hint fires once, and not at all if you’ve already worked the answer out. A color miss is a guess at its colors (“Is it red?”, “Is it Simic?”, “Is it colorless?”); asking how many colors (“Is it multicolored?”) isn’t.
Good questions vs. less useful questions
Good questions
Questions that split the remaining possibilities roughly in half, whatever the answer:
Is it a creature?
Is its mana value 5 or greater?
Is it blue?
Is it removal?
Less useful questions
Questions about something rare, like an obscure keyword or a very specific word in the rules text, usually get a No that rules out only a handful of cards. They’re great for confirming a hunch late in the game, and costly early on.
Question strategy
Instead of asking “What do I want to know?”, ask yourself:
“How many possible cards does this question eliminate?”
That’s information gain. A question that splits the possibilities evenly halves them whichever way it goes, so a handful of good questions narrows thousands of cards down to a few.
- Search numbers. Ask about the middle of the range (“5 or greater?”), then the middle of what’s left.
- Let hints and deductions work for you. Two color misses reveal a color, and a few range questions can pin down an exact number for free.
- Watch the match count. What you know, the “cards match” count and the card list under the guess box show how much you’ve narrowed things down.
- Guess carefully. Your free guess after each answer costs nothing, but extra guesses count as questions.
What counts as a question?
- Compound questions count as one, and only the first part is answered.
- Statements such as “It’s red” are read as the question they imply.
- Follow-ups are read on their own. Each message is a complete question, so “What about green?” is read as “Is it green?”.
- Natural phrasing is fine. “Can it kill creatures?” and “Is it removal?” are the same check.
- Several characteristics in one question (“Is it a red creature?”): the Oracle answers one of them. Your history shows exactly which fact was checked.
- Guesses are free right after an answer. An extra guess, or one before your first question, counts as a question.
- Card names go in the guess box. The question box only asks questions, so “Is it Grizzly Bears?” is read as “Is it a Bear?”.
- Free: questions the Oracle doesn’t understand, questions you already know the answer to, deductions and hints.
How the card pool works
- Eligible cards are a hand-picked pool: popular, widely played cards (mostly Commander staples) mixed with lesser-known commanders and chaos cards, across every color, type and era. So the answer isn’t always a card everyone knows.
- The daily card is picked from a fixed, shuffled order of the pool, so everyone gets the same card on the same day.
- No repeats until every card in the pool has been used.
- The pool will grow over time. It’s kept hidden so the card list can’t be used to solve the puzzle.
- Card details come from Scryfall. Set, rarity and year refer to the card’s first printing (Rhystic Study counts as common, from Prophecy in 2000), and a promo first printing counts as its main set (Esper Sentinel is from Modern Horizons 2). A two-faced card is answered as its front face. Hybrid and Phyrexian symbols count toward each of their colors, and “The One Ring” starts with T.
- Format legality, Game Changers and reprints are a snapshot that’s refreshed now and then, so a very recent ban, Standard rotation, Game Changers update or reprint can take a little while to show up.
- Practice games (the replay button, or “Play again with a random card”) draw from a separate, rotating practice set, not the daily cards, so practicing never hints at what the daily card can be.
The daily game
- One game per day, and everyone gets the same mystery card. It resets at midnight in your time zone.
- 20 questions. Your score is how many you used; fewer is better.
- Guessing: one free guess after each answer, and extra guesses count as questions. The game ends when you name the card, or when the guess after your 20th question misses.
- Stats and streaks: games played, win rate, current and longest streak, and average questions. A streak is a run of consecutive daily puzzles won.
- Spoiler-free sharing: 🟩 yes (confirmed true), 🟦 no (confirmed false), ✨ a no that earned a free deduction, ⬜ a wrong guess that used a question, 🎯 the winning guess. It never shows your questions or the card.
- Accounts are optional. Progress is saved in your browser; an account (email, Google or Discord) syncs it across devices and keeps your streak.
- Anonymous totals: each finished game is counted once for site-wide statistics, using a random ID for your browser plus the puzzle, whether you won, and how many questions and guesses you used. Each answered question is also counted by kind (say “Color: Red”, and whether it was a yes), and questions the Oracle can’t understand are kept so it can learn them. Nothing identifies you. See how everyone is doing on the statistics page.
- Question requests: signed-in players can star kinds of question the Oracle can’t answer yet on the requests page. A star is stored with your account so you can take it back; everyone else sees only the totals.
Why the Oracle doesn’t just tell you everything
The Oracle is designed to
answer what you asked while giving you only the information that question logically provides.
It knows the card, but it won’t volunteer unrelated facts. The puzzle is in choosing what to ask. What it will do is point out anything that has become logically certain from your own answers, since you’ve effectively earned that already. Oracle hints are the one deliberate exception, a small push when you’ve missed the same kind of question a few times.