My strategy is to never use the same start word, I like using a random word each time as that changes the path to solving. I might pick a random word from another open browser window, maybe from a news article or something like that. If I get stuck after the second or third try, I'll pick another random word, even if it contains a few discarded letters in order to ferret out 1 or 2 more valid letters.
For me, the best part of Wordle is not in finding the answer but in the steps prior as I work out the path to solving the correct answer.
I'm with you, I try and use a different word whilst keeping it plausibly common, it's part of the cerebral challenge. Using an algorithmic approach is efficient but not "fun". Target is always 3.
Another vote for this strategy to keep the game interesting.
Putting the same words in at the start got boring after the third game, and the first word dictates the on-the-fly logic to select the next, and so on.
It all comes down to how you want to play. Using maths optimises "winning", but not using maths challenges ones situational intellect. I prefer the latter.
This is what I've done from the start. Always pick a random word as the first entry (or I guess "random" since I'm still ensuring that there's no duplicate letters). It helps keep the game from getting stale.
Using the "same words every time" (i.e., a fixed set of initial guesses) has been studied before [1,2,3]. It is one of the few remaining open questions about Wordle. Interestingly, it can be done in 6+1 guesses [4] (COMBO FATTY GRRRL SPUDS VENGE WHILK, then the possibilities are always narrowed down to a singleton). However, it is unknown whether it is possible within 5+1 guesses, which would make Wordle 100% solvable even with such a constrained approach!
It is unfortunate that 3Blue1Brown's excellent video has been so often misquoted as providing "optimal" guesses for Wordle. Of course, one can legitimately argue that using maths takes the fun out of the game... but if we *are* going to use maths, then the information entropy approach is simply not the one most suited to this specific game (because the dictionary is fully known).
I'll also take this opportunity to plug Semantle, which is an infuriating game that I can't resist. It's based on word2vec. You get infinite guesses, and you're given feedback on how "semantically similar" each guess is. It often takes me 100+ guesses.
Yikes that was hard. To explain to others, the Similarity scores range from -100 to 100, and your goal is to get a similarity score of 100. It says the tenth-nearest has a score of 48-50, that’s a good guide for when you actually have a similar word or not. If you haven’t hit a similarity of at least 50, you’re still close but not there yet. The hints given might not be any good until you hit a similarity in the 40s. Also don’t let guesses with lower similarity scores distract you, the word tends to be a part of, synonym or related word by the time you get to a similarity score in the 50s…
I got it in 33 on my first play today but my guesses were way off and it was saying things were highly similar that ultimately threw me off the course. I could see that being a lot of fun to train yourself to think along semantic lines like that.
I used to start with ADIEU because it contains 4 vowels and the most common ones at that. But I've since moved on from that because I found I would be forced to reuse vowels in words 3 or 4.
I'm vaguely tempted to take the dictionary and figure out an optimal strategy that 1) guarantees success and 2) minimizes the number of guesses. If you assume every word is equally possible (it isn't; they're manually chosen) there would be an optimal starting word. I'm sure others have looked into this but this is something I'd like to do myself.
After that it branches depending on what hits you get on that first word. It may be possible that a second fixed word (or a small set of second words easily memorized) would be near-optimal but not actually optimal. I'd be curious to know this too.
But I'm curious how good an optimal strategy would be vs some of the naive strategies we've all chosen.
It's also an interesting question as to when it's worth switching from finding what letters are in the word vs locking down their position. If you get COAST and AT are in the word but in the wrong position, should your next attempt be 5 new letters or a word containing AT in different positions? The disadvantage of this of course is you're only testing 3 new letters.
I also toy around with Quordle where you have 9 guesses to find 4 words. That one's harder and the strategy is a little different. There I've pretty much settled on finding a set of 3 words that covers all vowels (and Y) and 9 of the most common consonants.
I've tried going both ways with vowels (lots or sparingly), and I found it seems to work better to conserve them. Vowels are where you can maximize information by inference rather than by spending your limited slots, by downplaying or ruling out more vowels once you've got what are likely enough.
I always go 5 new letters for the second guess and usually for the third. Knowing more letters are in or out is a lot more information than finding out the same letter isn't in a second position. Particularly for Quordle, you really want more letters through three or even four guesses.
> I also toy around with Quordle where you have 9 guesses to find 4 words. That one's harder and the strategy is a little different. There I've pretty much settled on finding a set of 3 words that covers all vowels (and Y) and 9 of the most common consonants.
Yes, in Quordle you're compelled to use a standard opening. Regardless of what you learn from early guesses, no followup can be more informative than just querying more letters.
It's kind of disappointing to me how much luck is involved in Quordle. I've just established that the only common letters in a word are S, A, and Y, with A second and Y fifth. All three other words are already guessed. Do I guess SASSY or SAVVY? The game gives the appearance of being much more skill-based than it actually is.
There are lots of words like this and it affects Wordle too. Take SHA_E. You have SHAME, SHARE, SHAVE, SHALE, SHAKE and SHAPE. It's highly likely your earlier guesses haven't eliminated all of these. So what you have to do is recognize this situation and throw in a word with all of the possible missing letters to tell you which one is the right one. Obviously you need 2 guesses left at least for this to work.
But I've found the same applied with Quordle. Recently I had _LARE and even that allows FLARE, GLARE and BLARE. It turns out solving another word (BRINE) eliminated one of the options for free.
I've noticed whoever chooses the word of the day for Worlde likes these types of words. Like I've ended up with S_ILL before leaving SPILL, SKILL, STILL and SHILL.
> I've noticed whoever chooses the word of the day for Worlde likes these types of words.
Josh Wardle said on a podcast[1] that him and his partner chose the set of words by taking all 5 letter words in the dictionary and filtering out the "bogus" ones. He then randomized the order of these words so that him and his partner could play. However, I'm not sure if the New York Times has changed how word selection works.
> So what you have to do is recognize this situation and throw in a word with all of the possible missing letters to tell you which one is the right one. Obviously you need 2 guesses left at least for this to work.
This advice applies much better to Wordle than it does to Quordle, though. Quordle gives you nine guesses. Four of them are already taken by the need to provide four answers. Three or four go to your fixed opener. That leaves you either one or two discretionary guesses. If your opener ended up giving you trouble with more than one word, you're basically hosed.
I play a bit of practice Quordle with a best streak of 97. I start TREAD and SONIC*, then usually HULKY or sometimes BULKY, LYMPH or PLUMB. Whether I use HULKY third usually depends on the prevalence of Ts or Cs. If I'm confident of a word after the first 1-2 lines, I'll obviously go with that as it's a free hit in uncovering more letters or position clues for the other words.
My 9yo loves Quordle also but usually starts TREAD and MINOS.
Obviously anyone can play however they like and how they enjoy the game. But I feel like this "enter three optimized words and win" is too easy. So instead I go with hard mode, and have to think some more. I still start with the same word every time, but from there on most parts vary. I also allow myself to fall back to normal mode if finding the word depends on pure luck (e.g. _ATCH).
Bonus: If you and your friends play hard mode, you can try to do a reverse wordle on their solutions.
My strategy on hard mode is to start with words containing uncommon letters, using new starting words each time to keep it fun. That way I find the "*atch" problem is reduced because (hopefully) the number of possible options has already been narrowed down.
My histogram is mostly 4's, with 2 losses in around 100 games, so I think I'm doing something right.
Since switching to hard mode, I almost never start with “good words”. Too much risk of getting the 10 possibilities with 3 guesses remaining type situation.
> In hard mode, Wordle can be solved in 3.5076 guesses on average (with 6 guesses at worst, i.e. 100% of the time). Or, with a different decision tree, it can be solved with a slightly worse average, but always within 5 guesses
"Cheating" is subjective, but you are correct. To put it in more neutral terms: Using the known 2309-word list of solutions, hard mode can be solved 100% of the time within 6 guesses. Not using it, you would need 7 guesses to guarantee that you solve it every time [1].
The subset of words that are solutions is an internal data structure of the game. It is clearly cheating to refer to that list. If you are willing to examine the inside of the game's black box, the optimal strategy is to simply extract today's answer.
As a human, you already have knowledge which is similar to knowing the internal dictionary: you can judge what other kinds of words Wordle is likely to avoid (uncommon words, proper nouns, plurals). So I don't think it is cheating.
Thing is, you can develop a very strong intuition for which words are in that list: they're all common words and mostly "main dictionary entries" (you probably won't find a 4 letter noun or verb with "s" tacked onto the end, probably not "ed" or "es" either)
Certainly using the exact list is a step beyond that; but you can in fact get very accurate at guessing which words might be on it without memorizing the list
In the limit, if there was one day of Wordle left and you looked at what it was going to be and then told me that the optimal strategy of which words to guess is to just guess the solution as the first guess, that's cheating. Your "strategy" becomes a question of how well can you compress the list of solutions into the entropy allowed by how you chose to express your strategy, in this case "four 5-letter words" or somewhere between 20 and 94 bits of entropy[0].
[0] 4*5 assuming that 1 letter is one bit of entropy, which is a rule of thumb when you're compressing gigabytes of text to log2(26^(4*5)) if we just count the combinations of letters
Sorry, can you explain what you mean here: I play hard mode and I always get the word within 6 guesses — and I’m no savant, I’m an idiot, and I don’t cheat — no looking at the word list. The same is true for my friends who play on hard mode: no cheating, always finish. Why is luck required for hard mode?
I lost this week on FOYER without repeated letters, there are plenty of words with ER endings and O as a second letter.
—-
VOICE
LONER
TOWER
JOKER
POSER
HOMER
——-
The last two were unlikely to be the word, but I try to finish in only a few minutes and those popped to my head before foyer. But I could have guessed BOXER or DOPER or some other words I’m sure. When I’ve lost it’s usually because I have 3 letters and guessing only 2 new letters at a time doesn’t let you eliminate the other options fast enough.
Our wordle slack group lots of people lost on foyer.
I had started with some random word with lots of consonants (all black result), so when I got to having just o e and r, I only had two clear possibilities I could see left with three guesses.
Wordle works with a 12k-word dictionary, which is very much comprehensive (that includes "words" like "grrrl"). However, the hidden/secret words are picked from a much smaller set of 2k reasonable words (i.e., frequent ones, and that most people would describe as English words).
There are two reasons that you always win within 6 guesses:
1. It has been shown [1] that in hard mode you can always solve wordle in 6 guesses (but not always in 5) if you assume that the hidden word is "reasonable", i.e., taken from the 2k-word list. However, if you know 12k 5-letter English words and if you don't assume that the hidden word is "reasonable", then you will sometimes need 7 guesses.
2. Even then, the average number of guesses that you need is much lower, at 3.5 guesses (or 4.5 using all 12k words). So if you play optimally or close to it, it is only in very rare cases (the worst case) that you will need the full 6 or 7 guesses.
So, surprisingly, the game is easier if you are not a "savant", or to be more precise, if you are not a computer :-).
That's luck, which is fine. But there is not an optimal strategy which if followed always result in a win for the actual Wordle games. Some of them require at least 7 guesses in hard mode.
It’s true you can get into a hole, but there are strategies that help, like the one the person you’re replying to mentioned. It also helps to get a sense for what kinds of words wordle uses (e.g. no plurals). That being said, it’s the possibility of losing that makes it exciting!
I’ve played 113 times on hard mode and only lost once so far, guessing between two possibilities on the sixth word.
For a while now I’ve just been starting with the previous day’s word to mix things up, rather than trying to use the most optimal starters.
Same. I start with QUARK every time. It’s mostly to one day achieve that elusive 1-guess solution but it often ends up making my life easier by avoiding pigeonholing myself on hard mode.
Huh. Am I alone in thinking that hard mode (must reuse previous letters) should be the only option? Being able to choose a completely different word to fill in the gaps seems overly easy... (There's still luck, but it's vastly easier).
I hadn't seen anyone else use that pair when I suggested it in early Jan https://twitter.com/bazzargh/status/1477998502747779079 ... but I guess there aren't many choices with those letters and other people were bound to hit on those words! Seems quite popular now.
For octordle/sedecordle I found that the 2 standard guesses wasn't enough to stop getting stuck, particularly it's missing Y for words that use that as a vowel. For those puzzles, I use INPUT SHAME GLORY, but I've not attempted to check if this is anything like optimal - I very rarely fail those puzzles with these 3 tho.
I don't regularly play octordle but came to the conclusion that you really want at least 3 start words. Otherwise, you start trying to solve with guesses that aren't globally optimum and run out of guesses.
STARE DOING CHUMP is usually enough for me to have a good stab at Quordle. For regular Wordle I try to stick to "hard mode", with the constraint that discovered letters have to be used.
I have settled on DREAM PILOT (I often don't use pilot if I get 2 or more yellow/green) with no reason other than I'm pig headed and refuse strategies others suggest.
I've also been playin on hard mode almost since the beginning. While it usually does make the game more interesting, it also annoyingly often makes your win depend on pure luck. Ironically this happens more often when your first guesses were very good.
Imagine having SHA?E confirmed and now being forced to stupidly guess between SHAPE, SHAKE, SHARE, SHAME, SHAVE or SHADE instead of eliminating half of those possibilities with a single word.
So maybe this implementation of a hard mode isn't well suited anyway if you're looking for a bigger challenge in skill.
I found that only happens if you try to hit the most common letters in your first guesses. If you use middle of the road letters you tend to get only a couple while eliminating a lot - s and e would be bad early letters because they don’t reduce enough words
Opinions vary based on the input text, but the English letter frequency list I use starts ETAONRISH in descending frequency.
Based on that, coupled with a little bit of insight on English words (as opposed to random collections of 5 letters) I've ended up with ALIEN, STORM, CHUMP as my regular first 3. Sometimes I deviate if the earlier rows offer insight, and sometimes I swap the first 2 based on a whim.
And unless it is obviously worth a try, I use my first 3 words to eliminate letters and not to guess.
I usually get it in 4, sometimes in 5, almost never needing 6, and have failed once. So not necessarily the best, but good enough.
For the variants I've programmed, I've enforced that they start with a random word guessed (same random word for everyone though.)
Wordle is good as it is but personally I think most of the variants would be better off adopting this feature.
That said I saw someone speedrunning 10x Xordle games and for speedrunning they just ignore the starting word and make the same 4 guesses every time: CARVE SHIFT GODLY WHUMP
Wordle sparked off sort of a global game jam around variants, maybe because it was such a wholesome project. And these variants have found players; there's a lot of people who play a whole bunch of daily word games, there's some youtubers, etc.
I've programmed 3; the way that went down is I was curious if an idea I had would be fun (two Wordles on the same board, sort of a mix between Wordle and a word jumble since you don't know which word the clues go to.) Since there was an open source implementation this only took me one night, and it was in fact fun and started gaining some traction as https://xordle.xyz
Then some game designer friends suggested their own takes and I implemented those too: https://fibble.xyz lies to you once per row, and https://warmle.org tells you if you're close to the right letter rather than if a letter is in the wrong spot which gives the gameplay a very different feel
This is what I do, just use some word. Don't care so much about the score, just to get it within the allowed guesses. Would be boring to make a recipe.
I thought it was in that linked video or somewhere that I found a list of starting words to try to make it easier to solve it based on letter frequency and getting a few vowels in within the scope of the Wordle word list (not a dictionary list).
These are the words I use:
SOARE
TREAD (or TRADE, which is an anagram)
ADIEU
By default my game play is like “hard mode” (only use letters already found to be correct and avoid letters found to be incorrect). That helps most of the time.
One of the annoying (or challenging) things about Wordle is that its word list has many sets of words that differ by just one letter. For example, you may get _ATCH right and then have to really guess if it’s going to be CATCH or BATCH or PATCH or MATCH or WATCH or LATCH or HATCH (maybe there are more words with a different first letter in this range). I’ve seen several sets of words like this. Just can’t do this within six total guesses. So chance does matter.
I thought this too at first, but this is where the real strategy comes out, because you can guess MULCH and that will tell you if it's LATCH or MATCH. or CLIMB, that would also eliminate BATCH. There's intelligent ways to guess to eliminate more letters. Hard mode is only hard because it eliminates this possibility of constraint narrowing.
I think Hard Mode is not a good name for it. The game is already plenty hard.
My wife and I don't care about the end result in those situations since the fun has been removed at that point. We just iterate through until we're right or have run through our chances.
I've settled on the following word list to determine if 20 of 26 characters are present within 4 guesses:
thank
fuels
crowd
gimpy
If there are fewer than 5 letters present, "bevvy" will eliminate another two ('b' and 'v'), leaving j, q, x, and z as the remaining possible characters. These occur rarely (though "pizza", "bijou", and "vixen" can be challenging targets. More usually, it's words with either doubled characters, such as "onion" or "lalai", or those which have viable anagrams ("spams" and "spasm", "donor" and "rondo") that will throw me.
The thing is that this doesn't help you with double letters, still leaves out some less common letters, and leaves you with very little margin for error at that point.
On Octordle, 4 words seem easiest although it makes the game more of an anagram solving exercise and in my limited experience didn't give me much of a consistent score advantage vs. three.
I've used that strategy on 100s of standard Wordle games to good effect. It's not perfect, but it solves ~98-99% of all games attempted.
(That stat comes from one of the multi-play Wordle imitations which does in fact track success rates.)
I don't recall mean attempts, though I believe it's still between 3-4 tries.
Once most characters are tested, it's usually pretty clear what the actual word is, and some familiarity with the actual wordlist (which I picked up merely by playing the game, not by examining source), the harder cases also become pretty obvious.
Doubled letters and anagrams are the most difficult cases.
I usually employ two entry words that employ the vowels, sometimes the second word isn’t necessary, but often I’ll just throw it in - it kinda makes the first two lines of play redundant as a game, but I can’t really think of a better way to solve!
I always start, very ego-centrically, with my last name (matching my HN account name), and then usually follow up with BEADS and JUICY to cover all the vowels.
But I'll try guess the target word if the first or first two words give me >= 3 hits.
Then I'll spend WAY more time trying to solve the 5 daily chess puzzles at https://www.chess.com/puzzles/rated ...
I actually made a Wordle variant me and my family have been playing that prevents you using a word you have ever used before (unless it is that day's word). It's a really small tweak and it takes a while to have any impact, but it makes the game so much more interesting for me. (jezzamon.com/wordle if anyone wants to try it)
No idea if it'll become unplayable eventually but I'm having fun seeing it play out as more and more common words get locked out
I start with two words that use all the vowels plus Y, and the most common consonants TNLR etc. e.g. AUDIO ENTRY then to save having to mentally step through all the possibilities (the least fun part of wordle) I use a word finder that takes the letters in the word, the letters not in the word, and any green letters in the right place to produce a list of the remaining possibilities.
Originally I used BLACK, WHITE, SOUND, and tried to guess from there, occasionally throwing in GRUMP if the word wasn't obvious. I didn't like the repeated vowel, and knowing if there was a Y in the word would help, so I switched to:
AMUCK
FETID
SWORN
GLYPH
100% so far. One or two of those may have involved aggressively grepping /usr/share/dict/words, though.
I've used STEAL, ROUND for almost the entire ~80 games I've played, and I'm still 100% so far, no assistance or grepping. (Notably very close to the author's first two guesses he came up with without assistance)
I have a go-to third guess when needed, I just can't think of it right now.
Having a set of words is important when trying to complete as fast as possible. This pretty much sets up the game into 2 phases, one where you just brute force typing from memory and another where you solve the puzzle at once. If it's too hard sometimes it's better to just reset and start a new puzzle.
The sequence has the upside of containing every vowel, without repeating a single character. Downside is it does not contain some pretty common consonants, I'm gonna have to try some of the sequences suggested in the post to compare what seems better now
There are 28 letters. With 3 initial same words you can cover up to 53.5% of the alphabet. With 4 initial same words you can cover up to 71.4%. There are only 5 vowels A, E, I, O and U which can be covered with the first two words. You can even add Y in the first two words played.
But I don't think it is a good idea for the average person to have too many set words. Unless you have great knowledge of the dictionary of legal words you might need extra clues to word order etc. ie I recently got a game with a letter repeated 3 times.
Spoiler, stop reading if you started playing worldle in the past 3 months.
I used could a lot until it became one of my o my 2-guess wins. Now I hate using it knowing there is zero chance it’s the answer. I still do occasionally if someone after 3 guesses all 5 of those letters are available, which is almost never.
For octordle I like CORNY and ADIEU for my first two guesses. Gets all the vowels and enough to start. For wordle, starting from STERN or a few other words like that is fine.
Josh Wardle is a cultural icon. He's really done an amazing thing.
An average of about 3.8 is not hard to achieve. Much harder is to reduce the number of failures, (i.e., requiring more than 6 guesses). At least in hard mode, that's true. I'm experimenting with easy mode now.
Easy mode is different. The goal is essentially to get as many 3s as possible while minimizing number of 5s. With any skill, actually losing will be almost impossible
CRANE first. If the A and E are not both present and there are no green consonants,
HOIST, else figure out a word that tests common di- or triglyphs (ch, cri) based on known letters that also contains O.
For me, the best part of Wordle is not in finding the answer but in the steps prior as I work out the path to solving the correct answer.