How to Master Pairdle, Column by Column

Most Pairdle advice treats the three columns as three copies of the same problem. Look at the pool, find a word, place your pairs. That's how I described it myself for a long time.

Then I actually counted, and it turns out the columns aren't remotely alike. One of them is close to a coin flip between five options. Another is a genuine free-for-all with three hundred possibilities. Knowing which is which changes the order you should solve them in — and it's the single biggest improvement available to most players.

So here is every number, column by column, for both Pairdle and Pairdle Pro. Then the playbook I actually use.

How I measured, and the honest caveats

I took every Pairdle puzzle we have on the schedule — 3,372 of them, three difficulties a day across the whole calendar — and sliced each secret word into its letter pairs. A six-letter word gives you three pairs: characters one and two go in FIRST, three and four in MID, five and six in LAST. Then I counted how often each pair landed in each column. Pairdle Pro got the same treatment across its 3,246 puzzles, where eight-letter words make four columns instead of three.

Four caveats you should carry while reading:

These are answer tiles only. Your board shows 24 tiles, but only three of them are correct. The other 21 are slices of seven decoy words. I've counted nothing but the pairs that actually belonged to the secret word, because "what is most likely correct here" is the question worth answering. Decoys are noise by design.

This is closer to a census than a sample. Pairdle draws its answers from a pool of common six-letter words, and that pool is about 3,059 words wide. Our 3,372 puzzles use 3,038 distinct ones. That's 99.3% of everything the game is capable of asking you. So these percentages aren't "what happened to come up recently" — they're very nearly the complete shape of the game. Pro covers 99.4% of its own pool. That's what makes the "never" claims further down worth printing.

A few answers repeat. There are more puzzle slots than there are eligible words, so 317 words appear more than once across the calendar. That could in principle skew the counts toward whatever repeated. I checked by running every number twice — once per puzzle, once per distinct word — and the two never disagreed by more than half a percentage point. The tables below are per-puzzle, which is what you actually encounter.

The board you see can differ very slightly. When two of the eight words would put the same pair on the board twice, the game swaps the surplus copy for a filler pair so no two tiles ever show identical letters. That's under one tile per board on average, and it never touches the answer tiles. It doesn't move any number here, but it's why your board occasionally contains a pair that belongs to no word at all.

Pairdle's LAST column: where the game is won

Start here, because this is the finding that matters.

The top five LAST pairs account for 35.4% of every Pairdle answer ever scheduled. Not the top twenty — the top five. Widen to twenty and you're at 61.5%. Nearly two-thirds of all Pairdle endings come from a list you could memorise over one cup of coffee.

Rank	Pair	Share of all LAST columns
1	ED	13.0%
2	ER	8.5%
3	ES	6.6%
4	NG	4.2%
5	TS	3.2%

The next fifteen, in order: LE 2.9%, LY 2.6%, RS 2.6%, EN 1.8%, ON 1.8%, AL 1.8%, LS 1.7%, ET 1.7%, DS 1.5%, KS 1.5%, NS 1.3%, NT 1.3%, ST 1.2%, TY 1.2%, RE 1.2%.

ED alone is one in eight. If you learn one number from this entire article, learn that one.

The reason is grammar, not vocabulary. English builds its past tense, its agent nouns, its plurals and its gerunds by bolting the same handful of suffixes onto the end of words — -ed, -er, -es, -ing, -s. Six-letter words are long enough to be inflected forms of shorter words, and short enough that the inflection eats the entire final pair. WALKED, FASTER, BOXES, TALKING, EVENTS. The last column isn't really testing your vocabulary. It's testing whether you noticed that English has about six ways to end a word.

LAST is also the narrowest column: only 176 distinct pairs have ever appeared there, against 197 in FIRST and 308 in MID. Fewer options, and the options that exist are wildly top-heavy. It is the most predictable real estate on the board by a distance.

Pairdle's FIRST column: broad, but with favourites

FIRST is a middle case. The top five take 11.8% and the top twenty take 32.4% — real signal, nothing like LAST's dominance.

Rank	Pair	Share of all FIRST columns
1	RE	2.9%
2	ST	2.8%
3	CO	2.3%
4	CA	1.9%
5	DE	1.9%

Then: SH 1.8%, PA 1.7%, MA 1.6%, SP 1.4%, MO 1.4%, BA 1.4%, BR 1.4%, SE 1.3%, CH 1.3%, LA 1.2%, RA 1.2%, TR 1.2%, IN 1.2%, BO 1.2%, CL 1.2%.

Two families do the work here. There are the prefixes — RE, DE, CO, IN are the front ends of REPEAT, DEMAND, COMBAT, INTENT — and there are the consonant openers, ST, SH, SP, BR, CH, TR, CL, which are simply the blends English is willing to start a word with. Between them they cover a third of all openings.

What makes FIRST feel harder than LAST is that the distribution is flat where it matters. RE at 2.9% is the single most likely opening in the game, and it is still wrong 97 times out of 100. You can't guess your way into a first column. You can only narrow it.

Pairdle's MID column: the one you should never guess into

MID is the flattest column in either game. Top five: 8.3%. Top twenty: 25.0%. And it holds 308 distinct pairs — more than either neighbour, out of a smaller total, which tells you it sprays.

Rank	Pair	Share of all MID columns
1	RI	2.0%
2	CK	1.7%
3	AR	1.6%
4	LL	1.6%
5	EA	1.5%

Then: ST 1.5%, TI 1.4%, TT 1.3%, IL 1.2%, ND 1.2%, RA 1.2%, RE 1.2%, AN 1.1%, IN 1.0%, LI 1.0%, EE 0.9%, EN 0.9%, RO 0.9%, VI 0.9%, NG 0.9%.

Look at the shape of that list. Three-quarters of all MID columns come from outside the top twenty. The middle of a word is where English keeps its actual variety — the doubled consonants (LL, TT, CK, EE), the vowel clusters, the fragments of stems that no suffix ever touches. There's no grammar shortcut here, because the middle of a word isn't doing grammar.

The practical consequence: MID should be the last column you commit to, and you should almost never guess it cold. Treat it as the residual. Once FIRST and LAST are locked, the middle is usually forced — there simply aren't many real words that fit ST _ _ ED. Working the other way round, trying to guess MID and build outward, is playing the hardest column first with the least information.

What never shows up where

Because the corpus covers 99%+ of what Pairdle can ask, "this pair has never appeared in this column" is a much stronger statement than it would normally be. But absence isn't the same as impossibility, so I split the claims into two honest tiers.

Genuinely impossible — no six-letter word in the entire dictionary puts these pairs here, common or obscure:

• Never FIRST: CK, DS, KS, LS, NS, NT, RS, TT. No English word begins with any of them.
• Never LAST: BR, CL, TR, VI. Nothing of this length ends that way.

If you see CK on your board, it is not a first column. Ever. That's a free elimination, and CK shows up often — it's the second most common MID pair in the game.

Impossible in Pairdle, but only because of our word list. These do exist in English somewhere, just not among the common words Pairdle draws from. I'm naming the counterexamples rather than hiding them, because "never" claims deserve their own disproof attempt:

• Never FIRST: LL, NG, TS, IL, ND — LLAMAS and TSETSE are real words, they're just outside the pool.
• Never LAST: BA, BO, CA, LI, MO, PA, SP, TI, TT — AMOEBA, ALPACA and GRAPPA all end this way and none of them are in Pairdle's dictionary.
• Never MID: DS, KS, LY, TY.

Notice what's missing from that last line. MID has no hard exclusions at all among the game's common pairs — not one. Every frequent pair in Pairdle can legally sit in the middle of a word. It's the permissive column in every sense: widest, flattest, and the one that rules out the least.

And three pairs in the top twenty live in exactly one column and nowhere else: DS and KS are LAST-only, TT is MID-only. When one of those turns up, you have a free placement.

Pairdle Pro: same story, louder

Pro's eight-letter words give you four columns, and the pattern doesn't just hold — it intensifies.

Column	Distinct pairs	Top 5 share	Top 20 share
FIRST	179	18.1%	40.6%
SECOND	309	8.3%	25.3%
THIRD	273	16.6%	37.4%
LAST	143	43.1%	72.6%

Pro's final column is the most predictable thing in either game. The top five take 43.1%, the top twenty take 72.6%, and it draws on only 143 distinct pairs — the narrowest column anywhere.

LAST: ED 14.4%, NG 14.4%, ES 7.4%, ER 3.5%, RS 3.4%. Then TS 3.1%, AL 3.0%, LY 2.5%, ON 2.5%, TE 2.3%, NT 2.2%, IC 2.0%, TY 1.7%, NS 1.7%, SS 1.7%, LE 1.5%, ST 1.4%, RY 1.3%, GS 1.2%, NE 1.2%.

ED and NG are dead level at 14.4% each — between them, more than a quarter of all Pro answers end in one of two pairs. Eight-letter words are overwhelmingly inflected forms, and -ing gets much more common as words get longer.

FIRST: RE 5.4%, CO 4.3%, PR 3.0%, IN 2.8%, ST 2.6%. Then DE 2.6%, DI 2.2%, PA 1.7%, MA 1.6%, TR 1.4%, CA 1.4%, CH 1.4%, SE 1.4%, SU 1.4%, BA 1.3%, EX 1.3%, SH 1.3%, EN 1.2%, SP 1.2%, GR 1.2%.

Pro's opening is noticeably sharper than Pairdle's — 18.1% against 11.8% — because long words lean much harder on Latin prefixes. RE, CO, PR, IN, DE, DI, EX, SU are eight of the top seventeen.

SECOND: ST 2.1%, EA 1.7%, TI 1.6%, LI 1.5%, ER 1.4%. Then AR 1.4%, ND 1.3%, TE 1.3%, RE 1.3%, VE 1.2%, IN 1.1%, LL 1.1%, RA 1.1%, RI 1.1%, ME 1.0%, RR 1.0%, AN 1.0%, MP 1.0%, AC 1.0%, DI 1.0%.

THIRD: TI 5.7%, LI 3.3%, RI 2.6%, IN 2.6%, ER 2.3%. Then EN 1.9%, DI 1.8%, NI 1.7%, SI 1.7%, AT 1.5%, RA 1.5%, CH 1.3%, ST 1.3%, LE 1.2%, MI 1.2%, KI 1.2%, PI 1.2%, UR 1.2%, RE 1.1%, TE 1.1%.

Pro's third column has a quirk worth exploiting: look at how many of those pairs end in I. TI, LI, RI, NI, SI, MI, KI, PI — eight of the top twenty. That's the -tion, -ical, -ity, -ing machinery lining up, and it means an I-ending pair is a strong third-column candidate in a way it never is in Pairdle.

SECOND is Pro's flat column — 309 distinct pairs, the same 8.3% top-five share as Pairdle's MID. Same rule applies: solve it last.

Three Pro pairs are strictly LAST-only across all four columns: TY, RY and GS — if one of those is on your board, its home is already decided. Meanwhile NT, NS, RS, SS, LL, RR, MP and ND can never start an eight-letter word at all.

The playbook

Here's how I actually play, now that I've seen the numbers.

1. Solve the ending first. This is the whole point. The last column has the highest concentration and the fewest options in both games, so it's where a guess buys the most. Scan the pool for ED, ER, ES, NG, TS — in Pro, ED and NG above all — and ask whether the word could plausibly be a past tense, a plural, or an -ing form. It usually can.

2. Then the front. With the ending fixed you've usually collapsed the word to a shape. Now work FIRST, where prefixes and consonant blends give you real structure to lean on.

3. Let the middle fall out. Don't try to win MID. Once both ends are locked there are rarely many words left, and the middle is usually forced. That doesn't mean the middle slots are useless — see step 6, where they turn out to be the best probes you have.

4. Give yourself about a minute to hunt for a real word — then stop looking. This is the discipline that took me longest to learn. If you've stared at the pool for a minute and no actual word has surfaced, more staring won't produce one. What it will produce is a guess built on a hunch, which teaches you nothing when it comes back grey. Do put the minute in first, though — word-spotting is a trainable skill, and the more you play the faster real words jump out of the pool at you. That minute gets shorter and more productive with practice in a way that no amount of frequency-table memorising does.

5. If you do see a real word, guess it. The odds are much better than they look. Here's something about the board that isn't obvious from playing: every Pairdle board is assembled from eight real words, and one of them is the answer. The other seven are there precisely to donate plausible-looking pairs. So a genuine word you've managed to build out of the pool has roughly a one-in-eight shot at being today's solution outright — assuming it's one of the eight rather than an accidental crossing of pairs borrowed from three different words, which does happen. Those are good odds for a puzzle that gives you six guesses, and the downside is mild: a wrong real word still colours all three of its pairs. If you see DEFEND sitting in the pool, play DEFEND.

6. Spend the middle slots on information, not on the middle. In standard and challenge you place two pairs in every column — effectively two candidate words per guess. That makes the middle slots the cheapest real estate on the board, because MID is the column where a guess is worth least: even the top five only land 8.3% of the time. Trying to hit MID with them is close to throwing them away.

So aim them somewhere else. Deliberately drop a suspected ending pair into a middle slot. If it comes back orange, that pair is in the word but not where you put it — and the conditional odds here are brutal in your favour. When ED appears anywhere in a Pairdle answer, it is the ending 97.3% of the time. TS is 99.1%, ES 95.3%, ER 91.3%, NG 82.0%. So an orange ED out of a middle slot doesn't just tell you ED is in the word — it has all but handed you the last column. You converted your least valuable slot into a confirmed pair and a confirmed position.

The same move works from the front, but less decisively, so know which openers are worth it. CO is a first-column pair 88.6% of the times it appears — a strong probe. RE is only 55.4%, because REPEAT and BARREN both exist, so an orange RE genuinely leaves you guessing. Probe with the lopsided pairs, not merely the common ones.

7. When no word appears at all, fill the slots with the most frequent pairs for each column and submit that. This feels like giving up. It isn't — it's buying information. A guess made of high-frequency pairs is a guess where every tile has a real chance of being right, and Pairdle's feedback tells you three separate things per tile: blue means right pair, right column; orange means the pair is in the word but you've put it in the wrong column; grey means it's out entirely. An orange is worth as much as a blue, because it hands you a confirmed pair and eliminates a column at the same time.

That's the trade. A hunch guess might solve it and probably tells you nothing. A frequency guess rarely solves it and always tells you something. Over a full week of puzzles, the frequency guess wins comfortably — and it's the reason a "wasted" guess in Pairdle is almost never actually wasted.

The columns aren't equal. Play them in the order their odds deserve, and Pairdle gets a great deal easier.

---

Want the general-English version of these tables? The most common letter pairs in English covers the dictionary-wide picture. For tactics rather than numbers, try 5 strategies to get better at Pairdle. And for the days when none of this saves you, the 10 hardest Pairdle puzzles according to player data. Or just go play today's puzzle.

By Tim Nye, Logic Loft Games