DEV Community: RAMKRISHNA PRASAD

I Analyzed 1,994 Wordle Answers - Here Is What the Data Says

RAMKRISHNA PRASAD — Tue, 20 Oct 2026 21:16:00 +0000

I pulled every Wordle answer published from 19 June 2021 to 5 December 2026 — 1,994 of them — and ran some analysis on the dataset. The results are more interesting than I expected, and a couple of them contradict things people say about Wordle with a lot of confidence.

Everything below is computed from that list. The full dataset and charts are here: https://anywordsolver.com/wordle-answer-statistics.html

The dataset

1,994 answers, 19 June 2021 → 5 December 2026
9,970 letters in total
Every answer is a real five-letter word

Letter frequency

Letter	Count	Share
E	1,080	10.83%
A	848	8.51%
R	764	7.66%
O	670	6.72%
T	649	6.51%
L	615	6.17%
I	568	5.70%
S	562	5.64%
N	503	5.05%
C	412	4.13%
U	397	3.98%
H	344	3.45%
D	337	3.38%
Y	331	3.32%
P	309	3.10%
M	274	2.75%
G	267	2.68%
B	237	2.38%
F	197	1.98%
K	183	1.84%
W	165	1.65%
V	135	1.35%
Z	38	0.38%
X	34	0.34%
Q	28	0.28%
J	23	0.23%

E and A together account for nearly a fifth of every letter Wordle has ever used.

Where answers start

First letter	Answers	Share
S	300	15.05%
C	171	8.58%
B	150	7.52%
T	132	6.62%
A	126	6.32%

X has never once started an answer. Zero out of 1,994.

Where answers end

Last letter	Answers	Share
E	395	19.81%
Y	285	14.29%
T	225	11.28%
R	169	8.48%
L	139	6.97%

Nearly one answer in five ends in E. If you have one guess left and a green tile on position 4, the answer is very likely to end in E.

The findings that surprised me

1. "Wordle never repeats words" is not true

There are 1,962 distinct words across 1,994 answers. 32 words have been used more than once. The first repeat was puzzle #1690, on 2 February 2026, and the word was CIGAR — the same word that opened the game in June 2021.

The full list of recycled words: ACUTE, AGREE, AWAKE, BATON, BLURT, BREAK, BRING, CIGAR, CLICK, CLOCK, COMET, FEIGN, FIRST, FLUME, FOCUS, FORTH, GRADE, GRIPE, IVORY, LINEN, MAJOR, MOTOR, PROVE, SANDY, SMILE, SPICY, SQUAD, STAND, STINK, STOUT, THUMB, TRACE.

The idea that answers never repeat is a reasonable mental model until you check the data, and then you find out it was true for the first 1,689 puzzles and quietly stopped being true.

2. A third of answers have a repeated letter

619 answers (31.0%) contain at least one repeated letter. There are 633 double-letter occurrences and 14 triples.

This is why a first guess with five distinct letters is slightly more informative than one with a repeated letter, even if the repeated letter would otherwise be useful.

3. Vowel counts are remarkably consistent

Vowels	Answers	Share
0	8	0.4%
1	620	31.1%
2	1,158	58.1%
3	205	10.3%
4	3	0.2%

Nearly six in ten answers have exactly two vowels. Only eight answers in the entire history have none — and only three have four.

4. The common endings are highly concentrated

The three most common final letter sequences are LLY (16), ING (16) and TCH (15), followed by TER, ATE, VER, ILL, OUT, ONE and ASH.

That list is a small cheat sheet. A word ending in -ING or -TER is far more likely than a random five-letter string.

The hardest answers

Here is where I have to be careful. There is no official Wordle difficulty rating, so I built a proxy: rare letters (J, Q, X, Z) are weighted heavily, a repeated letter adds difficulty, and very low vowel counts add more. It is my scoring function, not Wordle's data, and the page says the same thing.

The top of that ranking:

JAZZY (score 12) — two Zs, no common letters
TIZZY, FUZZY, FIZZY, DIZZY (9)
PIZZA (8)
JOLLY, JIFFY, JELLY (6)
WOOZY, QUILL, QUEUE, QUELL, QUEER, QUEEN (5)

The JAZZY cluster is not a coincidence. A double Z, a double Y and a double F are the three rarest patterns in English five-letter words, so of course they cluster at the top.

Why I built solvers next to the data

While I was in the data, I also wrote three free solvers:

Chainword solver — give it a tile rack, get every valid chain where each word starts with the previous word's last letter
Clueword solver — reverse Wordle: enter the coloured clue rows, get every word consistent with all of them (including the duplicate-letter rule, which is where most naive solvers break)
Squareword solver — complete a 4×4 grid so that every row and every column is a real word

All three are client-side JavaScript, no account, no ads on the tools.

Try it yourself

The solvers and the full analysis are free, with no sign-up:

Wordle answer statistics: https://anywordsolver.com/wordle-answer-statistics.html
Full answer archive: https://anywordsolver.com/wordle-answer-archive.html

I would genuinely like to know whether X has never opened an answer in other people's datasets too, or whether that is specific to how this answer list was curated. If you have run the same analysis on a different source, I am curious how the numbers compare.

105K Impressions, 659 Clicks: A Search Console Teardown

RAMKRISHNA PRASAD — Sat, 10 Oct 2026 16:08:49 +0000

Over the last 92 days (2026-07-07 to 2026-10-06), a tax calculator site I have been working on pulled 104,826 Google impressions and 659 clicks across 182 pages that earned at least one impression.

A 0.63% click-through rate.

If you saw only that number you would close the tab and move on. But those 659 clicks and those 104,826 impressions are two completely different stories stacked on top of each other, and the split between them is the whole lesson.

This is a straight Search Console export, no smoothing, no "estimated" numbers.

The headline numbers

Metric	Value
Period	2026-07-07 to 2026-10-06 (92 days)
Clicks	659
Impressions	104,826
CTR	0.63%
Pages with impressions	182
Total pages in sitemap	326

Less than 60% of the site ever showed up in the results at all. The other 144 pages were indexed and simply did not surface for anything worth counting.

Story one: the page that carried everything

One page produced 73.6% of all clicks on the site:

/tax-refund-calculator-2027

Metric	Value
Impressions	9,081
Clicks	485
Average position	9.29
CTR	5.34%

Look at the query behind it:

tax refund calculator 2027 - 1,931 impressions, 125 clicks, position 10.4, CTR 6.47%
2027 tax refund calculator - 862 impressions, 60 clicks, position 10.29, CTR 6.96%
tax return calculator 2027 - 333 impressions, 17 clicks, position 9.41, CTR 5.11%

Two things were true at once here:

Year-forward targeting. It is currently 2026. People were already searching for 2027 tax figures, and almost nobody had published a 2027 page yet. The competition for a keyword that does not officially exist yet is close to zero.
It is a tool, not an article. The searcher lands on something they can use immediately instead of three paragraphs of preamble before an affiliate table.

Neither of those is a trick. Both are just "give the searcher the exact thing they typed, earlier than everyone else."

Story two: 840 queries sitting on page five

Here is the position-band breakdown for the top 1,000 queries exported from Search Console:

Position band	Queries	Impressions	Clicks
1-10	42	2,536	70
11-20	45	3,762	223
21-40	73	1,372	15
41-70	442	12,568	1
71-100	398	8,495	1

840 of 1,000 queries sit below position 40. They produced exactly 2 clicks combined.

And this is the part people misread. Those are not dead queries with no demand. Two pages make it obvious:

Page	Impressions	Clicks	Position
`/irs-payment-plan`	10,700	1	64.64
`/capital-gains-tax-rates`	9,618	1	67.33

Ten thousand impressions means Google is willing to show that page for a lot of different queries. It is ranking it at position 65 and nobody scrolls that far.

That is not an impression problem. It is not a content problem. The page exists, it is indexed, it matches intent, and Google knows about every query it could serve for. It is an authority problem.

Content volume does not fix it. The content is already there.

The one that should sting

This query, from the same export:

tax refund status 2027    776 impressions    0 clicks    position 8.53

Position 8.53. Top of page one. Zero clicks.

You cannot blame authority for that. When a page ranks in the top 10 and gets zero clicks out of 776 impressions, the problem is the title and description the searcher sees, not the ranking.

That single line was worth more fixing than another 20 new pages.

What we are doing with this

1. Repeat the year-forward pattern.
The 2027 calculator proved the concept. The rest of the series (brackets, standard deduction, retirement contribution limits, EITC) is the same move with the same low competition.

2. Fix the ranking-but-not-clicking queries first.
A query at position 8 with zero clicks is a cheaper win than a query at position 65. It needs a title rewrite, not a link.

3. Build links, because content has stopped being the bottleneck.
442 queries in the 41-70 band will not move because we wrote more paragraphs about them. They already have paragraphs. They need external signals.

How to run this on your own site

Search Console, Performance, set the date range to 3 months.
Make sure Position is checked as a metric.
Export as CSV.
Bucket the queries into position bands (1-10, 11-20, 21-40, 41-70, 71-100).
Sum impressions and clicks per band.

What you are looking for is the shape:

Healthy: clicks concentrated in bands 1-20, impressions trailing off behind them.
Stuck: a fat mass of impressions in bands 41-100 with near-zero clicks. That is your authority gap and no amount of publishing fixes it.
Wasted: clicks in band 1-10 that are underperforming on CTR. That is a title and meta description problem, and it is the fastest win available.

Most people only ever look at the total row. The total row hides both of these.

The takeaway

Impressions and CTR, looked at alone, tell you opposite stories about the same site.

Impressions up, CTR flat: you are showing up further down the page than you were.
CTR up, impressions flat: you are converting the same visibility better.

The position-band breakdown tells you which lever to actually pull. For us it was clear: one page was doing the work of 181, and the fix list was shorter than the publish list.

The site behind this teardown is TaxCalcHQ - every figure in it is checked against IRS announcements before it gets published, which is why the year-forward pages landed rather than getting filtered as thin content.

All numbers above are a direct export from Google Search Console for the 92-day window ending 2026-10-06. No projections, no third-party estimate tools.