Crossword Puzzle Generator

{{ summaryHeading }} {{ summaryValue }} {{ summaryLine }} {{ placementBadge }} {{ fillPercent }}% fill {{ crossingRate }}% crossed

Puzzle title:

Use a short lesson, event, or theme name.

Answers and clues:

Enter one ANSWER: clue per line. Answers normalize to uppercase A-Z.

{{ sourceMeta }}

{{ sourceHint }}

Grid size:

Choose 5-25 cells per side; start larger than the longest cleaned answer.

Placement strategy:

Use Quality search for final sheets or Greedy for a quick fit check.

Minimum answer length:

Use 2-10 letters; three is a practical default for classroom puzzles.

Block symmetry:

None maximizes fit; rotation and mirrors trade some fit for a more regular silhouette.

Best-of attempts:

Use 1-12 attempts. More attempts can find a better layout but require more work.

Search budget:

Use 200-3000 ms-equivalent effort per attempt; larger values explore more branches.

ms

Layout seed:

Keep the seed with the word bank when the layout must be reproduced.

{{ workflowFeedback }}

Header note:

Leave blank to keep the printable header compact.

High-contrast print:

Leave off for the normal print palette; turn on for low-quality photocopies.

{{ params.high_contrast ? 'On' : 'Off' }}

Show answer key

Correct the authoring inputs and generate a puzzle.

{{ params.title || 'Crossword' }}

{{ printHeaderNote }}

Across

{{ row.clue }}

Down

{{ row.clue }}

No.	Direction	Answer	Clue	Length	Copy
{{ row.number }}	{{ row.direction }}	{{ row.answer }}	{{ row.clue }}	{{ row.length }}

Answer	Clue	Length	Status	Copy
{{ row.answer }}	{{ row.clue \|\| '—' }}	{{ row.length }}	{{ row.status }}

A crossword becomes possible when answers share letters that can occupy the same square. A strong word bank therefore needs more than a common theme: it needs useful overlaps, lengths that fit the grid, and clues that still make sense after the answers are normalized.

Automatic construction searches for legal Across and Down placements, but it cannot make incompatible words interlock. A list full of rare letters or long answers may leave entries unused even when every line is valid. Enlarging the grid creates more positions, while replacing one awkward answer often improves the puzzle more than adding many search attempts.

Answer quality: use recognizable words with varied lengths and shared letters.
Clue quality: write clues for the intended audience; placement does not fact-check or rewrite them.
Grid quality: check both how many entries fit and how many cells are true crossings.
Print quality: leave enough grid space for readable cells, especially on classroom sheets.

Symmetry is a visual preference, not proof of publication-style construction. A rotational or mirrored bias can make the filled-and-blocked pattern look more regular, but the extra preference may reduce the number of answers that fit. Choosing no symmetry gives placement coverage the most freedom.

A generated puzzle is a draft for review. Read every clue, inspect unused entries, confirm numbering, and solve the printed version before distributing it. Blank clue text falls back to showing the answer, which is useful during layout work but unsuitable for a finished puzzle.

How to Use This Tool:

Prepare the word bank first, then spend more search effort only after the entries themselves are suitable.

Enter one ANSWER: clue line at a time or load one TXT file. Answers are converted to uppercase A–Z, duplicate answers are removed, and clue text after the first colon is kept.
Choose a Grid size from 5 to 25 cells per side and a Minimum answer length from 2 to 10 letters. Start with a grid at least as wide as the longest cleaned answer.
Use Quality search for a final draft or Greedy for a quick fit check. Add a symmetry bias only when a more regular silhouette matters more than maximum placement.
Set the attempts, search budget, and Layout seed, then generate. Keep the same cleaned entries and all placement settings with the seed when the grid must be reproduced.
Review unused entries and crossings before printing the puzzle or answer key. Replace weak-fit answers or enlarge the grid before simply increasing the search budget.

Interpreting Results:

The placed count is the clearest coverage measure. A high fill percentage can still describe a poor crossword if most letters belong to only one direction, so compare it with the crossing-cell count and crossing rate. Unused entries are not necessarily invalid; they may be below the minimum length, longer than the grid, or simply impossible to place legally in the winning layout.

Crossword result checks
Result	Useful reading	Next check
Placed entries	Coverage of the cleaned word bank	Read the unused status for every leftover answer.
Crossing cells	Squares shared by Across and Down answers	Make sure the grid interlocks rather than forming isolated runs.
Fill percentage	Letter cells divided by all grid cells	Do not treat density alone as puzzle quality.
Symmetry percentage	How closely occupancy matches the selected mirror or rotation	Confirm that symmetry did not cost too many entries.

Technical Details:

Construction uses a deterministic, repository-authored heuristic rather than a publication rulebook. It accepts only placements that stay inside the square, preserve matching letters at crossings, keep answer ends clear, and prevent newly placed letters from touching parallel words side by side.

Rule Core

Normalize each answer to uppercase A–Z, discard empty results, and keep the first occurrence of a duplicate.
Exclude answers shorter than the selected minimum or longer than the grid. They remain visible as below-minimum or too-long entries.
Place the first eligible answer near the center. Later answers must share at least one matching letter with the existing grid.
Rank legal candidates mainly by crossing count, then by closeness to the center, with a penalty for missing the selected symmetry.
Choose the winning attempt by placed entries first, then occupied letter cells, crossing cells, and symmetry percentage.

The candidate score gives each crossing 100 points, adds up to 20 points for central placement, and subtracts 0.75 for each occupied cell whose mirrored partner is empty. This score orders candidate positions inside an attempt; it is not shown as a universal crossword-quality rating.

Q = 100 C + \max (0, 20 - \frac{D}{L}) - 0.75 M

In the placement score, C is the number of crossings, D is the total distance of the answer’s cells from the grid center, L is answer length, and M is the number of symmetry misses.

Formula Core

The result percentages describe the accepted grid. Each is rounded to one decimal place.

\begin{matrix} Fill & = & \frac{letter cells}{{grid size}^{2}} \times 100 \\ Crossing rate & = & \frac{crossing cells}{letter cells} \times 100 \\ Symmetry & = & \frac{matching occupancy pairs}{tested symmetry pairs} \times 100 \end{matrix}

With no selected symmetry, the symmetry percentage is reported as 100 because no mirror pairs need to be satisfied. Quality search explores up to five leading candidates at each branch and converts the selected 200 to 3000 ms-equivalent budget into a deterministic node allowance. Greedy accepts only the highest-ranked next placement. Neither strategy guarantees an optimal grid.

The layout seed changes entry order after the first attempt. Reproduction requires the same normalized answers, grid size, strategy, minimum length, symmetry, attempt count, budget, and seed.

Worked Examples:

Two answers with one crossing

CATS: Household pets and ART: Creative work fit a 5 × 5 grid with one shared T. Six letter cells are occupied, so fill is 24%; one of those six cells is crossed, giving a 16.7% crossing rate.

Valid entries that remain unused

A themed list can normalize cleanly and still leave leftovers when the answers share few letters. The entry review distinguishes below-minimum, too-long, and unused answers, which tells the author whether to change a length rule, enlarge the grid, or replace a hard-to-cross word.