The AI scoring rubric is not meant to declare which poem is “best.” It is a way to look at each Worku from several angles and ask a simple question: how well did the poem turn a set of required words into an actual haiku?
Each poem is scored out of 25 points across five categories, with a maximum of 5 points in each.
- Imagey /5
Does the poem create something the reader can see, hear, feel, smell, or otherwise experience? Haiku usually benefits from concrete detail. A poem built entirely from ideas can still work, but a strong image often gives the reader somewhere to stand. - Constraint Integration /5
How naturally do the required words fit into the poem? This is especially important in Worku. The best constraint words feel as though the poet chose them freely, even when they were imposed by the game. A lower score here usually means a word feels bolted on, grammatically strained, or disconnected from the scene. - Emotional Resonance /5
Does the poem leave some feeling behind—humor, sadness, tension, tenderness, surprise, unease, delight? This does not mean every poem needs to be profound. A comic poem can score well if the joke genuinely lands. - Rhythm & Flow /5
How smoothly does the poem move? For traditional Worku entries, the AI also pays attention to the 5–7–5 syllable structure. But meter is not the only issue. Awkward syntax, clunky phrasing, or a line that has to be reread can lower this score even if the syllable count is correct. - Craft & Cleverness /5
This is where the AI looks for the extra move: wordplay, structure, ambiguity, compression, a strong metaphor, or a satisfying turn. A clever use of a constraint word can help here, but cleverness alone is not enough if the poem is too hard to understand.
One thing the scoring has increasingly focused on is the turn.
A haiku often begins by giving the reader an image or situation, then shifts in some way. The last line may reframe what came before, reveal something, contradict it, make it funny, or suddenly deepen it. That turn is often what makes a short poem feel like a haiku rather than simply a seventeen-syllable description.
Worku makes this especially difficult. The poet is handed words that may be abstract, technical, awkward, or completely unrelated. Those words can give you nouns and verbs. They can suggest a scene. But they do not give you the turn.
That still has to be found.
The AI therefore tends to reward poems in which the constraint words disappear into a believable scene and one of those words—often the final one—does some structural work. It may deliver the joke, reveal the subject, change the emotional temperature, or explain why the earlier lines mattered.
The rubric also values clarity. A poem does not need to explain everything, and some mystery is often a strength. But there is a difference between productive ambiguity and a scene the reader simply cannot reconstruct. Sound, rhyme, or clever constraint use cannot fully compensate if the basic action is impossible to follow.
And the scores are not intended as permanent judgments. They are closer to a close reading: one way of identifying what is already working and where a poem may be under pressure from its constraints.
The larger goal is not to make every Worku score 25.
It is to notice the craft involved in making a handful of stubborn words feel inevitable.