Haiku Rubric
The exact scoring rubric behind this site's tools — a 100-point structural rubric and a separate 100-point literary rubric — broken down category by category.
Reviewed for accuracy · Last updated August 25, 2026
Two different rubrics
This site scores haiku two different ways, depending on what you're checking for. The Haiku Checker uses a structural rubric that mostly measures whether the poem follows the conventions of the form. The Haiku Rater uses a separate literary rubric that goes further, scoring originality, sound, and how strongly the poem ends. Both are out of 100 points, both are fully deterministic (no AI judgment involved), and both are documented in full on the accuracy and methodology page.
Structural rubric (100 points)
| Category | Points | What it measures |
|---|---|---|
| Structure | 25 | Three lines, with the traditional short/long/short shape (or a comparable modern shape). |
| Syllable accuracy | 20 | How closely the poem matches its target pattern — 5-7-5 in strict mode, a typical modern range otherwise. |
| Imagery | 15 | Sensory, concrete detail (10 pts) plus a penalty for abstract or evaluative language (5 pts). |
| Seasonal context | 10 | Whether a recognized seasonal word (kigo) grounds the poem in time — a missing kigo is scored as neutral, not a failure. |
| Juxtaposition | 10 | Whether a clear cut, marked by punctuation or a line-break shift, is detected between two parts of the poem. |
| Brevity & clarity | 10 | Penalizes abstract language and lines that run long for the form. |
| Natural language | 10 | Penalizes repeated content words that suggest padding rather than deliberate repetition. |
Literary rubric (100 points)
The Haiku Rater splits 100 points evenly across ten categories, 10 points each:
| Category | What it measures |
|---|---|
| Originality | Penalizes known clichéd phrases in favor of a more specific image. |
| Image clarity | Rewards concrete imagery; penalizes abstract or evaluative language. |
| Brevity | Rewards a spare average word count per line. |
| Sound | A small bonus for alliteration or repeated sound — not required, just noticed. |
| Movement | Whether a sense of motion comes through, versus a fully static scene. |
| Juxtaposition | Whether a clear cut is present between two parts of the poem. |
| Seasonality | Whether a recognized seasonal word is present. |
| Emotional subtlety | Penalizes emotion named directly ("happy," "sad") instead of implied through image. |
| Word economy | Penalizes filler words ("very," "really") and repeated content words. |
| Ending strength | Rewards ending on a concrete word rather than a light, trailing-off word. |
Using a rubric for your own grading
If you're a teacher grading haiku by hand, both tables above work as a plain checklist even without running the poem through a tool: structure and syllables first, then imagery, seasonal grounding, and the cut, and — if you want to grade for literary quality rather than just form — originality, sound, and how the poem ends.
Frequently asked questions
Why are there two different rubrics?
They measure different things. The structural rubric (behind the Haiku Checker) mostly asks "does this follow the conventions of the form?" The literary rubric (behind the Haiku Rater) asks a harder question — "is this a good poem?" — and scores originality, sound, and ending strength on top of structure.
Is a perfect 100 the goal?
Not necessarily. Both rubrics are deterministic, formula-based estimates, not a definitive verdict on literary quality — a poem that deliberately breaks a convention for effect can still be excellent. Treat a low score in one category as a prompt to look closer, not as a final judgment.
Are these scores generated by AI?
No. Every sub-score comes from a documented, deterministic formula — word lists, pattern matching, and counting — not a language model. The same input always produces the same score.
Can I use this rubric to grade student haiku?
Yes — the category breakdowns above work as a plain-language grading checklist, whether or not you use the tools themselves to generate the numeric score.