A good rubric does two jobs. It makes grading faster and fairer, and it tells students what good work looks like before they start. A poor rubric does neither. It is full of words like "excellent" and "good" that mean different things to different readers, so the same essay can earn two different marks from two different teachers, and students are left guessing what was wanted.
This guide covers the three main rubric types and when to use each, a six-step method for writing one, worked examples you can adapt, and how to use an AI rubric maker without ending up with something generic. If you want the short version, write three to five criteria you actually taught, describe each level in terms you could observe in the work, and test the rubric on real samples before you use it.
What are the three types of rubric?
An analytic rubric is a grid, with criteria down the side and performance levels across the top. Each cell describes what that level looks like for that criterion, so the grid gives students the most detailed feedback. The cost is time, because it takes the longest to write and the longest to use.
A holistic rubric is one scale with a single description for each level. It is quick to apply and works for tasks where the parts cannot really be separated, but it tells students less about what to improve. A single-point rubric lists only the target level for each criterion and leaves space on either side to note where a student falls below or rises above it. It is fast to write, focuses feedback on the standard, and is popular with teachers who find full grids too long.
Which should you use? An analytic rubric suits a major assignment that you will mark carefully. A single-point rubric suits regular work and feedback, when you want to say something specific without filling in a grid. A holistic rubric suits quick, low-stakes tasks where a single overall judgement is enough.
How do you write a rubric in six steps?
Start with the goal. Ask what a student should be able to do after this task, because the rubric should measure that and not everything you could possibly mark. Then choose three to five criteria. More than five and the rubric becomes hard to use, and students stop reading it, so pick only things you actually taught.
Next, decide the levels. Three or four are enough, and it helps to use words that describe the work, such as "meets" and "approaching," rather than labels that judge the student. The most important step is to describe each level in observable terms. "Uses two or more pieces of evidence that support the claim" can be checked by reading the paragraph, while "good use of evidence" cannot. A useful trick is to write the top and the bottom levels first, describing what strong work and weak work look like, and then fill the middle from those two ends.
Finally, test the rubric on real work. Grade three pieces of student work, or samples from last year, and see whether you and a colleague would give the same scores. Wherever two people would disagree, rewrite that descriptor until they would not.
What does an analytic rubric look like?
Here is a four-level analytic rubric for a paragraph. Every descriptor can be checked by looking at the writing, and none of them relies on an undefined word like "excellent."
| Criterion | 4 – Strong | 3 – Meets | 2 – Approaching | 1 – Beginning |
|---|---|---|---|---|
| Claim | Clear, specific claim that answers the question | Clear claim that answers the question | Claim is present but vague or partly off topic | No clear claim |
| Evidence | Two or more relevant pieces of evidence, accurately used | At least one relevant piece of evidence | Evidence is present but weak or not clearly linked | Little or no evidence |
| Explanation | Explains how the evidence supports the claim, with insight | Explains how the evidence supports the claim | Repeats the evidence with little explanation | No explanation |
| Writing conventions | Few or no errors, and clear sentences throughout | Some errors that do not affect meaning | Errors sometimes make meaning unclear | Errors often make meaning unclear |
Notice that the levels differ in a real way. The step from "meets" to "strong" in the evidence row is about quantity and accuracy, and the step from "approaching" to "meets" is about relevance and linking. If your levels only swap adjectives, students cannot tell what to do differently.
What does a single-point rubric look like?
The single-point version of the same task lists only the standard and leaves the outer columns blank for your comments.
| Below the standard (notes) | Standard | Above the standard (notes) |
|---|---|---|
| Claim: states a clear claim that answers the question | ||
| Evidence: uses at least one relevant, accurate piece of evidence | ||
| Explanation: explains how the evidence supports the claim | ||
| Conventions: errors do not get in the way of meaning |
You write short comments in the outer columns, such as "Your evidence is strong, but I couldn't see how it connects to the claim." Students get specific feedback without having to decode a grid, and you spend your time on comments instead of on choosing between boxes.
How can you use an AI rubric maker?
An AI rubric generator can give you a draft in seconds, and a draft is a real help. The catch is that a vague request produces a vague rubric, so give it the grade, the criteria, the levels and the style you want.
Create an analytic rubric for a grade 8 persuasive paragraph. Use four criteria: claim, evidence, explanation and conventions. Use four levels: Strong, Meets, Approaching, Beginning. Write every descriptor in observable terms, with no words like "good" or "excellent" unless they are defined. Keep each cell under 20 words.
Then check the result before you use it. Confirm that the criteria are the things you actually taught, and delete any that are not. Read each descriptor and ask whether you could see it in student work, and rewrite anything vague. AI often produces levels that only swap adjectives, such as "some evidence" against "good evidence," so check that each step up describes something different. Look for criteria that overlap, because then students get marked twice for the same thing. Adjust the levels or weights to match your grading scale, and trim the whole thing if it takes more than a couple of minutes to read.
When you use an AI tool to help with rubrics or feedback, do not paste student names or identifying details into a general tool. Check your school's rules on what may be entered.
How do you share a rubric so it actually helps?
A rubric only helps students if they see it before they start. Hand it out with the task and talk through one criterion, so students know how to read it. One of the quickest ways to make the standards clear is to show two sample pieces of work and ask students to score them against the rubric, then discuss where they disagreed. Before handing in, students can self-assess by marking one thing they did well and one thing to check. And when you give feedback, use the rubric's own language, so that your comments and the rubric point to the same things.
What mistakes do rubrics usually contain?
The most common is too many criteria, since five is usually the ceiling. Close behind is vague wording, because words like "good," "excellent" and "sufficient" mean different things to different people. Teachers sometimes mark things they did not teach, but if it is on the rubric, students should have had a chance to learn it. A quieter problem is rewarding length or neatness by accident, so check that the top level is about quality. And a rubric that is never revised keeps its flaws, so after grading, note which descriptors caused disagreement or confusion and rewrite them.
Frequently asked questions
How many levels should a rubric have? Three or four. More levels make it harder to tell them apart.
Should students help write the rubric? For older students it can work well. Asking them to describe what strong work looks like builds understanding, and you can keep the final wording.
Can I turn rubric scores into a grade? Yes, but decide how before you start, and tell students. A simple approach is to give each criterion equal weight and convert the total, adjusting the weight only if one criterion matters more.
Is it fine to reuse a rubric? Reuse the structure, but check the descriptors against each new task. A rubric for a persuasive paragraph does not fit a lab report without changes.