How to write an assessment rubric (with a worked workplace example)
To write an assessment rubric, pick the one task you are scoring, list three to six observable criteria, choose three or four named performance levels, and write a parallel description for each cell. Then test the rubric on real work before anyone is graded with it. The wording of the cells decides whether two scorers agree.
Most pages on this topic that we read were written by universities for instructors grading student papers. This guide is for trainers, instructional designers and L&D managers who score work done for an employer, such as a written report or a demonstration. The steps are the same, but the stakes and the scorers differ.
Key takeaways
- A rubric earns its place only when the answer is a judgment of quality. If the work is right or wrong, use an answer key or a checklist.
- Criteria matter more than level labels: research on rubric design concludes that appropriate criteria are the key to an effective rubric.
- Write descriptors that depict the work at each level, with the same sentence structure in every column, so only the degree changes.
- For pass or fail decisions, write the “meets” column first, because that is the line your decision turns on.
- Pilot the rubric with two scorers on sample work, and treat any criterion where they disagree as a writing problem to fix.
What is an assessment rubric, and when do you need one?
An assessment rubric is a scoring tool that lists the criteria a piece of work is judged on and describes what performance looks like at each level of each criterion. You need one when the work is open-ended and judged on quality, such as a written report, a presentation or a demonstration.
A rubric is not the only scoring tool. A checklist records yes or no decisions, a rating scale asks the scorer to pick a point without describing the points, and a marking guide gives model answers and mark allocations, as the Australian National University’s teaching guidance sets out. A rubric describes degrees of quality.
Objectively scored items do not need a rubric. For hands-on tasks with clear right or wrong steps, a checklist often fits better, and our guide to observation checklists for practical training sessions covers how to build one. Use a rubric when a competent learner can succeed in several ways.
Which type of assessment rubric should you use?
For training programs with more than one scorer, start with an analytic rubric: it scores each criterion separately and gives specific feedback. Choose a holistic rubric for a single overall judgment, and a single-point rubric for coaching conversations where comments matter more than scores.
NC State’s rubric guide defines the three types and lists the trade-offs, summarized below with a training-room reading added by us.
| Type | How it scores | Fits best | Watch out for |
|---|---|---|---|
| Analytic | Each criterion scored separately, then combined | Several scorers, detailed feedback, weighted criteria | Slow to write and to score; can overwhelm learners |
| Holistic | One overall level for the whole work | Quick final judgments of overall quality | Little feedback detail; no weighting of criteria |
| Single-point | Only the proficient level is described; comments go either side | Formative coaching, small groups, one scorer | More subjective; the comments take time to write |
Do not choose by reliability alone. Susan Brookhart’s 2018 review of rubric research in Frontiers in Education found no discernible pattern showing that one form of rubric was more reliable or valid than another. Pick the type that matches the decision, then put your effort into the wording.
How do you choose the criteria for an assessment rubric?
Start from what the job or course objective needs a learner to do, then list the observable evidence of it in the work. Keep three to six criteria, each measuring one thing. Criteria that only count surface features, such as the number of sources, let learners skip quality.
Brookhart’s review concludes that appropriate criteria are the key to effective rubrics, and it warns that rubrics built on counting elements or rating-scale language do not describe quality. For workplace tasks, John Peters, writing in Training Industry Magazine in 2023, recommends starting from the job task analysis and measuring only the tasks that need measuring.
On how many criteria, published guidance overlaps. MIT’s Teaching + Learning Lab says most rubrics have 3 to 8 criteria and that longer ones are harder to score. Risepoint’s faculty guide suggests about 4 to 7. For one training task, four is a sound default, as the worked example below shows.
Three tests catch weak criteria before you write a descriptor:
- Can a scorer point to evidence in the work for this criterion?
- Does the criterion measure one attribute, so one mistake is not penalized twice?
- Would an employer or client agree this criterion matters to the job?
How many performance levels does a rubric need, and what should you call them?
Use three or four levels. Fewer than three cannot show partial achievement, and MIT’s guide notes that more than five makes consistent scoring difficult. Name the levels so the minimum acceptable one is obvious, for example Exceeds, Meets, Approaching and Not yet.
Kansas State’s assessment toolkit, drawing on Huba and Freed, also lists three or four levels. MIT, citing Linda Suskie, suggests naming the minimum acceptable level. We prefer four levels with no midpoint, an editorial choice rather than a research finding, because it forces the scorer to decide whether the work reached the standard.
Should you write the top level first or the passing level first?
Sources disagree. Risepoint and the University of Illinois Chicago advise writing from the top level down. Radboud University advises setting the norm first, describing a barely passing performance for each criterion. Top-down suits grading and feedback. Passing level first suits pass or fail decisions.
Pro tip: Write The Meets Column First. In training, the decision that matters is usually “ready to work unsupervised or not”. Describe the Meets level for each criterion first, in plain terms a client would accept, then write the levels above and below it.
How do you write descriptors that two scorers read the same way?
Describe what the work contains at each level, not how good it is. Use the same sentence structure across the row, change only the degree, and avoid vague words like “good” or “adequate”. MIT’s guide and Brookhart’s review both prefer descriptive language over evaluative language.
Brookhart found that evaluative language does not give learners a depiction of work at that level. Illinois Chicago adds two rules: use mutually exclusive language, so a phrase at one level does not reappear at the next, and avoid subjective terms such as “effort”.
The table shows four common defects and a rewrite for each, using the workplace example in the next section.
| Weak descriptor | Why scorers will disagree | Rewrite |
|---|---|---|
| “Good analysis of the cause” | “Good” is a verdict, not a description | “Names the direct cause and supports it with a fact from the scenario” |
| “Includes at least three details” | Counts items and ignores whether they are the right ones | “Includes who, where and when, so a supervisor can act on the report” |
| “Excellent, very good, good, fair” | Adjectives replace content, so each scorer sets their own bar | One sentence repeated across the row, with the degree changed |
| “Shows effort and care” | Not observable in the finished work | Remove it, or score a visible feature such as completeness |
What does a finished assessment rubric look like?
A finished assessment rubric is a grid with one row per criterion, one column per level, and a description in every cell. The example below scores a short incident report that learners write after a near-miss scenario in a workplace safety course. It is illustrative, not a standard.
Learners read a one-paragraph near-miss scenario and write a half-page report. The four criteria are the four things a supervisor needs from a real report, and the Meets column is the pass line.
| Criterion | Exceeds (4) | Meets (3) | Approaching (2) | Not yet (1) |
|---|---|---|---|---|
| Facts and sequence | Events in time order, with who, where and when; nothing a reader would need to ask about | Events in time order, with who, where and when; minor gaps that do not change the account | Events present but out of order, or one of who, where or when missing | Events missing or mixed with opinion, so a reader cannot rebuild what happened |
| Cause | Names the direct cause and one contributing condition, each supported by a fact from the scenario | Names the direct cause, supported by a fact from the scenario | Names a cause the scenario does not support, or describes only the outcome | No cause given |
| Immediate action | Lists the steps taken in order, including making the area safe, plus the next step needed | Lists the steps taken in order, including making the area safe | Lists steps taken but leaves out making the area safe | No action listed |
| Usable by a supervisor | A supervisor could act on the report as written and find each point on first reading | A supervisor could act on the report after minor clarification | A supervisor would need key information before acting | A supervisor could not tell what to do |
Notice what the rubric leaves out. Grammar and spelling are not scored, because the objective is reporting a hazard, not writing style. The University of Illinois Chicago guide warns that a criterion unrelated to the objective can introduce bias.
Equal weights keep the total easy to explain: four criteria scored 1 to 4 give 4 to 16. Weight a criterion only when it matters far more than the others, and say why in writing.
Pro tip: Add A Gate, Not Just A Total. A total can hide a fatal gap: a learner can score well on facts and clarity and still omit making the area safe. Mark one or two criteria as gates that must reach Meets regardless of the total. Peters describes the same idea for skills checklists as “terminal” steps that cannot be skipped.
How do you turn assessment rubric scores into pass or fail?
Set the decision rule before anyone is scored, for example “Meets or above on every criterion, with the gate criterion at Meets or above”. A rule based on levels is easier to defend to a client than a percentage chosen by habit.
If you need a numeric pass mark, set it with evidence rather than a default. Our article on cut scores for professional training courses shows how a small provider can do that with a panel.
Rubric totals have limits. Peters notes that quantitative rubrics cannot capture behavioral factors such as how an employee treats customers. Our article on case-based assessment of soft skills shows a rubric for that area.
How do you test an assessment rubric before you use it?
Have two scorers mark the same 6 to 10 pieces of real or realistic work, independently, then compare criterion by criterion. A criterion where they disagree usually has unclear wording. Rewrite it, re-score, and only then use the rubric on learners.
MIT’s guide says to test on varied work and calibrate with other graders. Peters suggests three or four hypothetical scenarios to check that evaluators score consistently.
The table shows an illustrative pilot for the incident-report rubric: two scorers, eight sample reports, and a count of reports where they gave the same level.
| Criterion | Same level (of 8 reports) | What to do |
|---|---|---|
| Facts and sequence | 7 of 8 | Keep; review the one disagreement |
| Cause | 4 of 8 | Rewrite: scorers read “supported by a fact” differently |
| Immediate action | 8 of 8 | Keep |
| Usable by a supervisor | 6 of 8 | Add a sample report at each level to anchor it |
In this example the Cause row needs a rewrite, perhaps by listing what counts as a supporting fact. Keep the scored samples as anchor examples for new scorers. Agreement statistics such as Cohen’s kappa are covered in the checklist guide linked earlier. We set no pass threshold here, because the right level depends on the stakes.
Testing is not a one-off. The technical documentation for the National Assessment of Educational Progress says calibration sets serve as tools to prevent scorer drift, with about 10 to 20 papers re-scored after any sustained break. Your program is smaller, but the habit scales down: re-score a few anchor samples at the start of each session.
What should you do next?
Pick one open-ended task in your course that is currently scored by feel. Write four criteria, a four-level scale and the Meets column first, then run the two-scorer pilot above. Share the finished rubric with learners when you introduce the task, so it works as a guide as well as a scoring tool.
If you deliver assessments online, check how your platform handles scored responses. SimpliAssess’s features page lists item banking, test management and analytics, but its published pages do not describe rubric scoring, so ask about it directly. You can request a SimpliAssess demo and bring your rubric as the example.
Frequently asked questions
What is an assessment rubric?
An assessment rubric is a scoring guide that lists the criteria used to judge a piece of work and describes what each level of performance looks like for each criterion. It is usually laid out as a grid, with criteria as rows and performance levels as columns. It helps scorers apply the same standard and tells learners what is expected.
How many criteria should an assessment rubric have?
Published guidance varies: MIT’s Teaching + Learning Lab says most rubrics have 3 to 8 criteria, and Risepoint’s faculty guide suggests about 4 to 7. Longer rubrics are harder to score and harder for learners to use. For a single training task, start with four, and merge criteria that overlap.
What is the difference between a rubric and a checklist?
A checklist records yes or no decisions on specific items, so it suits steps that are either done or not done. A rubric describes levels of quality for each criterion, so it suits work that can succeed in several ways. Many practical assessments use both: a checklist for critical steps and a rubric for quality.
Should you share the rubric with learners?
Yes. The University of Washington’s teaching guide recommends sharing the rubric when you introduce the assignment, and MIT’s guide says rubrics tell learners what is expected before they start. Sharing it also lets learners check their own work. For a certification exam, share the criteria but keep the scenario content secure.
Can AI write a rubric for you?
AI can produce a first draft, and NC State’s rubric guide lists it as one starting point, with the institution’s AI guidelines applied and no identifiable learner data. Treat the draft as raw material. You still need to cut the criteria to those that matter, rewrite vague descriptors, and run a scoring pilot.
