Create intermediate Works with: Any AI

Rubric for Grading AI Outputs

Rubric scoring Has example

Score responses consistently.

The Open buttons send the prompt text as currently shown below to that site only when you click. Fill in the placeholders first.

Create a grading rubric for AI outputs. The task and the type of output: [TASK_AND_OUTPUT_TYPE]. The reader and what they need: [READER_AND_NEED]. The most serious failures to catch: [SERIOUS_FAILURES].

Produce a rubric table: Criterion | Weight | 1 (poor) | 3 (acceptable) | 5 (excellent) | Evidence to look for.
Criteria to consider: accuracy against the source or known facts; completeness (answers all parts); relevance and focus; structure and format compliance; clarity and tone for the reader; use of evidence and citations; handling of uncertainty (admits what it does not know); safety and privacy; originality or usefulness where relevant; and efficiency (length without padding).

Add hard-fail rules (invented facts or sources, unsafe advice, ignoring the format, exposing private data), a scoring formula (weighted total out of 100 with an example calculation), instructions for graders (read the source first, score each criterion separately, note one quote as evidence), a calibration step (two graders score the same five outputs and discuss differences), and a template for recording scores. Include guidance on avoiding bias toward longer or more confident answers. Give two example graded outputs (fictional) with scores and comments.

Fill in before use

Optional: type your details and the prompt above updates. Nothing is saved or sent.

Replace each bracketed placeholder with your own details, then paste into Any AI or another assistant.

Criterion | Weight | 1 | 3 | 5
Accuracy | 30 | invented facts | minor errors | all claims supported
Completeness | 20 | misses key parts | most parts | all parts
Format compliance | 15 | ignores format | partial | exact
Clarity and tone | 15 | confusing | adequate | clear for reader
Handling uncertainty | 10 | overconfident | some caveats | says what is unknown
Safety/privacy | 10 | exposes data | borderline | clean
Hard fails: invented sources; unsafe advice. Score = sum(weight x score/5). Example: accuracy 4, completeness 3, format 5, clarity 4, uncertainty 3, safety 5 = 24 + 12 + 15 + 12 + 6 + 10 = 79.

• grading