AI marking tools ingest your rubric or success criteria, break them into individual criteria and performance bands, then judge each student response against those descriptors to draft a suggested mark and feedback. You review and edit every judgement, so the standard stays yours rather than the model's.
What does it mean for an AI marking tool to use your rubric?
It means the tool works from the exact criteria you assess against rather than a generic notion of quality. You give it your rubric, success criteria or marking guidelines, and it treats those descriptors, performance levels and weightings as the yardstick for every response. The judgement stays anchored to your standard, so the feedback reflects the task in front of your class instead of the model’s own opinion about what strong writing looks like.
This matters because a rubric encodes decisions you have already made about what counts and how much it counts. A tool that ignores that structure can produce comments that read well but do not line up with how you actually award marks. Using your rubric is what turns a general AI assistant into something aligned with your specific assessment, which is the whole point of reaching for a marking tool rather than a chatbot.
How do AI marking tools read a rubric or success criteria?
Most tools accept a rubric in one of three ways, then parse it into its parts: the individual criteria, the performance levels or bands beneath each one, and the descriptors that define them. That parsing step is what lets the tool check a response against each strand separately instead of forming a single vague impression. It is also where the structure of your original document does most of the work.
The cleaner the parsing, the better the marking. A well-organised rubric with distinct criteria and clearly separated band descriptors gives the model unambiguous targets to test against. Vague or overlapping wording, or a rubric dropped in as one undifferentiated block of text, leaves more room for the tool to guess. In practice the way you express your criteria directly shapes the quality and reliability of what comes back.
- Upload: attach an existing rubric or marking guideline as a document.
- Paste: copy your criteria and band descriptors straight into the tool.
- Build: enter criteria, levels and descriptors into a structured form the tool provides.
How does an AI marking tool apply your criteria to student work?
Once your criteria are loaded, the tool compares each response against them one criterion at a time. For every criterion it looks for evidence in the writing, decides which band or level that evidence best matches, and drafts a short explanation of the judgement. The output is a suggested mark or level for each strand, together with commentary tied to specific descriptors rather than a single unexplained grade.
Stronger tools cite evidence in the student’s own words instead of simply asserting a result, which lets you check the reasoning in seconds. They also apply the same descriptors across a full set of scripts, so the thirtieth response is measured against the same standard as the first. That consistency is one of the main reasons teachers turn to these tools when a large marking pile would otherwise drift as fatigue sets in.
What should you look for so a tool marks the way you do?
Look for faithful capture of your rubric’s structure, use of your own wording, evidence for each judgement, and consistent application across responses. A tool can summarise a rubric neatly and still blur the distinction between adjacent bands, which is precisely where marks are won and lost. The practical test is to run it on a few scripts you have already marked and compare its calls with yours before you trust it on new work.
Weighting and control matter just as much. If your rubric weights analysis more heavily than expression, the tool should reflect that rather than treating every criterion as equal. And every suggestion needs to be editable, because you remain the marker of record and may read a response differently. The table below sets out stronger and weaker approaches to each of these so you know what to ask a vendor to demonstrate.
Where do AI marking tools get rubrics wrong?
They struggle most where a rubric relies on holistic judgement, context or knowledge you carry in your head. Band language such as sophisticated or perceptive means something specific in your subject and cohort, and a model can read it too generously or too literally. It may also over-reward surface features like length or advanced vocabulary that often correlate with quality but are not the criterion you are actually assessing.
A further risk is invented evidence: a tool can claim a response does something it does not, or miss a subtle point a student genuinely made. That is why review is not optional. Treat the draft as a well-organised first pass against your criteria, then correct the calls that need your professional judgement. Used this way the tool speeds up the routine comparison while you keep authority over the finer decisions.
How do success criteria and comment banks fit alongside a rubric?
Success criteria and comment banks give the tool more of your voice to work with. Where a rubric sets the bands, success criteria spell out what a strong response actually contains, and a comment bank supplies phrasing you already trust. Feeding all three in means the drafted feedback sounds like you and steers students toward the same next steps you would have written by hand, not a generic set of tips.
This combination also supports consistency across a class and across a faculty. When several teachers share one rubric, one set of success criteria and one comment bank, an AI tool applies them uniformly, which underpins the comparable, defensible judgements assessment bodies expect. You still review each result, but you start from a common baseline rather than a blank page, and moderation conversations become easier because everyone is working to the same descriptors.
How can a tool like JeddAI apply your criteria consistently?
A tool like JeddAI applies your criteria consistently by working from your own rubric, success criteria and comment banks, then drafting feedback and marks aligned to them for you to review. It handles the repetitive job of comparing every response against each descriptor, while you keep control of the judgement and edit anything that needs your read of the student.
Built in Australia and used across schools, it is designed to save marking time without taking the decision out of your hands. If you want to see how your existing rubric behaves, you can Get started with JeddAI and try it on a task you have already marked, so you can compare its calls with your own before relying on it for a full class.
| Capability | Stronger approach | Weaker approach |
|---|---|---|
| Rubric ingestion | Parses your criteria, bands and descriptors as separate, structured parts | Treats the rubric as one block of text and forms a single impression |
| Wording | Marks against your descriptors and language | Substitutes generic quality standards for yours |
| Evidence | Points to evidence in the student's own words for each judgement | Asserts a band or mark without showing why |
| Consistency | Applies the same descriptors across the whole class | Judgements drift between the first and last response |
| Weighting | Reflects how your rubric weights each criterion | Treats every criterion as equally important |
| Teacher control | Every suggested mark and comment is editable before use | Outputs are hard to override or presented as final |
Frequently asked questions
Can I upload my existing rubric, or do I have to rebuild it?
Most tools let you upload or paste an existing rubric, or build one in a structured form. A clearly organised rubric with distinct criteria and band descriptors parses more reliably than a single block of text.
Will an AI marking tool give the same mark I would?
Often it lands close on clear criteria, but not always, especially where judgement is holistic. Test it on scripts you have already marked and compare, then treat its suggestion as a first pass you confirm or change.
Do these tools work with holistic rubrics or only analytic ones?
They handle both, but analytic rubrics with separate criteria and bands give clearer targets. Holistic rubrics with broad descriptors leave more room for the model to interpret, so review is even more important.
Can the tool use my own comment bank wording?
Yes, if it lets you attach a comment bank. Feeding in your phrasing helps the drafted feedback sound like you and point students to the next steps you would normally suggest.
Is the AI's suggested mark final?
No. You remain the marker of record, and every suggested mark and comment should be editable. The tool speeds up the routine comparison against your criteria; the final judgement stays yours.
Get started with Jeddle
Jeddle gives teachers and students instant, syllabus-aligned feedback powered by JeddAI.
Looking for study material? Browse Jeddle's Australian-English subject resources, or explore more articles on EdTech & Tools.
