Read: ~8 min · Try it: 20–25 min
You know that feeling when you're grading the 12th essay in a row and silently thanking Past You for taking the time to write a good rubric?
This post is about that version of you.
And it's also about what happens when we bring AI assistants like TA39 into the mix—because suddenly your rubric isn't just a guide for you. It becomes part of how the AI understands quality, applies standards, and generates feedback.
Why rubrics matter even more with AI
Most teachers already believe in rubrics. We've seen how they:
- make expectations visible for students
- support fairer grading
- help us give faster, more focused feedback
But when you bring AI into the process—whether it's co-grading, suggesting feedback, or helping you review student work—your rubric stops being just a helpful tool and becomes a much more active part of the system.
Humans can work around a fuzzy rubric. AI can't.
It will follow what's written, not what you meant.
If your rubric says "uses evidence effectively," an experienced teacher brings years of classroom context, content knowledge, and judgment to that phrase.
An AI doesn't have your gut. It has your words.
So if those words are vague, overlapping, or open to interpretation, the AI will still do its best—but its "best" may not look much like yours.
What AI is actually doing with your rubric
Tools like TA39 are not just "scoring" writing. They are doing something more demanding:
- reading your rubric as a set of rules
- matching student work to those rules
- explaining that reasoning in language a student can understand
- sometimes aligning evidence or comments to specific parts of the work
That is a lot to ask from a rubric.
If the rubric is clear, the AI has a much better chance of doing work you will recognize and trust.
If the rubric is fuzzy, the AI will still produce an answer—but now it is filling in gaps you may not have realized were there.
That is where problems tend to start.
Three quiet ways a "good enough" rubric breaks down with AI
Here are a few common patterns.
1. The double-counting problem
Imagine a rubric with these two criteria:
- Use of Evidence
- Accuracy & Relevance of Details
On paper, that can look fine. In practice, they often end up judging the same thing twice.
A human teacher may balance that out without even noticing. An AI is more likely to follow the structure literally.
That means a strong quote might get rewarded in both places. Or one weak detail might affect multiple criteria.
The issue is not that the AI is doing something wrong. The issue is that the criteria overlap more than we may realize when we write them.
2. Vague level descriptions
Consider level labels like:
- 4 – Strong thesis and clear ideas
- 3 – Mostly clear ideas
- 2 – Somewhat unclear ideas
- 1 – Limited clarity
We all know what those feel like when we're grading.
But what does "mostly clear" mean in practice? Is it about organization? Sentence clarity? The thesis? How much confusion is too much?
An AI has to pick a definition. And once it does, it will apply that definition very consistently—even if it is not the one you had in mind.
So you end up with scoring that is consistent in one sense, but not necessarily aligned with your standard.
3. The holistic fog
Sometimes a rubric is really a checklist wrapped in a paragraph.
For example:
Student shows strong writing skills, with good organization, appropriate vocabulary, and clear ideas.
That might sit under one criterion such as "Overall Writing Quality."
Humans are often quite good at reading that as a holistic judgment. AI can mimic that. But when we also ask it to justify the decision—Why is this a 3 instead of a 4? What in the writing supports that?—the fuzziness becomes harder to ignore.
Holistic criteria are not wrong. But they are harder to use as the main engine for AI-supported feedback.
What a stronger rubric usually looks like
The goal is not to turn your rubric into a legal document.
The goal is to make your professional judgment more visible and easier to apply consistently—for you, for students, and for the AI.
A few shifts make a big difference.
Make levels more observable
Instead of:
Uses evidence effectively
try:
Includes at least 3 pieces of relevant evidence that support the claim, and explains how each piece connects to the argument.
Now:
- you know what you are looking for
- the AI knows what it is looking for
- the student knows what stronger work looks like
Separate criteria that are doing different jobs
If one criterion says:
Organization and language are clear
that is probably two things: how ideas are structured and how clearly those ideas are expressed.
When those are separated, the feedback becomes more useful too. A student can see whether the issue is with structure, sentence-level clarity, or both.
Anchor levels with clearer differences
Instead of:
- 4 – Strong thesis
- 3 – Clear thesis
- 2 – Weak thesis
- 1 – No thesis
you might write:
- 4 – Thesis is specific, arguable, and clearly placed. It previews the main reasons or points.
- 3 – Thesis states a clear main idea, but may be general or not fully preview the reasons.
- 2 – There is an attempt at a main idea, but it is vague, off-topic, or hard to locate.
- 1 – No identifiable thesis or main idea.
That gives both you and the AI something more concrete to work with.
Where TA39 fits in: "Optimize with AI"
In TA39, this is exactly the kind of work Optimize with AI is designed to support.
You start with the rubric you already have. Then TA39 can help you review it by identifying places where the wording may be too vague, where levels overlap, or where performance distinctions are harder to apply consistently.
It can suggest a revised version that keeps your general intent, but makes the structure clearer and the level descriptions more explicit.
That matters because this feature is not just about formatting. It replaces the old idea of a separate rubric-conversion step with something more useful: a review-and-improvement step inside the rubric workflow itself.
Think of it less as a judge and more as a rubric coach.
It does not replace your expertise. It helps surface parts of the rubric that may need clearer language before you save and use it.
Optimize with AI helps identify vague wording, overlapping distinctions, and places where a rubric can be made clearer before you save it.
If you want guidance on that workflow, see Rubrics formatting guidelines for TA39.
Why this is not about replacing teacher judgment
A clearer rubric does not replace your judgment. It documents it.
It helps when students ask why they received a score. It supports consistency across classes or teaching teams. It gives students a fairer picture of what strong work looks like.
And yes, it helps an AI assistant work more in line with your standards instead of guessing at them.
When you tighten a rubric, you are not giving up control. You are making your expectations easier to see and easier to apply.
Try it: The Rubric Analysis Protocol
About 20–25 minutes. Pen and one sheet of paper, beside a rubric you actually use.
Pick a rubric you have applied to a real class — recently enough that you remember where it worked and where it didn't. Then work through these three steps. Keep a single sheet of notes beside you as you go. The notes themselves are what matter; they are the artifact you'll bring into the next step.
Step 1 — Initial Assessment (about 5 minutes)
Read the rubric through once, without grading anything. Write a single line in answer to each:
- What stands out as particularly clear, or particularly confusing?
- Which criteria seem most useful? Which seem least useful?
- How long would this take to apply, fairly, to twenty-five student papers?
Step 2 — Dimension-by-Dimension Analysis (about 15 minutes)
Go through the rubric again with a pen. Mark the criteria directly.
Clarity. Circle any vague terms — good, adequate, limited, several, many. Underline any words students would not understand. Mark any pair of levels you cannot tell apart in practice.
Usability. Count the criteria. More than six is usually too many. Identify any criteria that overlap. Note any criterion that is hard to observe in real student work.
Accuracy. Compare the rubric to your learning objectives for this assignment. What is missing? Check the language against the curriculum framework — IB criterion descriptors, AP scoring guidelines, CBSE marking scheme, IGCSE assessment objectives.
Ease of applicability. Think of three recent submissions. Could you score them quickly and confidently with this rubric? Would it help you write better feedback, or only justify a grade?
Step 3 — Prioritize (about 5 minutes)
Write down the three changes that would most improve the rubric. Then write a single line at the top of the page naming your assessment intent — what this assignment is actually trying to assess. That intent is the anchor for the next step.
Your sheet should end with three labeled notes, in this shape:
Assessment intent: ___________________________________
Vague criterion(s): __________________________________
Overlapping levels (or overlapping criteria): _______
Those three labels — assessment intent, vague criterion, overlapping levels — are the exact terms the next video will use. Bring this sheet with you when you watch The Rubric Optimizer: Deep Dive. The video walks through running the same rubric through TA39's Optimize with AI and comparing its suggestions against these notes. The comparison is where professional judgment sharpens. The optimizer is most useful when read against this analysis, not in place of it.
What to read or watch next
Once your Rubric Analysis Protocol notes are in hand, watch The Rubric Optimizer: Deep Dive and run the same rubric through the optimizer. Compare what it flags against your notes.
Continue Layer 3 with Module B: Feedback Voice & Structure, beginning with Presenting Your Voice in Feedback.
For ongoing rubric refinement once you have run a few assignments, see Calibrating Quality: Is Your Feedback Good Enough? in Layer 4.
Final thought
AI will not fix a fuzzy rubric.
If anything, it will make the fuzziness easier to notice.
But when you give it a clear, thoughtful, well-structured rubric—the kind many teachers are already close to writing—it can support some of the best parts of your practice:
- clearer expectations
- more consistent grading
- more actionable feedback
That work has always mattered.
What has changed is that tools like TA39 make it easier to see which parts of a rubric are doing that work well—and which parts need a clearer next draft.