Ask ten teachers what AI essay grading does and you will get ten different guesses. Some picture a black box that spits out a number. Others assume it just counts words and checks commas. In practice, a rubric-based tool reads each essay against the criteria you set, and a novel unit like Ender's Game is a good place to see how that plays out.

Take a typical assignment: argue whether Colonel Graff was justified in isolating Ender. The teacher's rubric might cover thesis, use of textual evidence, reasoning, organization, and conventions. The AI reads the essay and matches what it finds against the descriptions for each level. It then drafts a score and comments for each criterion.

That process is strongest where the rubric is concrete. Does the thesis make a claim someone could disagree with? Does the evidence come from named scenes, and does each one get explained? Those questions have visible answers on the page, so a rubric-driven reader can apply them the same way to paper one and paper one hundred.

It is less reliable on things that live outside the page. A student who builds on an idea from Tuesday's seminar, or who writes in a voice you recognize as a big improvement, gives you context an AI does not have. That is why the draft score is a starting point. You still read, adjust, and decide.

What the AI actually looks at

Most of the work comes down to matching evidence in the essay to the descriptors in the rubric. If a student claims Ender becomes colder as Battle School goes on, the tool checks whether the paper backs that up with moments from the book, like his fights with Bonzo or the way he runs his army. It also checks whether the student explains why those moments matter. Papers that only retell the plot tend to land lower on the reasoning line, which is what most teachers want.

• Whether the thesis makes a clear, arguable claim about the book
• Whether evidence comes from specific scenes and characters rather than general statements
• Whether each piece of evidence is explained and tied back to the claim
• Whether paragraphs follow a logical order and stay on topic
• Whether spelling, grammar, and citation errors follow a pattern worth flagging

Where teacher judgment still matters

Plot errors are a good example. A student who says the Formics are invading Earth during the story has misread the timeline, and you will want to catch that yourself. Nuance is another. A paper that argues Card wants readers to feel uneasy about Ender's victory may be doing sophisticated work that a rubric line only partly captures.

Tone and growth are harder to score too. If a student who usually writes flat summaries suddenly takes a real risk, you may want to reward it even when the execution is uneven. Use the AI draft to handle the routine parts, then put your own attention where a human read adds the most.

Setting up a novel unit for AI-assisted grading

Give the tool what you would give a new colleague covering your class: the prompt, the rubric, and a short note on what you expect. If you want at least three scenes cited, say so. If you accept both a sympathetic and a critical reading of Graff, say that too.

Run a few papers first and compare the drafts to your own scores. Where they differ, adjust the rubric wording, not the papers. After a round or two of tuning, most teachers find the drafts land close enough to review quickly.

Questions teachers ask before trying it

The most common worry is accuracy, and it is a fair one. Treating AI output as a draft, with a teacher signing off on every grade, keeps that risk manageable. Students also deserve to know how their feedback was produced, so many schools spell that out in their syllabus.

The second worry is fairness across students. A tool applying the same rubric to every paper can reduce some of the drift that comes with long grading sessions. Still, check results across student groups now and then, and keep the human review step in place.