AI writing feedback for students in Writing, no kidding comes from a single tool: the AI Correction Tool (feature AI-002 in the platform's published AI Instructions for Use). It's worth understanding exactly what it checks and what it doesn't, since it's the first pass on every essay before a teacher ever touches the grade.

What the tool actually analyzes

When a teacher clicks Run AI Analysis on a writing submission, the tool reviews the student's text and returns:

• Inline corrections with type tags — each flagged issue is labeled as one of four categories: grammar, vocabulary, structure, or register. This is deliberately narrower than a vague "AI feedback" blob — you can see at a glance whether a student's main issues are sentence-level grammar or bigger structural/organizational problems.
• A feedback summary — a short written overview of how the piece reads overall.
• Strengths and weaknesses — specific things the student did well, and specific things to work on, rather than a single averaged impression.
• Prioritised action steps — a short, ordered list of what the student should focus on next, so feedback translates into something actionable rather than just a list of errors.
• A proposed score (0-100) — a numeric starting point for that writing task.

This is what makes the tool useful as automated grammar and vocabulary feedback and for structural comments together, in one pass, instead of teachers manually tagging each issue type by hand.

Inline essay corrections are a draft, not a verdict

Every correction the tool produces is meant to be read and adjusted, not applied blindly. Per the platform's own documentation, the AI may miss cultural references, intentional stylistic choices, or context only the teacher knows — and it does not know the student personally, their learning history, or any special circumstances. Some flagged corrections may simply be wrong or unnecessary. Review them before they reach a student.

The proposed score is a starting point, not a final grade

This is the most important limitation to internalize: the proposed score is a starting point, not a final grade, and it can be inaccurate — particularly for creative writing or highly contextual tasks where a formulaic reading of the text misses what the student was actually trying to do. The tool does not "understand" language the way a person does; Gemini 2.5 Flash is a probabilistic model producing statistically likely outputs, not a considered human judgment. Treat the number as a reference point you weigh alongside your own reading, never as something you forward to a student unreviewed.

Where this fits in the grading flow

The AI Correction Tool's output feeds into the broader review process described in How to Grade ESL Essays with AI — it's the analysis step that happens before a rubric score or composite grade is computed. Once you've reviewed the corrections and summary, the next step is deciding what grade to actually assign; see How to Review and Override AI Grades for that part of the workflow. If your assignment grades against a named framework rather than a custom rubric, see How to Use Grading Standards: CEFR and Cambridge for how that changes what descriptors the AI applies.

Related reading

• How to Grade ESL Essays with AI
• How to Review and Override AI Grades
• How to Use Grading Standards: CEFR and Cambridge