Three AI Essay Feedback Tools: My Honest Review After a Semester

At the end of last semester, I let out a sigh as I set down a stack of Bumi Manusia essays submitted by thirty-two 11th-grade students on my desk. It's a scene that repeats every semester, but this time I wanted to try something different. I picked three free AI writing feedback tools, ran them simultaneously on ten student essays, and compared the results with my own red-pen feedback.

Here are the tools I tested: Grammarly (free version), ChatGPT (free access based on GPT-3.5), and Quillbot's grammar check feature. The selection criterion was simple: all three could be used at no cost. Running multiple paid subscriptions on a Bali public high school teacher's salary simply isn't realistic.

Grammar and Structure: Where Machines Do Pretty Well

The most common mistakes students make when writing essays in English are subject-verb disagreement and tense mixing. Grammarly exceeded expectations here. In eight out of ten essays, it caught roughly 70% or more of the grammar errors I had marked. Quillbot performed at a similar level, and its sentence restructuring suggestions were actually more varied.

ChatGPT took a fundamentally different approach. When asked to "find the grammar errors in this essay," instead of correcting line by line, it tended to rewrite the entire flow. This can be useful at times, but it's not well suited for helping students learn what they need to fix in their own writing.

Does It Understand the Context of Indonesian Literature?

This is where all three tools showed their limitations. One student, while analysing Pramoedya Ananta Toer's narrative structure, used the expression "hierarchical language of Javanese feudal society." Grammarly flagged this sentence as "unclear." Quillbot went so far as to suggest replacing "feudal language hierarchy" with "language barrier." That is exactly the kind of misfire you get when context is missing.

ChatGPT was relatively better here. When given an additional prompt asking it to explain the background of Pramoedya's works, it provided fairly accurate context about the social structure of Java during the Dutch colonial era. However, writing separate prompts like this for each of the ten student essays took even longer than writing comments directly with a red pen.

Tone of Feedback: Is It Appropriate to Hand to Students?

This was the aspect I cared about most. When a machine spits out "Your argument lacks coherence and depth" to a 17-year-old student, that is not feedback — it is a verdict. Grammarly was neutral in tone, and Quillbot had almost no tone at all — it simply listed correction suggestions.

When I instructed ChatGPT to "give feedback in an encouraging tone for a high school student," the comments that came out were quite decent. Things like, "You made an interesting point about Minke's identity — try expanding it with a specific scene from the novel." I actually showed this to two students, and the response was not bad.

So Is the Red Pen Retiring?

Not yet. All three tools saved time on English grammar correction, but providing feedback on literary interpretation and argument structure — the core of an Indonesian literature class — remains the teacher's job. The practical approach I settled on is this: have students write a first draft and polish the basic grammar with Grammarly first, then I personally comment on the content and argumentation.

On a rainy afternoon in Ubud as the wet season begins, I still wrestle with stacks of essays while listening to the rain outside the staff room window. The only difference now is that I spend a little less time drawing red lines under grammar errors.