She believes that is when AI is most helpful. With some tips, chatbots can provide immediate writing feedback for each student’s needs. A student may need to practice writing shorter sentences. Another might struggle with story structure and overview. In theory, AI can meet the individual needs of the entire classroom faster than human teachers.
In Meyer’s experiment, she inserted the AI only after the first draft was completed as part of the revision process. In a study published in 2024, she randomly assigned 200 German high school students and received AI feedback after writing a draft English paper. Their revised papers are better than the articles of 250 students who were also told to revise but did not receive AI help.
In the survey, those with AI feedback also said they were more motivated to rewrite than those who did not. This motivation is crucial. Usually, students are reluctant to rewrite, and without revision, they cannot become better writers.
Meyer doesn’t think her experiments prove that AI is an excellent writing teacher. She doesn’t compare it to student writing improvements after human feedback. Her experiment only compared AI feedback without feedback.
Most importantly, a dose of AI writing feedback is not enough to improve students’ writing skills. On the second fresh paper topic, students who received AI feedback before were no better than those without AI help.
It is unclear how many rounds of AI feedback will be needed to improve students’ writing skills more permanently, rather than just helping to modify the article at hand.
And Meyer doesn’t know if students are going to discuss writing with AI robots over and over again. Maybe students are willing to interact with it in this experiment because it is a novelty, but they will soon get tired of it. This is the next one on Meyer’s research agenda.
Virus MIT Research
A smaller MIT study published earlier this year echoes Meyer’s theory. “Your brain is on Chatgpt” is popular because it seems to say that using Chatgpt to help write papers makes students less and less brains. The researchers found that students who did not have any online tool to write papers had stronger brain connectivity and activity than students who used AI or consulted Google to search for original materials. (Using Google while writing is not as different from AI in the brain.)
Although these results make headlines, there are more experiments. The students who initially wrote a paper on their own later got chats to help improve the paper. The shift to Chatgpt enhances brain activity compared to what neuroscientists found during the initial writing process.
These studies add to evidence that after some initial thinking and drafting, a slight delay in AI may be the best place to learn. This is more testing needed by researchers.
Despite this, Meyer is concerned about providing AI tools to very weak writers and young children who do not develop basic writing skills. “It could be a real problem,” Meyer said. “It can be harmful to use these tools too early.”
Deceive learning methods?
Meyer believes that having students ask Chatgpt to write for them is not a bad idea.
Just as young artists learn to draw by copying masterpieces in museums, students may learn to write better by copying good writing. (The late great New Yorker editor John Bennet taught Jill to write in this way. He called it “copy work,” and he encouraged his journalist students to be legendary writers weekly by copying Longhand, not AI.)
Meyer advises students to ask Chatgpt to write sample articles that meet their teacher’s tasks and grading criteria. The next step is the key. If a student pretends to be his work and submits it, it is cheating. They also uninstalled cognitive work into technology, but learned nothing.
However, in theory, an AI paper can be an effective teaching tool if students study arguments, organizational structures, sentence structures, and vocabulary before writing new drafts in their own words. Ideally, if students learn through that analysis and internalize the style and technology of the model paper, the next task should be better, Meyer said.
“My assumption is that as long as I do cognitive efforts, as long as I have a lot of time to do the task, and like critical thinking about the output, it should be fine.”
Rethink praise
Everyone likes to praise. But too much praise can flood the learning, just as too much water can prevent the flowers from blooming.
Chatgpt tends to pour praise on a dense trend and usually starts with mediocre flattery, such as “Excellent work!” even if students’ writing requires a lot of work. In Meyer’s test of whether AI feedback can improve student writing, she deliberately told Chatppt not to start with praise, but to accept constructive criticism directly.
Her minimalist approach to praise comes from a 2023 writing study that prompts students to make revisions. The study found that when teachers started with universal compliments, students were doing well enough that they did not put in extra effort to rewrite.
In Meyer’s experiment, no complimentary feedback effectively allowed students to revise and improve the paper. But she didn’t compete directly between the two methods – no compliment and praise – satisfaction – so when students interact with AI, we’re not sure which method works more.
Praise the real teacher in the wrong way. After Meyer canceled the compliment from the feedback, the teacher told her they wanted to recover. “They want to know why the feedback is so negative,” Meyer said. “That’s not what they’ll do.”
Meyer and other researchers may one day solve the problem of how to turn an AI chatbot into a great writing coach. However, whether a student has the willpower or desire to give up an article immediately is another matter. As long as Chatgpt continues to allow students to take simple ways out, that is human nature.
Shirley Liu is a graduate student in education at Northwestern University. Liu reported and wrote the story with Jill Barshay of Hechinger Report.
