5 ‘Good’ AI Use Cases That Are Actually Bad for Students

20 years ago, when Dr. Rachel Bedard was a student navigating medical school, she kept a well-worn copy of Pocket Medicine tucked away in a pocket of her white lab coat.

At the time, the analog support was standard-issue, placing evidence-based clinical guidance at her fingertips. “I consulted it all day long,” she recalls in a recent New York Times opinion piece. “I had to understand exactly what I was looking for when I flipped through its pages, and it could not tell me what to do with whatever knowledge I found there.” 

Today’s medical students have access to something slightly more sophisticated: OpenEvidence, a chatbot described on its website as an “AI copilot for doctors that helps them make high-stakes decisions at the point of care.” Despite noting that its diagnoses are not always “useful or even accurate,” Bedard has no qualms about experienced doctors using the tool. In fact, she believes access to the platform is making her a better physician: “My patient care has improved as conversations with the agent push my thinking beyond its usual limits,” she writes. 

The same can’t be said, she worries, for future doctors-in-training. 

While the app’s “speed and anticipatory helpfulness” might be a welcome “enhancement” for seasoned medical practitioners, the risk that students will uncritically follow the bot’s authoritative output—without interrogation or skepticism—jeopardizes the health and wellness of patients in their care. “When I skim a chatbot’s response to a question about back pain, I’m intuitively testing its suggestions against my hard-won mental model for how to approach back pain,” Bedard argues. “Could I have developed that if I’d had a robot leading my thinking all along?”

STUNTED SKILLS

The question of what foundational skills students can (or will) develop with generative AI riding in the passenger seat is both murky and consequential—not just in medical school, but across all of K-12 education. Mounting evidence underscores the risks of unsupervised student AI use, though key questions around potential productive use cases remain unanswered. “We have no idea the best way to teach a kid to use AI to develop as a writer, or a thinker,” explains Justin Reich, an MIT associate professor and director of the Teaching Systems Lab. “At some point, we will have more research to help us say, ‘This is a good age to start introducing AI tools, and these are some good tools and approaches to use that don’t replace key skills.’”

Rather than wait, New York City’s Department of Education announced policy that restricts elementary and middle school students from using AI inside classrooms—a move echoed by the Los Angeles public school system. But inside many schools across the country, school leaders defer to the discretion of individual teachers regarding how and when AI might be used by their students, if it’s used at all. That’s a degree of responsibility some educators feel unprepared to navigate without clearer guidance and support: “I’ve interviewed nearly 100 teachers about AI in schools, and they unanimously say they need policy and professional development,” Reich notes.

With an aim of preparing students for the workforce, a popular usage model has emerged, writes cognitive scientist Daniel Willingham on his Substack: Student as the lead thinker, and AI as the capable assistant. While a lead/assistant positioning is reasonable “for life after graduation,” Willingham acknowledges, this philosophy is far less optimal for K-12 students. In a classroom, the product—whether it be a worksheet of solved math problems or an essay—is not the point. What’s at stake is the development of the foundational skills required to complete them, Willingham argues: “The point of the mental processes is learning.”

Before allowing students to consult an ever-obliging AI assistant, it’s worth considering whether they have mastered the core mechanics of the task at hand. When deciding if your students should use AI to help them brainstorm or conduct research, for example, “a teacher should ask whether doing the task themselves is still part of what students are meant to learn or practice,” Willingham suggests. 

Here, we examine a number of scenarios where students using AI is widely viewed as harmless (or even helpful), yet the available research and thinking of leading experts suggests each instance deserves a much closer look.

BRAINSTORMING

Behind every good idea are a thousand false starts. That was the conclusion of the early 20th century advertising executive Alex F. Osborn, a few decades before the Madmen era, who found that his staff members, paid to generate brilliant slogans, often produced more bad ideas than good ones. Every day, it seemed, he filtered through the mess and pulled out words and phrases that could be fashioned into award-winning advertising campaigns.

To increase his odds of striking gold, Osborn invented what his employees later called "brainstorming," a creative activity whose premise conceded the limits of the human imagination: Employees gathered to produce dozens of half-baked ideas in collaborative, judgment-free working sessions. Fragments from a multitude of bad ideas, Osborn discovered, carried the seeds of genius within them; human discernment allowed them to be combined into something extraordinary.

That may explain why brainstorming for creative tasks like fiction or essay writing has been so frequently mentioned by early adopters of AI as tasks better suited to machines. Indeed, having no concept of shame and no limits to their work ethic, LLMs brainstorm with astonishing brio and efficiency. In a 2025 study, researchers concluded that humans produced more creative suggestions for open-ended problems like “designing toys from everyday objects” and “repurposing household items” when aided by ChatGPT. A 2026 study found that “access to generative AI” to brainstorm produced short stories that human evaluators scored as “more creative, better written, and more enjoyable.”

But across both studies, AI brainstorms generated suggestions that were also far more predictable. Truly unique ideas—the stuff that Osborn and his teams turned into groundbreaking material, and the stuff students should aspire to—tended to be missing from the brief, hypnotic bursts of language that followed a prompt.

One job of humanities teachers is to provide—and protect—the space for students to express themselves authentically, says high school English teacher Chanea Bond. In her classes, brainstorming is a foundational skill that teaches valuable lessons: The task requires students to sit patiently within the messy realm of nascent thinking; to use their emerging taste and discretion to identify promising ideas; and to sequester and refine the best or most surprising notions. "While some teachers say that AI can serve as a brainstorming partner or an outline generator," she wrote in a prescient July 2025 article for Edutopia, "I believe it’s so important for students to learn these skills themselves."

The research is increasingly clear about what happens when these crucial activities are outsourced to computers. In a 2026 study, for example, researchers concluded that participants with access to chat AI were highly creative. When the tools were later withdrawn, however, the creative skills quickly faded, and signs of long-term creative atrophy emerged. Use of the AI tools to spur creative thinking, the researchers concluded, had left a “scar” that may have permanently altered the “creative trajectory” of its users.

THESIS WRITING

Like brainstorming, thesis writing is often thought of as "pre-work"—preparation for an argumentative or persuasive essay, but not the essay itself. Writers may convince themselves that a set of thesis statements generated by AI on Shakespeare’s Sonnet 18, commonly known as “Shall I Compare Thee to a Summer’s Day?,” will merely allow them to tap into their existing knowledge of the poem. 

Asking Google’s Gemini to provide five thesis statements for a 1000-word essay, however, reveals patterns that the research predicts: Three of the resulting thesis statements cluster around a single theme—the poem is not “a declaration of romantic love” as much as a way for Shakespeare to “immortalize his own literary genius”—revealing the software’s preference for fixating on one or two themes rather than trying something surprising. Several of the theses are ambiguous and represent superficial readings of the language. Nothing in the mix is remarkable.

Good thesis statements, meanwhile, emerge from deep thinking about a subject, and distill that thinking into a provocative, focused statement that a writer is ready to defend. Because thesis statements are the end product of deep, personal reflection about how and why the subject matter is truly important, they are unique by nature and are not a separate process from the writing.

When Bond gave her students “the option of writing their own literary analysis thesis statements or feeding their notes to AI and asking it to write a thesis statement,” kids who used AI wrote terrible papers. “I realized there was a problem,” she said: “my students don’t have the skills necessary to be able to take something they get from AI and make it into something worth reading.”

CONDUCTING RESEARCH

Teaching research skills to students is “one of the most important jobs of an educator,” writes high school teacher and librarian Benjamin Barbour, but the mere mention of the word elicits groans from students. They’re not alone. “As adults, when we are told to do our own research, we don’t always jump for joy,” writes Sarah Said, an English teacher for English language learners. That makes it a particularly attractive target for AI assistance.

Some contend that AI enables students to retrieve information faster and provides a jumping off point, but the tool’s credibility remains open to question. “Let's say you want to ask ‘Who are the most important 18th century poets in the English language?’” explains Robert Malkin, a professor at Duke University and founder of the International Research Institute of North Carolina. “One of the problems with asking AI that question is AI might just go to any random website and just pull the list.” At least in its current iterations, Malkin protests, “AI is terrible at selecting these sources.” 

What’s more, to a student still developing the ability to carefully and critically evaluate, a list of sources—credible or not—may look like flat terrain. At best, the highly motivated and skilled pick a few to investigate more deeply; more often, students may choose arbitrarily or ask AI to choose for them. “I think it’s asking a lot of students—or better, asking a lot of human nature—to expect them to take a much harder mental path,” Willingham says, “even if it is in the interest of their own learning.”

Nevertheless, use is outpacing guidance as more than half of high school students are already employing AI to “conduct research and find sources,” according to a survey from the College Board. While initially skeptical of students adding AI to their research toolbox, Barbour now counts himself among the (partially) converted. He’s warming to the idea of his AP Research students using AI to “find the weaknesses and deficits in one's argument,” for example, or help develop the seeds of a research plan they’ll continue to grow themselves.

Whether students leverage AI tools or not, Barbour ensures they know what’s expected of them, as well as what’s at stake: “‘You will have to present information on this topic verbally in front of an audience. You will be asked questions. You need to be able to speak fluently about the research,’” he tells his students. “If AI does everything, the student never learns an adequate amount about his or her research to present the information satisfactorily.”

SUMMARIZING LEARNING

It’s an unfortunate reality of all learning: “Without reinforcement, we forget almost everything we learn,” writes high school English teacher Cathleen Beachboard. “Unless an effort is made to retain new information: 50 percent of new information disappears within an hour, 75 percent is gone by the next day, and up to 90 percent fades away within a week.” 

When studying, there’s a common student misconception that rereading your notes or a relevant text is enough to help encode the information into long-term memory. Students might snap a photo of their notebook and create polished AI-generated summaries to review, for example, because they appear superior to their own notes. But regardless of the quality of the summary, reading it is “passive in nature” and often provides students with a “false sense” of security that they’re comprehending something more clearly, according to the researchers behind a 2018 review of decades of insights on studying techniques. 

Effective studying is effortful, and requires an element of active retrieval—challenging oneself to recall then rearticulate what they’ve learned on their own. Asking AI to create the summary removes the real work of reorganizing the material from memory. In a 2021 study, students who wrote a short summary about a challenging text in their own words vastly outperformed their peers who simply read a summary that was provided to them. 

There are exceptions: Asking an AI tutor to create a summary of a topic like photosynthesis that can then be compared to class notes allows a student to surface gaps in understanding as they review—but that review must ultimately hinge on effortful retrieval, forcing the learner to actively recall and reorganize the material from memory.

WRITING FIRST DRAFTS

For generations, students have struggled to walk the tightrope between taking their raw, unformed thoughts and turning them into polished prose. The act of staring down the dreaded blank page is further complicated by factors like passion for the topic and mastery of relevant vocabulary.

Just as students solve complex math problems using pen and paper rather than solely in their heads, a similar inclination may lead students to conclude that AI can help them write essays or stories without compromising the quality of the work. Research disagrees: According to a 2025 study, even after high school, there’s a tendency to blur the line between using AI as a sounding board or a stand-in: When given access to ChatGPT as they worked on writing assignments, college students mindlessly “follow[ed] the thinking” of the tool, producing “statistically homogenous” essays that felt “soulless.” Habitual users who asked AI for assistance on tasks like drafting text for writing assignments “had science scores an average of 20 points lower than students who said they never used AI—equivalent to about a year of schooling,” recent reporting in The Wall Street Journal uncovered.

Students can and do produce great writing in classrooms; the trick, according to high school English teacher Jen Roberts, is getting them to stick with it. Periodically, to speed up the feedback process, Roberts allows her students to submit drafts into tools like Brisk or MagicSchool AI—a tactic that often spurs kids to take a second or even third swing at the draft, instead of waiting for her. A student “gets some feedback, and he goes back to writing,” Roberts says. “Over time, instead of spending 10-15 minutes writing, they spent 40 minutes writing that period.” 

Roberts understands the tradeoffs, and prohibits the use of AI when she uses peer writing groups, for example. “Writing groups have been a huge indicator of authenticity for me,” she says. “A student will happily turn in AI writing to me, but they will never put AI writing in front of their peers.