The math problem had Palo Alto High School sophomore Brian Chung stuck for over an hour. It was 11 p.m., his Analysis Honors exam was mere hours away, and that answer wasn’t clicking — until he reached for his phone, took a screenshot, and let Claude be the key that finally opened the door.

Chung is not alone. According to a May 2025 College Board survey, 84% of high school students reported using generative AI tools to assist with schoolwork and study for tests, up from 79% in January.

As admissions rates at selective universities continue to decline, a growing wave of AI-powered study platforms have been reshaping how students prepare for tests. In Palo Alto, home of Stanford University and Silicon Valley, for the many who aspire to be admitted in selective universities, the pressure to perform well on standardized and school tests has never been higher. 

“It [college admissions] is getting more competitive, and there’s a lot of pressure to get good grades,” Chung said.

In this environment, any new tools promising a competitive edge quickly gain traction.

The Phenomenon

When Paly computer science teacher Kathryn Widen was in college, she studied using textbooks and old materials. Now, she says she is noticing the effects of that shift in studying methods in real time.

“Students have said they’ve used AI for generating extra practice problems if they’re struggling with that concept,” Widen said.

Chung said that Anthropic’s Claude chatbot appeals to him because it helps him understand advanced concepts.

“[For] the concepts I didn’t know, I screenshotted some images into Claude, and I prompted Claude to explain some conceptual stuff and create questions, like multiple choice or free response,” Chung said. “AI just helps me understand better.”

Chung also said that he uses AI tools to study the SAT.

“I go through a bunch of practice tests, then go over the questions I get wrong,” Chung said. “If I still don’t get it, I just prompt it into Claude.”

Paly senior Theresa Wang said she uses AI tools to study for her AP Physics C class. 

“The last time I used AI was to help me with my physics homework, so like going over problems and targeting concepts,” Wang said.

Wang said that she enjoys using OpenAI’s chatbot, ChatGPT, because of its ability to act as a tutor.

“I usually use ChatGPT … I think it’s good at giving personalized answers and being a tutor in place of an actual human in specific contexts,” Wang said. “So for example, when you already partially know the answer, but you just want a bit of confirmation.”

Chung added that AI can have an advantage over Google or Youtube because its ability to give personalized answers allows it to answer all types of questions.

“Even if I search it up on Google or YouTube, there might not be an answer. But AI can literally prompt it specifically, and can still give me an answer.”

Beyond general-purpose chatbots like ChatGPT or Claude, Chung said he uses Turbo, an AI study tool designed specifically to create notes, summaries, flashcards, and quizzes marketed to help students prepare for tests in academic subjects.

“When you put a YouTube link it just gives you a whole summary of it,” Chung said. “I’ve used that before and it’s been pretty helpful when I didn’t really have any time and just needed a quick summary.”

Chung noticed that AI study tool companies, like Turbo, have increasingly turned to social media platforms to market high school students, running advertisements featuring students who claim that these specific tools helped them improve efficiency, understanding of concepts, and test scores on exams.

Dangers

However, doubts regarding accuracy have been raised about Turbo and similar AI tools. Turbo struggles with solving higher-level complex subjects and often oversimplifies nuanced topics, AI consulting firm Christopher Queen Consulting reported.

As students resort to using AI tools that provide them with information to complete assignments or study for tests, the concern of inaccurate information has become deadlier, especially as each individual test score gains more weight.

Wang said that while AI is useful for conceptual review, it struggles with calculations and should not be trusted without human verification.

“I think it [AI] is good at reviewing concepts, but I would be careful with the calculations,” Wang said. “It always gets it wrong. I think AI is quite poor when it comes to basic arithmetic … that’s why I think it’s always a good idea to double check your own work and just make sure it sounds right.”

Hallucinations

According to the University of California San Diego Computer Science professor Charles Elkan, who studies machine learning and data science, AI hallucinations refer to when an AI model produces an output which is believable, but not actually correct.

“What these AI models are doing is they’re starting with the contacts that you provide as input, and then they’re looking for the next word, which is high probability, high likelihood, according to all the patterns that the model has learned from its training data,” Elkan said, “Then it’s looking for the second next word, the third next word, and so on. And so the output is always something which is high probability, but sometimes things with such a high probability are still not actually correct.”

Wang said that she frequently encounters AI chatbots generating false or unsourced information.

“It [AI hallucinations] is very common,” Wang said. “When it [AI chatbots] gives you results for something, it doesn’t tell you what the source is, or the source is completely fake.”

Over-reliance

Widen also raised concerns about students using AI to fully complete assignments, which she said ultimately detracts from learning.

“I see a lot of text generation for just generating an answer that they can copy into a document,” Widen said. “The class that I see it the most in is my AP class … it’s really important that they try and fail now, especially because there’s opportunities to revise versus having ChatGPT solve it for them.

Wang warned that artificial intelligence poses risks to critical thinking and has strained the relationship between students and teachers.

“There’s definitely a danger with critical thinking here [use of AI],” Wang said. “AI caused a bit of a trust issue between students and teachers … nowadays it can be hard to figure out if the student has used AI or not.”

Widen said generating practice problems and receiving step-by-step guidance are valuable functions of artificial intelligence, but added that educators have concerns about misusing the technology.

“Generating practice problems is awesome, Widen said. “Walking you step-by-step through how to do something for a different problem is great. Those are all really useful tools … but when your teachers don’t want you using AI, it’s because we see students using AI as a shortcut that is ultimately preventing learning,”

Looking Forward

For now, AI tools remain a pivotal part of many Paly students’ study habits, while their accuracy of the information they produce remains imperfect. 

As college admissions grow in competitiveness, and each test score carries more weight, finding a balance between efficiency, reliability, and dependence is something students, educators, experts and parents must figure out how to address together.