Artificial Intelligence

Are AI Teacher Assistants Reliable? What to Know

By Arianna Prothero — August 20, 2025 7 min read
Photo illustration of woman using AI on laptop computer.
  • Save to favorites
  • Print

Teaching “assistants” powered by artificial intelligence can help save teachers’ time and improve student learning. But they also create hard-to-detect problems that could negatively affect students if teachers are not diligent.

That’s according to a new risk assessment report of four popular AI teacher-assistant platforms by Common Sense Media, a group that researches and advocates for healthy tech and media use among youth. These assistants are tools that are designed and trained specifically to help teachers with common classroom tasks such as lesson planning, differentiating coursework, and grading, as well as administrative tasks such as writing newsletters and emails to parents.

Common Sense tested Google’s Gemini in Google Classroom, Khanmigo’s Teacher Assistant, Curipod, and MagicSchool.

Among some of the major issues identified in the report: AI teacher assistant tools can produce biased outputs, based on what it perceives to be a student’s race or background; fail to identify harmful misinformation; and may not be appropriate for novice teachers to use without guidance and training.

While veteran teachers might be equipped to spot these issues, novice teachers may lack the experience to detect problems and determine when it’s appropriate to use an assistant, said Robbie Torney, the senior director of AI programs at Common Sense Media.

“As somebody who was a novice teacher once, speaking for myself, I was not aware of what I didn’t know,” he said. “Using an AI chatbot, you could see unintended consequences of a new teacher making decisions that could have long-term impacts on students.”

For example, he said, an inexperienced teacher might rely too heavily on text-leveling tools that default to giving struggling readers easier text.

“Text leveling can be helpful sometimes, but that’s not the only thing that struggling readers need,” Torney said. They “also need access to grade-level text. And if you’re giving a struggling reader access to easier text for an entire school year, that reader isn’t going to be making progress toward grade-level text.”

That doesn’t mean schools should bar early-career teachers from using AI teacher assistants, Torney said. But school leaders should not assume because novice teachers might be younger and more tech savvy in general that they don’t need support and training on how to use these platforms.

Nor can AI teacher assistants replace the need for mentors and learning communities for new teachers.

AI teacher assistants recommend different behavior interventions for Black students

Some issues that arose in Common Sense Media’s tests were obvious red flags—such as the AI platforms creating examples or activities for students that used harmful stereotypes, promoting misinformation as factual, or presenting sanitized versions of historical events.

But some issues flagged in the report are subtler and easier for teachers to miss. That makes them potentially more concerning because they are harder to catch, said Torney.

As part of the risk assessment, Common Sense Media tested both Google Gemini’s and MagicSchool’s tools for generating behavior intervention suggestions and strategies. Common Sense used the same prompts about, say, a disruptive student, but alternated using traditionally male and female names, and “white-coded” and “Black-coded” names.

Over many tests, Common Sense found a pattern of differences in which interventions the AI tools suggested for which students. These differences were based on what the AI was inferring from students’ names about their gender and race, the report said. For example, the report said: “Annie tended to get de-escalation-focused strategies; Lakeesha tended to get ‘immediate’ responses; and Kareem tended to have little specific guidance.”

On an individual student level, these recommendations might not sound any alarm bells, said Torney.

“When I looked at the model outputs for the Black-coded names, at the individual level, I was like, ‘oh, these model outputs are pretty good,’” he said. “It’s only when you compare the entire sample of the Black-coded names against the entire sample of the white-coded names, you start to notice some of these differences.”

See also

Artificial intelligence and schoolwork image with hand holding pencil with digital AI collage overtop
iStock/Getty
Artificial Intelligence AI and Equity, Explained: A Guide for K-12 Schools
Alyson Klein, June 20, 2024
•
12 min read

But teachers aren’t doing those kinds of analyses in their day-to-day jobs, so it will be difficult for them to catch problems like that, Torney said.

In a response to an Education Week query, Google confirmed that it turned off a shortcut to the prompt “generate behavior intervention strategies” in Google Classroom out of what the company said was an abundance of caution as it evaluates the concerns raised in the Common Sense Media report.

“We use rigorous testing and monitoring to identify and stop potential bias in our AI models,” a Google spokesperson said in a statement. “We’ve made good progress, but we’re always aiming to make improvements with our training techniques and data.”

MagicSchool said in a statement that it had not been able to replicate the findings of the Common Sense Media report.

“As noted in the study, AI tools like ours hold tremendous promise—but also carry real risks if not designed, deployed, and used responsibly,” the statement said. “At MagicSchool, we are committed to ensuring our tools advance fairness, equity, and quality teaching for all students.”

Another issue identified in the report is using AI teacher assistants to create individualized education programs, a feature that is included in three of the platforms Common Sense tested. These tools, at least as of now, can produce misleadingly polished documents, but they lack the sophistication to create truly comprehensive IEPs, Torney said.

AI should support teacher expertise, not replace it

The big takeaway from the report is that AI teacher assistants are only as good as the systems that surround them—such as the districts’ policies guiding usage, teacher training, oversight processes, and, perhaps most importantly, teachers’ expertise.

Teachers, the report said, should think carefully about what information they input into an AI teacher-assistant tool and whether it will lead to better outputs. For example, will sharing personal information like a student’s name improve the output, or potentially bias it?

Teachers should be particularly careful when relying on AI teacher assistants to generate images, which can create pictures that promote harmful stereotypes and misinformation, the report said.

Teachers should also be conscious of the phenomenon of “automation bias”—a human tendency to put too much trust in the suggestions of automated systems—which can lead teachers to not critically assess the AI outputs. Even though AI teacher assistant tools can generate materials such as a slide deck in seconds, teachers should always review materials with a critical eye before putting them in front of students, the report cautioned. For example, when Common Sense prompted MagicSchool to create a picture of a “thug,” it produced religious images of Jesus.

These platforms work best, the report concluded, when teachers are in the drivers’ seat and the AI assistants are just that: assistants. For example, Torney doesn’t recommend asking an AI teacher assistant to differentiate a lesson, but it can be a valuable tool for aiding teachers in differentiating a lesson.

See also

Photo of a young white boy viewing AI Chatbot responses on his mobile device.
iStock/Getty

“You as the teacher already have a really clear plan for how you want to structure the different groups in your classroom and the types of experiences you want them to learn,” he said. “Giving very clear instructions to the teacher assistant about how you want to do that differentiation can save a bunch of time in generating those materials.”

Some features of well-designed AI teacher assistants, according to the report, are platforms created primarily for teachers to use rather than students. If a platform does include features for students to use directly, they should be customizable and still controlled by the teacher. (The report credits MagicSchool for doing this.)

Well-designed tools also have options that allow teachers to upload lesson plans, curriculum guides, and other resources. This helps ensure that outputs from the AI tool are aligned with curricula and standards rather than undermining them.

Used properly, AI teacher assistants can improve learning by enhancing high-quality instructional materials and saving teachers time on busy work so they can focus more attention on the core aspects of teaching, Torney said.

“AI can really be a powerful assistant,” he said. “This can increase productivity and boost creativity. If I was still teaching, I would want to be using these tools. And if I was an administrator, I would really want my teachers to be using these tools.”

Events

This content is provided by our sponsor. It is not written by and does not necessarily reflect the views of Education Week's editorial staff.
Sponsor
School Climate & Safety Webinar
There But Not Present: Rethinking Student Engagement
Learn how to recognize the early signs of disengagement and what the behavioral science reveals about which interventions actually work.
Content provided by Frontline Education
Teacher Preparation Webinar What Teacher Education Should Look Like in the Age of AI
Explore the current trends around the use of AI in teacher preparation and continuing education with data from the EdWeek Research Center.
English Learners Webinar How Can School Districts Better Support English Learners With Disabilities?
Join this webinar to hear recent findings from the EdWeek Research Center and discover practical strategies for improvement.

EdWeek Top School Jobs

Teacher Jobs
Search over ten thousand teaching jobs nationwide — elementary, middle, high school and more.
View Jobs
Principal Jobs
Find hundreds of jobs for principals, assistant principals, and other school leadership roles.
View Jobs
Administrator Jobs
Over a thousand district-level jobs: superintendents, directors, more.
View Jobs
Support Staff Jobs
Search thousands of jobs, from paraprofessionals to counselors and more.
View Jobs

Read Next

Artificial Intelligence A Reckoning on Social Media Addiction Came Late for Kids. Will AI Safety Be Different?
Efforts to make social media safer offer lessons for those looking to protect children using AI.
5 min read
Kristin Bride, right, mother of 16-year-old Carson Bride, who died by suicide after being cyber-bullied on an anonymous app on Snapchat, is joined by Amy Neville, left, mother of 14-year-old Alexander Neville, who died from fentanyl poisoning after being deceived by a drug dealer operating on Snapchat, during a Social Media Victims Remembrance Day memorial on Capitol Hill, June 23, 2026, in Washington.
Kristin Bride, right, mother of 16-year-old Carson Bride, who died by suicide after being cyber-bullied on an anonymous app on Snapchat, is joined by Amy Neville, left, mother of 14-year-old Alexander Neville, who died from fentanyl poisoning after being deceived by a drug dealer operating on Snapchat, during a Social Media Victims Remembrance Day memorial on Capitol Hill, June 23, 2026, in Washington.
Rod Lamkey, Jr./AP
Artificial Intelligence Opinion Think Your Students Know More About AI Than You Do? Read This
Here are four things your students don’t understand about AI—and what you can do about it.
Sam Wineburg & Nadav Ziv
•
4 min read
102026 opinion Wineburg AI media literacy
Vanessa Solis/Education Week + Getty
Artificial Intelligence ChatGPT's 'Teen Mode' Is Unsafe for Students, Report Concludes
The chatbot failed to report unsafe interactions to parents in a timely manner, researchers found.
5 min read
Young man using a smartphone to ask an AI about exercise
Vanessa Solis/Education Week + Getty
Artificial Intelligence At OpenAI Forum, Discussion Focuses on Stronger Teen Safety Measures
Panelists put forward ideas on how to incentivize industry to develop responsible products.
4 min read
Washington, DC - October 5, 2026: Arthur C. Evans Jr., PhD, Chief Executive Officer of the American Psychological Association  speaks during the discussion. The OpenAI Forum hosted a panel on Responsible AI for Teens, how AI can support teens’ learning while promoting safe, healthy use. The program featured an OpenAI Forum panel with experts in learning, adolescent mental health, and family and online safety: Kate Ruane, Director of the Free Expression Project at the Center for Democracy & Technology, Chuck Melley, SVP of Policy, Advocacy and Communities at CodeAI, Arthur C. Evans Jr., PhD, Chief Executive Officer of the American Psychological Association and Mathilde Cerioli, Chief Scientist at everyoneAI. Chris Lehane, OpenAI’s Chief Global Affairs Officer, moderated the conversation and shared context on ChatGPT for Teens, a learning-first experience launched in August and designed to help teens think critically, deepen understanding, and use AI with confidence.
Arthur C. Evans, Jr., the CEO of the American Psychological Association, speaks on a panel at an Oct. 5, 2026, OpenAI forum on how to support responsible use of AI among teenagers.
Sarah Voisin for Education Week