We recently conducted a survey and one-on-one interviews with students from different colleges to understand the role of AI in tutoring and academic support.

How are students using AI for academics today? How do they compare AI to a campus tutor? How are these trends likely to change in the future?

We conducted this research to help our learning center partners stay ahead of the evolving support landscape, as AI represents a significant shift.

The conversations were interesting, to say the least! This article aims to:

1. Share the takeaways from the research, including clips from the interviews
2. Predict the near-term future of AI performance for tutoring
3. Outline how learning centers can thrive even as AI tutoring becomes increasingly prominent

Summary

We collected over 150 responses on this survey and interviewed six students from four different public and private 4-year colleges. We found that students:

1. Use AI multiple times per week on average for schoolwork
2. Don’t trust AI to give them correct answers, otherwise they’d use it even more
3. Have already decreased their usage of campus tutors due to AI
4. Emphasize the value of the human component of their tutoring center

As AI improves, I expect a gradual but significant reduction in learning center traffic from those just seeking quick answers. While that may be a bit scary, we also found that students value the human component of learning centers deeply. Investing further in quality experiences can offset the decline in “answer machine” traffic and build a stronger center that fulfills a student need far into the future.

Student Interviews

Penji serves learning centers with student-friendly scheduling software, and we asked some of our partner centers if we could interview past students. While this pool of students does have some selection bias towards motivated and outgoing students, we felt it was important to get students who could compare tutoring and AI usage side by side.

Distrust of AI for Academics

All six interviewees agreed that AI will often give incorrect answers, notably to math problems. STEM majors felt this strongly, to the point of not even using AI due to this reputation.

The non-STEM majors we spoke to (marketing and real estate), however, were very pro-ChatGPT, and felt the AI was reliable enough to use heavily.

I expect that the distrust of AI will go away as the AI models improve. AI will become near-perfect for most undergrad subjects, including difficult courses like upper-division engineering - see my demo later in the article - so we need to start thinking with this assumption in mind.

How do you compare AI to campus tutors?

If AI was always right, would you reduce your usage of campus tutors?

Megan felt trustworthy AI would lower her usage of human tutors by about 50%.

Miranda has already taken tutoring usage down to zero because AI can answer the majority of her marketing major questions correctly.

Alexa felt similarly and uses it heavily, preferring AI to a human for the comfort, lack of judgement, and convenience.

Luke, Daniela, and Bennie didn’t seem to think trustworthy AI would affect their tutoring usage all that much, as they were motivated to attend by the human component.

In all, there was a mixed bag of comments indicating that (1) AI definitely is useful to all students in some ways, but also that (2) there is real value in the human component that at least some students will continue to be drawn to.

Value in the Human Component

Multiple students brought up the back and forth that occurs in a tutoring session with another person. The person learns about you, remembers things you’ve done in past problems, and is able to quickly get to the bottom of your misunderstandings.

There was also repeated mention of “writing on the whiteboard” or “pen and paper” across different interviews. There seems to be value in the physical medium shared with a real person. Megan’s clip above mentions that, as does Luke’s just below.

There was also some mention of the emotional support one gets from a person who’s been through it before, but I was surprised that longer-term relationships with the tutor were not a key piece for any student. Instead, the focus was on the comfort of the human interaction, and the benefits of verbal conversation with a person.

Contrasting Takes on Human Value

While most students agreed that the human component has value, one student felt strongly that the main goal is to learn, and AI can help them do this much faster and more conveniently: “The human component I can get elsewhere”. This was the marketing major who feels school is easy, so perhaps the higher quality of in-person learning becomes necessary in more difficult subjects.

Alexa, the real estate major, also wasn’t very interested in the human component. She felt that in some ways, AI is actually more comfortable, as “you can ask it lots of questions and it will never get annoyed”, and that “I went in for an Econ tutor and they were all very rude”. So, certainly possible to have negative experiences with tutors.

Overall Feelings on AI Trend

There was disillusionment amongst these students about what AI is doing to society. I received quite a few comments about this, as these students see their friends leaning on AI heavily and not learning to be resourceful. This does seem to be selection bias as these interviewees felt in the minority for being cynical on AI, and they were the ones who opted in to a call with me.

One interesting quote from our written survey:

"I believe it does more harm than good. It takes the humanity and creativity away from modern life, through art, written papers, emails, texts and advice. I'd be lying if I said I don't also use it. It tends to be my last resort when I'm struggling with an assignment or understanding a concept after scouring online resources. It's been super helpful, but I know if I overuse it I am stripping away my ability to problem solve."

Survey Results

With our written survey on AI usage, we received over 100 responses, and many of the themes mentioned above were confirmed and expanded upon.

In general, how do you feel about AI?

116 responses

‍

How frequently do you use AI for homework, studying, or other academic purposes?

117 responses

‍

How useful is AI for getting quick answers to homework or study guide problems?

118 responses

‍

How useful is AI for deep, meaningful learning?

119 responses

‍

How important are the following factors in motivating you to use the campus tutoring center?

‍

Has AI caused you to reduce your usage of campus tutors?

121 responses

Assessing AI’s impact on learning center usage

The value propositions of learning centers typically include:

1. “Answer Machine” - Students just scraping by, using a tutor last minute to get a quick answer because other methods aren’t working for them
2. Quality Learning - Students who are focused and motivated, consistently using tutoring to build skills and understanding
3. Social - Students who are motivated to meet and interact with their peer and enjoy the group experiences in tutoring
4. Role Models - Students who view tutors as great people to spend time around and model their behavior after

Let’s take a guess at the mix of motivations to visit the tutoring center, loosely based on our survey results above:

I believe AI will step in for 100% of the “answer machine” value proposition over the next five years as models get near-perfect in quality and students trust them more. This 35% slice represents a significant portion of learning center traffic - we can disagree about the exact size of the slice, but certainly there are many students in this group. AI is a direct fill in here.

The other slices, however, are likely durable. Human tutors will provide the highest quality, and obviously will win on the human components.

The reduction in traffic from the “answer machine” slice won’t happen overnight, and shouldn’t be viewed as a crisis, but I believe it’s important for learning centers to act now to improve elements that will provide long term value to students and protect your usage (and budgets).

Before moving on to recommendations, will AI actually get good enough to take over this “answer machine” slice?

Will AI ever become a good enough “answer machine”?

Student perceptions currently seem influenced by past experiences with older AI models that have since evolved. As AI models continue to improve, perceptions will catch up, and student trust in the models will build, eliminating one of the top objections to using them instead of tutors.

For example, Megan saw incorrect information on (1) Google’s AI search summaries (which used an old model until recently) and (2) Chegg’s AI, which likely was not state of the art either. These interactions convinced her that AI was untrustworthy.

However, each new AI model generation brings progress. Just this week, Claude 4 released with chart-topping benchmark scores. We are in an AI arms race with many billions being spent on improving them.

Below is a demo of using Claude 4 (released this week) and Google’s Gemini 2.5 model (released in March) for a Heat Transfer problem that I, at one point in my life, understood how to do.

Long story short, these new models were very close to getting a difficult engineering problem 100% right. The small discrepancy in the answer represents a problem that has plagued AI models for some time - basic math - but that the current generation of models is working to eliminate. It seems likely that these calculation errors will be eliminated soon, and that student trust will increase accordingly.

So what should learning centers do?

This discussion of reduced traffic may sound scary, but I believe this will just force learning centers to double down on their unique strengths. Tutors are not meant to be “answer machines”. We are being forced to focus on deeper learning value from the simple transactional tutoring that is too easy to lean on.

The mission is now to double down on the unique value propositions learning centers can offer, become better at highlighting your excellent staff, and build a smooth experience for interacting with those staff.

To continue with the business analogies, it is common that fundamental shifts in “the market” will occur, and it is up to businesses to adjust. This can be hard but is also a natural force to improve, and this will end up being good for students.

Check out Crystal from LA Pierce College discussing this:

Student-Friendly Scheduling

So, how can you work to improve? A good first step is making sure students can easily access your services, for example you can watch a new student try to book an appointment online and see where they get stuck. With the Penji scheduling platform we consistently see growth in traffic after launching a new partner because we focus on making the most student-friendly interface. Improving your student-facing experience can offset downward pressure from AI.

The Learning Center Funnel

I’ve also written a few articles with actionable advice on improving your center:

• Analytical Thinking at Learning Centers outlines the three-piece “learning center funnel” framework for measuring and improving your center
• Funnel Step 2: Activation focuses on a smooth initial experience for students using your center
• Funnel Step 3: Re-Engagement focuses on maximizing comfort of student visitors and the quality of your tutoring sessions

Beyond these, I do think that a human tutor will offer the best ceiling for high quality learning, always. This is a protected advantage due to the physical presence, the chance for positive human connection, and the unique back and forth only possible in real conversation. See Alexa discussing this.

Here are a few bonus items from recent conversations I’ve had with directors.

Social Events

I’ve been hearing more and more about monthly game nights at learning centers. This is a chance to truly come and be social, to mingle students with tutors in a format that they can actually become friends instead of rushing through a session. Attendance can be pretty good. Get pizza if you can.

A Strong Session Structure

To get ChatGPT to actually teach you well, you have to prompt it the right way. I guarantee the majority of students are not prompting it well.

Your tutors, on the other hand, offer the service of having a pre-defined structure that is intended to maximize benefit for the student. The student has to do no prompting.

The tutor will set learning goals, will set post-session goals, will encourage booking followup sessions at the end of a session, and will observe trends in the student’s thinking that the AI will miss.

They will also offer empathy for the struggle, and confidence that if they did it, then the student can do it too.

Iterate Based on Feedback

These AI systems are generic, computer-based things that can’t be tailored to your environment.

If you ensure you have strong feedback loops (post session forms, semesterly surveys, and act upon the feedback), you will create a more tailored service that feels fundamentally different than using AI.

Wrapping Up

From these interviews, it appears that AI will change the way learning centers must position themselves on campus. While this may end up being a positive, it does necessitate action and further investment into strengthening your center’s experience for students.

AI represents the biggest shift to student learning patterns, probably ever, and we’re only just beginning. Shifting your focus to featuring the human experiences you offer - and optimizing your start-to-finish experience - will position your center well for the future.