AI Eroded Faculty Trust in Students. Most Say the Relationship Survived.

Almost nine in ten faculty say AI has hurt their confidence that student work is real. Fewer than half say it has hurt the relationship. That gap is the part you can still use.

Those numbers come from a new Inside Higher Ed and Hanover Research survey on faculty trust in students, published today. The survey covers much more than AI, including politics and self censorship, and that part of the story is worth reading on its own. This post stays on the AI section, because it is the one that changes what you do before your next stack of final papers.

Half of the faculty who answered named the increased use of generative and agentic AI as a top concern about higher education, the largest share of any concern reported, ahead of federal interference at 45 percent. Then the survey asked what that AI use has done to the work between professors and students.

What the survey found about faculty trust in students

Inside Higher Ed asked about the effect of generative or agentic AI on five parts of the relationship with students. Here is how the answers came out.

• Confidence that student work or communication is authentic: 44 percent very negative, 44 percent somewhat negative, 3 percent positive.
• Sense of student engagement in coursework: 28 percent very negative, 51 percent somewhat negative.
• Trust between you and your students: 17 percent very negative, 50 percent somewhat negative, 29 percent no effect.
• Overall relationships with students: 5 percent very negative, 35 percent somewhat negative, 53 percent no effect.
• Ability to give timely or individual support: 17 percent negative, 65 percent no effect, 18 percent positive.

The doubt crosses disciplines. Arts and humanities faculty reported a negative effect on their confidence in authenticity at 92 percent, the social sciences at 90 percent, STEM at 86 percent and law or business at 82 percent.

One honest caution before going further. Hanover emailed 100,296 faculty members from September 21 to October 8 and received 1,665 full or partial responses, which is under two percent, and people who answer a survey like this may feel more strongly than people who do not. Inside Higher Ed reports a margin of error of 2.4 points. Read it as a strong signal from the faculty who answered, not a census of the profession.

The gap between the paper and the person

Read the third and fourth lines of that list together. Two out of three respondents say AI has damaged trust with students. Yet 53 percent say it has had no effect on their overall relationships with those same students.

Paul LeBlanc, former president of Southern New Hampshire University, described that split to Inside Higher Ed as “a testimony to faculty’s ability to separate these two things out.” He put the faculty view this way: “I’m not trusting what I’m seeing in the work that my students are submitting, but I still love my students.” José Antonio Bowen, a senior fellow at the American Association of Colleges and Universities, said something close: “Faculty mostly like students and we want them to succeed.”

The doubt has a real source. A University of California, Berkeley paper reported by The Wall Street Journal in May found A grades growing faster in classes heavy on writing and coding since ChatGPT arrived. Its author, Igor Chirikov, said the results suggest students relied on generative AI to do better in their studies, not that those students were learning more. A polished paper proves less than it used to. You are not paranoid for noticing.

But the doubt is also quiet, and I think that is the risk. Students cannot read what you now wonder when a clean paper lands. They can only read what comes out of it: the tone of a comment, a sudden new rule, a question that sounds like an accusation. You and your students are also on different parts of the AI adoption curve, so a silence that means one thing to you can mean something else to them. Forty percent of respondents already report a negative effect on the overall relationship. The rest of that relationship is still there to work with, and it will not protect itself.

Trust runs in both directions

Here is the part most of these conversations skip. Your students are making the same calculation about you.

Oliver Schilke and Martin Reimann ran thirteen experiments on what happens when people disclose that they used AI, published in 2025 in Organizational Behavior and Human Decision Processes. Disclosure lowered trust, and it kept doing so across framings, including a note that a human had reviewed the work. In the first experiment, 195 undergraduate business students trusted a professor less when she disclosed using AI for grading than when she disclosed using a human graduate teaching assistant.

So telling students is not free. The same paper also found that being exposed by a third party did more damage than disclosing it yourself. That last test used a tax advisor rather than a professor, so treat it as a strong hint about classrooms, not a classroom result. I read the two findings together this way: there is no clever framing that makes your own AI use cost nothing, and hoping nobody finds out is the worse bet. You owe your students one plain sentence.

Send one message before the final paper

A policy tells students what is prohibited. It does not tell them what you think of them. So before the last big assignment of the term, send one short class message that does six jobs. Call it the Trust Note.

1. Say what you believe about them. One true sentence. If you cannot write it honestly, that is useful to know too.
2. Say what you cannot tell from a file. Plainly. A finished paper no longer shows you how a student got there, and they already know that.
3. Say where AI is welcome on this assignment, and where it stops. Name the task, not the tool, because the tools keep changing.
4. Say what you will ask to see or hear. One thing that shows the thinking: a note on the hardest decision, a saved draft, a two minute conversation.
5. Say what happens when you have a question about someone’s work. You ask first, and a conversation comes before any conclusion. Check this against your institution’s academic integrity process, and name that process if it applies.
6. Say how you use AI yourself. One sentence. If you use it to draft feedback or class messages, say so, and say what you still check yourself.

Then add one invitation with a reason inside it. LeBlanc’s worry in the survey story is that students can now ask a chatbot what they used to ask in office hours, and a question asked in your office is also “a chance to connect.” An open door is easy to ignore. A specific request is harder to ignore: bring me the paragraph you trust least, and give me ten minutes.

What the note will not do

Straight talk. Nothing in this survey tests a message like this, and I am not going to tell you it restores anything. It is a judgment call, and it is small. It will not tell you who used AI, and it will not fix an assignment that still rewards a polished file with no thinking behind it. If that is your problem, AI Cheating Is a Design Problem. Draw It, Then Defend It. covers the assessment side, and Alaska’s Draft AI Policy Says a Human Decides. Your Syllabus Should Say Where. covers the policy language.

What the note does is narrower. It moves your doubt out of the place where students can only guess at it and into a sentence you chose. It costs one message. And you are still in the position the survey describes, where most of the relationship is intact enough to carry it.

Start with one class this week

The free session walks you through all six jobs on your own class and your own assignment. You will leave holding a finished note and an office hours invitation, in your words, ready to send.

Then the next thought arrives on its own. That is one message for one class. You may teach other sections, each with its own next assignment, and the replies start coming as soon as you press send. The Announcement & Email Drafter writes class messages that sound like you wrote them, grounded in your Course Record, the record of your course, your outcomes, your policies and your voice that every tool in the toolkit uses. Full access for 14 days, cancel in one click. Write this first note by hand anyway, because this one has to be yours. If you use a tool on the next one, your sixth sentence already covers it.

If you would rather try your first sentence on other instructors, the Harvest Kernel Learning Community is free, and the question there this week is simple: when a clean, polished paper lands in your stack now, what is the first thing you wonder?

Your students cannot see your doubt. They can read your next message.

Sources

• Emma Whitford, “Ahead of Midterms, Faculty Report Self-Censoring on DEI, Politics,” Inside Higher Ed, November 2, 2026. insidehighered.com
• Oliver Schilke and Martin Reimann, “The transparency dilemma: How AI disclosure erodes trust,” Organizational Behavior and Human Decision Processes, volume 188, 2025. sciencedirect.com
• “‘A’ Grades Are Suddenly Everywhere Since the Arrival of ChatGPT,” The Wall Street Journal, May 2026. wsj.com