AI Study Tools: High Use, Low Trust—What Cal State Found

Across the Cal State system—23 campuses, more than 460,000 students—a new survey found that students are widely embracing AI tools. The same survey found that most of them distrust the results. They also fear AI will eliminate jobs in their fields.

Those three facts together are more interesting than any of them alone. These aren't students who tried AI once and quit. They're returning, repeatedly, to tools they've already decided are unreliable. That says something specific about how AI study tools are actually functioning—and what educators are dealing with when they try to guide students toward "responsible use."

The Tool Isn't the Problem. The Incentive Is.

Students aren't confused about AI quality. They've already noticed the accuracy problem. What they've also noticed is that a bad AI first draft is still faster than a blank page, and that summarizing a hundred pages of reading into five bullet points—even imperfectly—is better than nothing when three assignments are due Thursday.

This reframes the evaluation question entirely. The standard checklist for AI tools asks: Is the output accurate? Is it pedagogically sound? Is it privacy-compliant? Those matter. But they miss the operative dynamic: a tool can be widely adopted precisely because convenience beats quality when students are under pressure. The tools winning in the student market right now weren't designed to win an accuracy competition.

"Students are pragmatic about AI in a way that many policy debates aren't: they use it because it's convenient, not because it's reliable."

If educators respond to this data by recommending better tools without addressing the underlying pressure, those recommendations will be ignored.

What's Actually in the Student Toolkit

The tools most students are using aren't education-specific. They're general-purpose LLMs—ChatGPT, Microsoft Copilot, Google Gemini—deployed at massive scale for a use case their developers didn't design for. Here's how the main options actually stack up for educators trying to make recommendations.

ChatGPT (OpenAI)
The default choice for most students. Low friction, no setup, available on any device. Hallucination rates on specific factual claims remain a real problem—particularly in domains like history, science, and law. There are no academic integrity features, no citation standards, and the consumer version's data terms are not appropriate for K-12 contexts. For adult learners who understand they're working with a drafting assistant and not a research tool, it has legitimate utility. For anyone expecting reliable facts, it doesn't.

Microsoft Copilot for Education
If your institution runs Microsoft 365 Education, this is already licensed and already compliant. The data handling is substantially cleaner than consumer ChatGPT—Microsoft's education agreements cover student data under FERPA in ways the consumer product does not. Output quality is comparable to GPT-4. The institutional controls are real. For schools already in the Microsoft ecosystem, this is the most defensible tool to recommend and the one with the least privacy exposure. It is not more accurate than ChatGPT; it's more compliant.

Google Gemini for Workspace (Education Edition)
The Google-side equivalent. If your district uses Google Workspace for Education, Gemini is bundled and covered under Google's education data agreements. Same logic as Copilot: not more accurate, but more defensible from a compliance standpoint. Choose based on your existing ecosystem, not on output quality claims.

Khanmigo (Khan Academy)
The most carefully designed education tool in this space, and also the least likely to be chosen over ChatGPT by a student under pressure. Khanmigo uses Socratic prompting—it guides students toward answers rather than providing them directly. That's pedagogically sound and genuinely protective of the learning process. It's curriculum-aligned, FERPA- and COPPA-compliant for K-12, and honest about what it is. The problem: it's slower, more demanding, and actively resists being used as a shortcut. Students who want to shortcut their work won't choose it. Teachers who care about whether AI is actually supporting learning should push for it.

Perplexity AI
An outlier worth knowing about. Perplexity cites sources by default, which ChatGPT does not—giving students at least a visible trail to verify. It functions better as a search tool than a writing tool, and some students are already using it that way. Not designed for education, privacy terms for minors are not robust, avoid for K-12. For college-level research workflows where students are taught to verify sources, it has more honesty baked in than the alternatives.

---

Key stat: The Cal State system's finding isn't a data point about one campus. It's a read on roughly half a million students in one public university system—making it one of the most significant ground-level snapshots of student AI behavior published this year.

---

Try This Tomorrow

Before recommending any AI tool to students, run a ten-minute test: ask it three factual questions from your subject area with specific, verifiable answers. Note how often it's wrong, how confident it sounds when wrong, and whether it acknowledges uncertainty. That test will calibrate your guidance more usefully than any vendor demo.

Tool Verdict

Usefulness: High (for task completion) / Low (for reliable information output)
Time saved: High across all tools
Setup friction: Low (consumer tools) / Medium (institutional tools)
Privacy/compliance clarity: Weak (ChatGPT consumer) / Acceptable (Copilot, Gemini education editions) / Strong (Khanmigo)
Learning curve: Low across the board
Output quality: Inconsistent; students have already figured this out independently

Best For / Not For

Best for: College instructors building AI literacy curriculum; IT administrators evaluating institutional licensing; instructional coaches helping faculty set clear guidelines before assigning work where AI use is permitted.

Not for: Anyone expecting accuracy without human verification; K-12 contexts that haven't run a privacy review; teachers hoping students will voluntarily choose the slower, more educational option.

What to Avoid

Don't recommend consumer AI tools as research tools and expect that framing to stick. Students are already using them that way, and they've already noticed the accuracy problem. Endorsing a tool without naming its limits signals that you haven't used it seriously—which undermines your credibility on the rest of your guidance. Be explicit: "Use it to draft, not to cite. Use it to brainstorm, not to verify."

Also avoid treating this survey data as a problem of student trust that better prompting will fix. The issue isn't that students need to learn to trust AI more. The issue is that current AI study tools don't deserve more trust than students are already giving them.

---

The NeuralClass Takeaway

The Cal State finding is clarifying, not surprising. Students are more clear-eyed about AI limitations than much of the policy debate gives them credit for—they've just decided the convenience is worth it anyway. Educators who want to actually influence how students use these tools need to start from that reality, not from the assumption that better guidance will produce better choices. The lever isn't trust. It's redesigning the conditions under which AI becomes the default.

Reflection Question

If your students are regularly using AI tools they don't trust for academic work, what does that tell you about what they're actually trying to accomplish—and whether your current assignments are structured in ways that reward completion over thinking?

Practical Next Step

Identify one assignment this week where AI shortcuts are most likely being used. Redesign one component—not the whole thing—so that completion requires something AI can't replicate from context alone: a specific observation from your classroom, a direct reference to a conversation from that week, or a comparison to the student's own prior work.

Related Reading

• AI Study Tools and the Productive Struggle Problem
• Khanmigo in the Classroom: What Teachers Actually Get
• AI Productivity Tools: Do They Actually Support Learning?

---

Homepage block: Tools Worth Knowing — a recurring NeuralClass series evaluating specific AI tools for classroom and school use, with honest verdicts rather than promotional overviews.