All Projects

Research · UX methods · 5 min read

AI: An Escape from Illusion

A UX research study on trust, efficiency, and understanding: what happens to learning when AI makes studying feel effortless.

Role

UX Researcher

Method

1:1 interview

Type

Academic research

Focus

Trust · Efficiency

Illustrated portrait of Ben, the graduate student persona at the center of this study

User persona

Meet Ben.

B

Graduate student, International Relations & Law

An ambitious learner searching for clarity in complexity. Ben’s coursework runs 80–120 pages of reading a week, with heavy conceptual loads and a real need for precise understanding: exactly the conditions where AI promises the most help.

“AI made studying fast.. but it made me question my understanding”

Ben

Habits & frustrations

A full AI toolkit, and a full list of ways it fails him.

How Ben uses AI

  • NotebookLM for structured summaries
  • ChatGPT for clarifications
  • Perplexity for verified citations
  • AI-generated study plans

Where AI fails him

  • Wrong interpretations
  • Hallucinated examples
  • Missing required angles
  • Verification burden
  • Shallow explanations

Setting

Before AI: learning was slow, but deep.

Ben’s work was extremely text-heavy: books, articles, reports. He spent hours synthesizing them by hand. Slow, but the understanding stuck.

“My work was extremely text-heavy: books, articles, reports. I spent hours synthesizing them.”

Ben

The real challenge

Processing, not the content.

  • Synthesizing multiple sources
  • Limited time for deeper reflection
  • Cognitive fatigue
  • FOMO of missing a key detail

What changed

Efficiency rose. Engagement and understanding fell.

01

Mental effort dropped fast, then had to be earned back.

Self-reported mental effort fell from a 9 (before AI) to a 4 during over-reliance, bottoming out at 3.5 during an engagement dip, before climbing back to 5.5 once Ben deliberately rebalanced his usage.

02

AI made studying easier, but learning weaker.

"I wasn’t reading, I was just skimming summaries," Ben said. Fluency with the material came from the AI’s explanation, not from his own processing of the source.

03

The hidden cost was verification time, not reading time.

Students re-check everything AI gives out. Time saved in reading became time lost in verifying, and trust in the tool fluctuated session to session rather than settling.

Mental effort across AI adoption phases

Mental effort
9

Before AI

6

First use

4

Over-reliance

3.5

Engagement dip

5.5

Balanced

By the numbers

The adoption is universal. So is the doubt.

92%

use AI regularly for academic work

75%

reduced study time without deepening learning

83%

worry about AI accuracy

58%

fear losing critical thinking skills

The aha moment

AI wasn’t helping him think faster. It was helping him think later.

Even when AI made work quicker, Ben consistently reported the same pattern: revising faster, but understanding less; feeling productive, but unable to explain the content afterward; depending on AI more than he realized. The real problem wasn’t time pressure: it was a hidden erosion of understanding.

Efficiency vs. understanding across AI adoption phases

AI efficiencyUnderstanding
4
8

Before AI

7
6

First use

9
5

Over-reliance

9
3

Aha moment

7
6

Balanced use

What the research revealed

Fluency isn’t mastery.

  • AI hides its reasoning steps
  • Fluency ≠ mastery
  • Over-reliance leads to shallow understanding
  • Students mistake speed for depth

Resolution

AI that supports thinking, not replaces it.

Think First, Ask AI Second

Students jump to AI because it saves time, but that also turns it into a shortcut through reasoning. Have people draft their own take before the AI weighs in.

Structured verification over blind trust

Across interviews, students said AI sounds confident even when wrong. Build in a deliberate check step, not just a chat window.

Reveal the reasoning path

The biggest aha moment was realizing AI hides the work: it gives answers, not reasoning. Tools that show the path let students catch errors and actually learn from them.

Measuring healthy AI use

What success looks like, if these recommendations work.

🧠

Think First → Ask AI Second

70%+ tasks start with a student-generated draft

Independent reasoning attempts ↑30% · AI-first prompting ↓40%

Builds reasoning before using AI.

🔎

Structured verification

Verification actions ↑40%

AI errors caught ↑25% · trust–accuracy deviation <10%

Reduces blind acceptance of AI.

🧭

Show reasoning path

Reasoning mode use ≥60%

Understanding score ↑20% · reasoning clarity ↑25%

Makes AI's thinking visible.

📖

Reinforce deep reading

Deep reading completion ≥80%

Shortcut substitution ↓40% · interpretation accuracy ↑25%

Keeps comprehension rooted in the source text.

Biggest learning

Students don’t need AI that makes studying easy. They need AI that makes learning meaningful.

“AI can improve great lawyers, but it cannot replace them.”

Ralph Losey