Effortful Learning Feels Worse and Works Better

Effortful learning has a marketing problem: it feels worse while it works better. In a randomized 2019 study in PNAS, college physics students in active-learning classes learned more than students in polished lectures, yet rated their own learning lower. The researchers’ explanation is the part every parent should know. When students feel the extra mental effort, they “initially take that effort to signify poorer learning.”

Now hand a child an AI tool that makes every assignment feel smooth. By that same logic, it will feel like the best teacher they ever had. That feeling is worth questioning.

Key Takeaways

• In Deslauriers et al. (PNAS, 2019), randomly assigned college physics students in active classes learned more but felt they learned less than peers in lectures from highly rated instructors.
• The cause, in part, was effort: students misread the strain of active learning as a sign they were doing badly.
• The same pattern appears elsewhere: people rated “massed” study as more effective than interleaved study even after their own test scores showed the opposite (Kornell & Bjork, 2008).
• A 53-study meta-analysis found that struggling with a problem before instruction beat instruction-first, though not for second to fifth graders, who did better with guidance first.
• When AI removes the struggle, a child may feel more successful and learn less. Teach kids what effort feels like and what it means.

What Is Effortful Learning? The PNAS Physics Study

Effortful learning means study methods that make the brain work harder in the moment: answering questions instead of listening, recalling instead of rereading, solving before being shown. Learning scientists Elizabeth and Robert Bjork named these “desirable difficulties”: conditions that make practice feel slower and harder but pay off in long-term retention.

The clearest evidence that effort feels bad comes from Louis Deslauriers, Logan McCarty, Kelly Miller, Kristina Callaghan and Greg Kestin. Their paper, “Measuring actual learning versus feeling of learning in response to being actively engaged in the classroom,” appeared in PNAS on September 4, 2019.

The design was careful. Students in large introductory college physics courses were randomly assigned to two kinds of class. One was active instruction, following best practices in the field. The other was passive instruction: lectures by “experienced and highly rated instructors.” Both groups got identical content and handouts, and instructors didn’t try to sell either method.

The results:

• Students in active classrooms learned more.
• Their perception of learning was lower than their peers in the lectures, though still positive.
• The authors showed this negative link was “caused in part by the increased cognitive effort required during active learning.”

Their warning about the lecture side is pointed: “a superstar lecturer could create such a positive feeling of learning that students would choose those lectures over active learning.”

One correction to how this study is sometimes described: these were college students, not children. It’s the best-designed test of the feeling-versus-learning gap, but extending it to an 11-year-old is an inference.

Why Hard Feels Like Failing

Here’s what is going on underneath. When you study, your brain is constantly estimating “am I getting this?” It can’t observe memory formation directly, so it uses proxies. The main proxy is processing fluency: how smoothly information flows. Smooth equals “I know this.” Strain equals “I’m lost.”

Fluency is a decent proxy for recognition, the feeling of “I’ve seen this before.” It’s a poor proxy for retrieval, the ability to produce the answer on a blank page later. The methods that build retrieval are almost always less fluent. You are pulling, reconstructing and sometimes failing. That friction is the work.

The same misreading turns up in other experiments:

• Rereading versus testing. Henry Roediger and Jeffrey Karpicke (Psychological Science, 2006) found that taking recall tests produced substantially better retention a week later than rereading, “even though repeated studying increased students’ confidence.”
• Blocked versus mixed practice. Nate Kornell and Robert Bjork (Psychological Science, 2008) had people learn painters’ styles. Mixing artists together worked better than studying one artist at a time, yet participants rated the one-at-a-time method as more effective “even after their own test performance had demonstrated the opposite.”

That last detail is striking. Seeing their own scores didn’t fix their beliefs. The feeling of fluency won.

Struggle first, but watch the age

Productive failure is a specific version of effortful learning: try the problem before being taught the method. Tanmay Sinha and Manu Kapur’s 2021 meta-analysis in Review of Educational Research pooled 53 studies with 166 comparisons. Problem-solving-then-instruction beat instruction-first with a moderate effect (Hedges’ g = 0.36), rising to as much as 0.58 when done with high fidelity to the method.

The age caveat matters for families. For “younger age learners (second to fifth graders),” effects favored instruction first. Young kids often need a model before struggle is productive. Struggle is a tool, not a virtue in itself.

What This Means When AI Makes Homework Easy

An AI chatbot is close to a perfect fluency machine. It answers instantly in clear, confident sentences. If a child’s internal gauge reads fluency as learning, an AI-assisted session will register as a great study session.

The field evidence suggests that feeling can be wrong in a costly way. In a 2025 PNAS field experiment, Hamsa Bastani and colleagues gave nearly 1,000 high school math students GPT-4. With a standard chatbot interface, practice grades rose 48%. When access ended, those students scored 17% worse than peers who never had it. The authors wrote that students used the tool “as a ‘crutch’.” A tutor version that gave hints, which kept more of the effort with the student, largely avoided the drop.

And in a 2024 study in the British Journal of Educational Technology, Yizhou Fan and colleagues found university students who revised essays with ChatGPT improved their scores most but gained no more knowledge than others. We break that one down in metacognitive laziness: better essays, no new knowledge.

Neither study proves AI always harms learning. Both show the same shape as the physics study: the easy path produced the better-looking session and the weaker learner.

Feels Like Learning vs. Builds Learning

Study method	How it feels	What the research found
Polished lecture or explainer video	Clear, satisfying	Lower learning than active classes, higher felt learning (Deslauriers et al., 2019)
Active problem-solving in class	Confusing, effortful	More learning, lower felt learning (Deslauriers et al., 2019)
Rereading notes	Familiar, confident	Better at 5 minutes, worse at 1 week (Roediger & Karpicke, 2006)
Self-testing from memory	Frustrating, slow	Much better at 2 days and 1 week (Roediger & Karpicke, 2006)
One topic at a time (blocked)	Smooth	Worse for telling categories apart (Kornell & Bjork, 2008)
Mixed topics (interleaved)	Jumbled	Better, though rated as worse (Kornell & Bjork, 2008)
Unrestricted chatbot for practice	Easy, fast	17% worse once removed (Bastani et al., 2025)
Hint-only AI tutor	Some struggle kept	Harm largely mitigated (Bastani et al., 2025)

How to Help Your Kid Trust Effortful Learning

Name the feeling before it happens

Tell your child, before homework starts: “If this feels hard and a bit confusing, that’s often what learning feels like. It doesn’t mean you’re bad at it.” The Deslauriers team suggested instructors address this misperception directly early on, and they described a successful example of doing so. Parents can do the same in one sentence.

Swap one rereading session for a blank-page quiz

Instead of reviewing notes, your child closes the notebook and writes down everything they remember, then checks. It will feel worse. Point that out, then show them the result a few days later. Our guide to retrieval practice and the testing effect has age-by-age versions.

Put AI after the struggle, not before

A simple order: try first for a set time (five minutes for younger kids, fifteen for teens), then use AI for a hint or to check, not for the full solution. This keeps the productive part of the effort while still letting the tool help. For kids under about 10, the meta-analysis suggests showing an example first is fine; struggle works better when they have something to build on.

Judge the method by next week, not tonight

Learning methods should be judged by delayed results. Have your child rate how well a session went on a 1–5 scale, then quiz them a few days later. Over a month, they’ll see for themselves whether “felt great” and “stuck” line up. For background on why harder practice sticks, see our piece on desirable difficulty.

What not to do

Don’t make work artificially miserable. Desirable difficulties are specific methods like recall, spacing and mixing, not general suffering. Confusing a child with material far beyond their level isn’t productive struggle; it’s just struggle. If your child is upset rather than effortfully engaged, back off and give an example.

What to Watch For Over the Next 3 Months

• Week 4: Your child can name at least one study method that felt hard but worked. They’re starting to say “this is hard” without adding “so I’m bad at it.”
• Month 2 red flags: Homework is consistently fast and easy, but test performance hasn’t improved. Your child avoids any task that can’t be done with AI. Frustration turns into shutting down rather than effort.
• Month 3 self-check: Your child chooses a self-quiz or a try-first approach without being asked at least some of the time. Delayed quiz results are stable or improving.

Frequently Asked Questions

Why does studying feel harder when it’s working?

Methods that build long-term memory, like recalling from memory or mixing topics, make your brain reconstruct information rather than just recognize it. That’s less fluent, and the brain reads low fluency as “I’m not getting this.” Studies from 2006 to 2019 show students consistently misjudge effortful methods as less effective even when they work better.

Was the PNAS study done with kids?

No. Deslauriers and colleagues studied college students in introductory physics. It’s the most rigorous test of the gap between felt and actual learning, with random assignment and identical materials. Applying it to children is a reasonable inference, since younger learners have less experience judging their own learning, but it hasn’t been tested directly.

Does this mean my kid should never get help?

No. The productive-failure meta-analysis found second to fifth graders did better with instruction first. Effort helps when a child has enough to work with. Give younger kids an example, then let them try. For older kids, a short solo attempt before help is a good default.

Is AI bad for learning because it makes things easy?

It depends on how it’s used. In the 2025 PNAS high school study, an unrestricted chatbot raised practice grades but hurt later performance, while a hint-only tutor largely avoided the harm. The issue isn’t AI itself. It’s whether the effort still happens in your child’s head.

How do I convince my child that hard is good?

Don’t argue; run an experiment. Have them study one set of words by rereading and another by self-testing, then quiz both a few days later. Most kids find the self-test set stuck better. Seeing it in their own results persuades more than any lecture from a parent.

---

About the author

Ricky Flores is the founder of HiWave Makers and an electrical engineer with 15+ years of experience building consumer technology at Apple, Samsung, and Texas Instruments. He writes about how kids learn to build, think, and create in a tech-saturated world. Read more at hiwavemakers.com.

---

Sources

1. Deslauriers, L., McCarty, L. S., Miller, K., Callaghan, K., & Kestin, G. (2019). “Measuring actual learning versus feeling of learning in response to being actively engaged in the classroom.” Proceedings of the National Academy of Sciences, 116. https://doi.org/10.1073/pnas.1821936116
2. Roediger, H. L., & Karpicke, J. D. (2006). “Test-enhanced learning: Taking memory tests improves long-term retention.” Psychological Science, 17(3). https://doi.org/10.1111/j.1467-9280.2006.01693.x
3. Kornell, N., & Bjork, R. A. (2008). “Learning concepts and categories: Is spacing the ‘enemy of induction’?” Psychological Science, 19. https://doi.org/10.1111/j.1467-9280.2008.02127.x
4. Sinha, T., & Kapur, M. (2021). “When problem solving followed by instruction works: Evidence for productive failure.” Review of Educational Research, 91. https://doi.org/10.3102/00346543211019105
5. Bastani, H., Bastani, O., Sungu, A., Ge, H., Kabakcı, Ö., et al. (2025). “Generative AI without guardrails can harm learning: Evidence from high school mathematics.” Proceedings of the National Academy of Sciences. https://doi.org/10.1073/pnas.2422633122
6. Fan, Y., Tang, L., Le, H., Shen, K., Tan, S., Zhao, Y., et al. (2024). “Beware of metacognitive laziness: Effects of generative artificial intelligence on learning motivation, processes, and performance.” British Journal of Educational Technology, 56(2). https://doi.org/10.1111/bjet.13544