The science · Why it works
In 1984, Benjamin Bloom set a challenge. We built ProfilEd to take it on.
One-to-one tutoring beats classroom teaching by a wide margin. It was always too expensive to give everyone. Here is how ProfilEd does it, and the research behind every part.
No account needed · about five minutes
Before you read anything else
A 60-second test.
Psychologists have run this experiment for twenty years. It shows the problem ProfilEd exists to solve better than any paragraph we could write.
-
1. Pick one.
2. How well do you understand it? 1 = not at all · 7 = I could teach it
Next -
3. Now explain how a zipper works, step by step, as if to a curious twelve-year-old. Every step, in order.
Stays in your browser. Nothing is sent anywhere.
I’ve tried -
4. Having tried, how well do you understand it now?
See my result -
6 → 4
In the original studies, most people’s ratings dropped after trying to explain. Researchers call it the illusion of explanatory depth.1 It happens to experts too, in their own fields.
That gap, between what you think you know and what you can explain, is exactly what ProfilEd finds. And then works on.
Find yours in a 5-minute check-inTry another
The problem
Most of what we call proof of learning measures something else.
Learning that feels easy usually doesn’t last. Learning that feels effortful, like recalling instead of re-reading or explaining instead of recognizing, sticks. Psychologist Robert Bjork calls these desirable difficulties.2
Most learning products remove the difficulty because it feels better. ProfilEd keeps the useful kind, and removes the rest.
The challenge
Tutoring works. Nobody could afford it.
“The tutoring process demonstrates that most of the students do have the potential to reach this high level of learning. I believe an important task of research and instruction is to seek ways of accomplishing this under more practical and realistic conditions than the one-to-one tutoring, which is too costly for most societies to bear on a large scale.”
An honest footnote. Later research puts the tutoring advantage lower than Bloom’s original “two sigma”, but still large. A 2011 review found human tutoring and well-designed step-by-step tutoring systems almost equally effective (effect sizes of 0.79 and 0.76).4 The lesson for us: the design of the tutoring matters more than whether the tutor is human.
Then AI arrived
The same AI. Opposite results.
In 2025, two experiments tested AI in real classrooms. Together they explain both the danger and the opportunity.
Researchers at the University of Pennsylvania gave high-school math students either no AI, a standard ChatGPT-style assistant, or an AI tutor that gave hints but not answers.5
With the standard assistant, practice scores rose 48%. When the AI was taken away for the exam, those students scored 17% worse than students who never had AI at all. They had practiced getting answers, not understanding.
The same year, a Harvard randomized trial tested an AI tutor built on learning-science principles against a well-run, active in-class lesson. Students using the AI tutor learned more than twice as much, in less time.6
The difference wasn’t the AI. It was the design.
During practice AI available
On the exam AI taken away
The six principles
Every part of ProfilEd answers to a principle.
Learning science has argued for a century. On a few things it now largely agrees. We built around those, and around one principle of our own.
Tutor, don’t tell.
“The tutoring process demonstrates that most of the students do have the potential to reach this high level of learning.”
Benjamin Bloom, 1984Read principle I IIIf you can’t explain it, you don’t know it yet.
“The first principle is that you must not fool yourself — and you are the easiest person to fool.”
Richard Feynman, 1974Read principle II IIIEvery idea needs a worthy opponent.
“Organizations frequently take actions in contradiction to what they really want to do.”
Jerry B. Harvey, 1974Read principle III IVFix the root, not the symptom.
“By repeating why five times, the nature of the problem as well as its solution becomes clear.”
Taiichi Ohno, 1988Read principle IV VChange the belief, not just the answer.
“If learning is to persist, managers and employees must also look inward.”
Chris Argyris, 1991Read principle V VIEarn the right to speak.
“Do not say that for which you lack adequate evidence.”
H. P. Grice, 1975Read principle VIHow ProfilEd thinks
Most AI says something every time you ask. ProfilEd stays quiet until it’s sure.
“The product earns the right to speak by the completeness and accuracy of its model of the person.”
How it compares
Where ProfilEd sits.
| Question | Generic AI chat | Video course / LMS | Human coach | ProfilEd |
|---|---|---|---|---|
| Gives you the answer | Yes | Yes | Rarely | No. It asks. |
| Remembers you across weeks | No | Progress only | Yes | Yes |
| Challenges your reasoning | Tends to agree | No | Yes | Yes, by design |
| Checks you understand before saying you’re done | No | A quiz | Yes | Yes, across several kinds of session |
| Says when it isn’t sure | Rarely | — | Usually | Yes. Confidence rules |
| Hard to outsource to someone, or something, else | No | No | Yes | Yes. Live, spoken, built from your last answer |
| Leaves evidence someone else can review | No | Completion and quiz scores | Their notes, if any | Yes. Transcripts, proof records, CSV export |
| Available at 11pm, every day | Yes | Yes | No | Yes |
| Cost per learner | Low | Low | High | Low |
For you
What this feels like from the inside.
In your first five minutes, you’ll say something out loud you’ve never quite put into words, and leave with one specific thing to do tomorrow.
In your first week, you’ll find at least one thing you thought you understood but can’t yet explain. That isn’t failure. It’s the most useful thing you’ll learn all week.
By week three, it will remind you of something you believed on day two, and show you how that belief has moved.
For institutions
Evidence of thinking that’s hard to fake.
A written essay no longer proves a student understood anything. Neither does a polished take-home case. You need to see reasoning happen.
ProfilEd’s evidence is spoken, live and built over time. Each follow-up question is built from the learner’s own previous answer, so rehearsed or outsourced answers tend to fall apart by the second or third question. In interviews it also notes how something is said: speaking pace, pauses before answering, filler words and tone.
Every concept a learner proves gets a proof record: a unique ID, the date, the score and a hash, kept in My Unified Journey and shareable by link. Transcripts can be reviewed, and session data exports to CSV, built to support direct-assessment evidence for assurance-of-learning reporting such as AACSB Standard 5.7
We don’t claim it’s impossible to cheat. We claim it’s much harder to fake than anything written, and that the evidence builds over weeks, not in one sitting.
Universities & business schools The evidence brief, for committees
Hard questions, straight answers
The questions serious people ask us first.
Isn’t this just ChatGPT with a clever prompt?
It runs on large language models, like most AI products. The difference is everything built around the model: separate roles that don’t share a goal (one questions you, one argues the other side, one hosts); a memory of you across weeks, with a confidence level on everything it believes; rules that stop it speaking until the evidence is there; and a quality gate that sends nothing rather than something generic. A chatbot answers your questions. ProfilEd is built to make you answer its questions.
Can someone cheat with a second screen?
They can try. Each follow-up is built from what was just said, so a read-out answer tends to collapse by the second or third question. Interviews also note pace, pauses and filler words, which change when someone is reading rather than thinking. And the evidence builds over many sessions, not one. We don’t claim cheating is impossible. We claim it’s much harder than with anything written.
What if it gets me wrong?
Sometimes it will. That’s why the more personal a claim is, the more evidence it needs before it says it. It asks before it concludes. It never decides you’ve resolved something just because you said so. And when it isn’t sure, it says nothing. See the exact rules.
Is my voice recorded and stored?
Your conversations are turned into transcripts, and that is what ProfilEd learns from. For exactly what is stored, for how long, and how to delete it, read the Privacy Policy. Nothing is shared unless you share your link.
Which AI model does it use?
Commercial large language models from leading providers, and we’ll switch as models improve. What makes ProfilEd work isn’t the model. It’s the rules around it, and those stay the same whichever model is underneath.
Why should an institution trust it without its own published trial?
It shouldn’t, on faith. The methods it’s built on have strong evidence, listed in full below. ProfilEd itself is new and hasn’t had its own published trial yet. That’s exactly why we offer an 8-week pilot with success measures you set before week one. You see the data either way.
Who uses it today?
We’re opening to our first learners and partner institutions now, so we don’t have user numbers or testimonials to show you yet, and we won’t invent them. Early partners shape what we build next.
What won’t it do?
It won’t write your essay, hand you the answer, flatter you, or claim to know something about you it hasn’t seen. It isn’t therapy, and it doesn’t replace a teacher, coach or manager.
Don’t take our word for it
Test it. Pick the version that fits you.
The fourteen-day test
On day one, pick something you’re sure you understand and explain it out loud in a study session. On day fourteen, do it again. Compare the two.
A pilot, measured your way
Before week one, we agree the success measures with you: for example, blind-rated oral explanations before and after, the trend in concepts learners can’t yet explain, depth of reflection, completion, and faculty review of transcripts. At week eight you get the data and a written report. If the numbers don’t move, you’ll see that too.
- Week 0Measures agreed
- Week 4Review
- Week 8Data & report
See it with your own people
Run a cohort through structured interviews and realistic judgment scenarios, and see who can reason out loud, not just who writes the best answer. The same pilot shape and measures apply.
Honest limits
What we’re not claiming.
- ProfilEd hasn’t yet been tested in its own published randomized trial.The methods it’s built on have been, many times. Our pilots are designed so that each one adds evidence, and we’ll publish what we learn.
- We’re just launching.So there are no user numbers or testimonials on this page. When there are, they’ll be real.
- The AI can be wrong.That’s why it asks before it concludes, and stays quiet when it isn’t sure. Some of its rules are still being built. We list them on the How ProfilEd thinks page.
- It doesn’t replace teachers, coaches or managers.It gives them evidence they don’t currently have.
- It isn’t therapy.It’s a thinking partner. If something serious comes up, talk to a person.
Sources (7)
Every figure and quote on this page, with a link to the original where one exists.
- Rozenblit, L., & Keil, F. (2002). The misunderstood limits of folk science: an illusion of explanatory depth. Cognitive Science, 26(5), 521–562. Source ↗
- Bjork, R. A. (1994). Memory and metamemory considerations in the training of human beings. In J. Metcalfe & A. Shimamura (Eds.), Metacognition: Knowing about Knowing, 185–205. MIT Press.
- Bloom, B. S. (1984). The 2 Sigma Problem: The search for methods of group instruction as effective as one-to-one tutoring. Educational Researcher, 13(6), 4–16. Source ↗
- VanLehn, K. (2011). The relative effectiveness of human tutoring, intelligent tutoring systems, and other tutoring systems. Educational Psychologist, 46(4), 197–221. Source ↗
- Bastani, H., Bastani, O., Sungu, A., Ge, H., Kabakcı, Ö., & Mariman, R. (2025). Generative AI without guardrails can harm learning: Evidence from high school mathematics. PNAS, 122(26). Source ↗
- Kestin, G., Miller, K., Klales, A., Milbourne, T., & Ponti, G. (2025). AI tutoring outperforms in-class active learning: an RCT introducing a novel research-based design in an authentic educational setting. Scientific Reports, 15, 17458. Source ↗
- AACSB International (2020). 2020 Guiding Principles and Standards for Business Accreditation, Standard 5: Assurance of Learning. Source ↗
Five minutes. Out loud. See what it finds.
Just your first name. Your first check-in is free, and you’ll leave with something to do tomorrow.
No account needed · about five minutes
- Do I have to talk out loud?
- For the daily check-in, yes: it’s a voice conversation and you don’t type. Interviews can be voice or fast text, and study sessions can be voice or chat.
- Can I do it on my phone?
- Yes. The check-in runs in the browser on your phone, tablet, or computer.
- Do I need an account?
- Not for your first check-in. A free account keeps your conversation and picks up where you left off.
- Is it private?
- The check-in is confidential, and nothing is shared unless you share your footprint link. For exactly what’s stored, read the Privacy Policy.
- What does it cost?
- The first check-in is free, with no account or card, and new accounts get starter gems to try deeper things. ProfilEd Pro is $19 a month for unlimited use. Full details are on the pricing page.