The science

The science of saying it out loud.

TalkGraph is built on what research says helps people think and learn: explaining in your own words, being asked how, and recalling later. The map is the scaffold that makes all three easy. Here's the evidence, where it lives in the product, and where it stops.

By Klevis Kokoshi, founder · Sources checked 12 October 2026

0.72 vs 0.43Building a concept map helps more than studying a ready-made one (142 comparisons, 11,814 learners)Schroeder et al., 2018
0.55average gain from prompting people to explain what they're learning (64 research reports)Bisra et al., 2018
61% vs 40%recalled a week later by students who tested themselves vs rereadRoediger & Karpicke, 2006
-17%exam grades of students who practised with an answer-giving AI, once it was taken awayBastani et al., 2025

Four principles

01You do the thinking

The ideas on the map are yours, in your words. TalkGraph structures and questions them; it doesn't think them for you.

02Ask, don't answer

A question about what you've just said does more for your thinking than an answer handed to you.

03Knowledge should compound

Every conversation joins one growing network instead of becoming another note.

04Keep what lasts

A connected map of what you know outlives any one-off AI answer.

Ten findings, and how TalkGraph uses them

Finding 1

New knowledge sticks to old knowledge

Learning lasts when new ideas connect to what you already know, which is why concept maps make links visible.

In TalkGraphEvery conversation becomes linked concepts, not a list. New points attach to existing branches, and your Brain links ideas across sessions.

Read the research

David Ausubel's foundational insight in educational psychology is that new learning only sticks when it connects to existing knowledge. His student Joseph Novak turned that into concept mapping, the practice of making relationships between ideas explicit and visible.

“The most important single factor influencing learning is what the learner already knows.”

Ausubel, Educational Psychology: A Cognitive View. Archive.org ↗

Finding 2

Building a map beats looking at one

Concept maps help people learn, most of all when you build the map yourself: across 142 comparisons, creating maps beat studying ready-made ones (effect sizes 0.72 vs 0.43).

In TalkGraphTalkGraph drafts the map from your words so you can keep talking. Then it's yours to work on: rename, move, merge and connect cards, and mark what's settled. The draft saves the slow drawing; checking and fixing it is where the learning happens.

Read the research

Two meta-analyses agree that concept mapping helps. Nesbit & Adesope (2006) pooled 55 studies and 5,818 learners; Schroeder and colleagues (2018) pooled 142 comparisons and 11,814 learners and found that constructing maps helped more than studying maps someone else made. An earlier trial at Barts Medical School found medical students recalled mind-mapped material about 10% better after one week than with their own study methods.

Honest limit: Nobody has measured an AI-drafted map yet, including us. Against other active methods, like writing a summary, the gains from mapping are smaller, and part of the benefit likely comes from the effort of building the map. So treat TalkGraph's map as a draft to work on, not something to watch.

Schroeder et al. (2018) ↗ · Nesbit & Adesope (2006) ↗ · Farrand, Hussain & Hennessy (2002) ↗

Finding 3

Explaining helps, if someone asks

Simply thinking aloud doesn't change how well people do. Being asked to explain does: across 94 studies, explaining raised performance, and prompts to self-explain have an average effect of 0.55.

In TalkGraphWhen you pause, TalkGraph asks one short, open question about what you've just said, to help you develop, challenge or finish the idea yourself. It never answers it for you.

Read the research

Fox, Ericsson & Best (2011) compared 94 studies of people talking while they worked. Plain thinking aloud had no measurable effect, but instructions to describe or explain raised performance above working in silence. Bisra and colleagues (2018) pooled 64 research reports on self-explanation prompts and found an average gain of 0.55 across subjects and settings.

Honest limit: These studies used questions written by researchers and teachers. We haven't yet measured whether TalkGraph's questions do the same.

Fox, Ericsson & Best (2011) ↗ · Bisra et al. (2018) ↗

Finding 4

AI that answers for you can weaken learning

Students who practised with an answer-giving AI scored 48% higher in practice, then 17% lower in the exam once it was taken away. A version built to guide rather than answer largely avoided the harm.

In TalkGraphThis is why TalkGraph questions your ideas instead of writing them, and why the map keeps your own words.

Read the research

Bastani and colleagues ran a field experiment in high-school maths, published in PNAS in 2025. A general chat assistant lifted practice grades by 48%, but students who had used it did 17% worse than those who never had access once it was removed. A tutor version with guardrails that pushed students to work things out themselves lifted practice grades by 127% and largely avoided the drop.

Honest limit: One study, in one subject and one age group. The lesson we take is about answers versus questions, not about any particular product.

Bastani et al. (2025), PNAS ↗

Finding 5

Your head holds about four things

Working memory holds roughly four chunks at once. A forty-minute meeting contains hundreds.

In TalkGraphThe map is external working memory. Every card parked on it is one your brain doesn't have to hold, and it stays there between sessions.

Read the research

Cognitive load research (Sweller, 1988; Cowan, 2001) shows how quickly working memory fills. Good visual organisation isn't decoration: it offloads effort, freeing attention for thinking instead of holding on.

Honest limit: Offloading frees attention in the moment. To learn something rather than just park it, pair the map with recall (below).

Cowan (2001) ↗ · Sweller (1988) ↗

Finding 6

Two channels beat one

Information encoded in words and visually together is remembered better than either alone.

In TalkGraphThe transcript keeps the words; the map adds position, type, colour and connections. Same content, two encodings.

Read the research

Allan Paivio's dual-coding theory: verbal and visual representations are processed in partly separate systems, and material encoded in both has two routes back into memory.

Honest limit: This isn't the "visual learner" idea, which lacks good evidence (Pashler et al., 2008). Words plus structure help everyone.

Paivio, Mental Representations (1986). Oxford University Press ↗ · Pashler et al. (2008) ↗

Finding 7

Copying words isn't understanding

Note-takers who copy word for word tend to process less than those who put ideas in their own words.

In TalkGraphYou're not transcribing; you're explaining in your own words, which is the processing the study rewarded. The map then sorts what you said into decisions, actions and questions: check it rather than accept it.

Read the research

Mueller & Oppenheimer (2014) found laptop note-takers tended to transcribe lectures verbatim and did worse on conceptual questions than those writing by hand, who had to summarise.

Honest limit: A larger replication in 2019 (Morehead, Dunlosky & Rawson) found the difference smaller and less consistent. The durable lesson is about verbatim copying, not screens.

Mueller & Oppenheimer (2014) ↗ · Morehead et al. (2019) ↗

Finding 8

Recall beats rereading

Students who tested themselves recalled about 61% of a passage a week later, against about 40% for those who reread it.

In TalkGraphCollapse a branch and recall what's under it, or take the Brain's 60-second workout: say five of your ideas back from memory.

Read the research

Roediger & Karpicke's Test-Enhanced Learning (2006): rereading looked better five minutes later, but retrieval practice won clearly after a week. It's the principle behind Make It Stick, the classic summary of decades of learning science.

Roediger & Karpicke (2006) ↗ · Make It Stick ↗

Finding 9

Saying it out loud helps it stick

Words you say aloud are remembered better than words you read silently: the production effect.

In TalkGraphThe input is speech. Every idea you map was produced out loud first.

Read the research

MacLeod and colleagues (2010) named the production effect: producing a word, by saying it, makes it more distinctive in memory than reading it silently.

Honest limit: The effect relies on contrast, so it's strongest for the key points you choose to say aloud, not for reading everything aloud.

MacLeod et al. (2010) ↗

Finding 10

People grow when they're seen

Recognition and psychological safety are linked to engagement and better-performing teams.

In TalkGraphEvery voice is attributed, praise is its own card credited to the person, and the People panel tallies who contributed what.

Read the research

Gallup's workplace research links regular, meaningful recognition to higher engagement and lower turnover; Amy Edmondson's work on psychological safety shows teams do best when every voice is heard.

Gallup ↗ · Edmondson, HBS ↗

Where TalkGraph sits

Concept-mapping research assumes you draw the map by hand, which few people do while they think. Chat AI is quick, but it answers for you. TalkGraph sits between: it drafts the structure of your own words as you speak, asks rather than answers, and brings your ideas back for recall.

  1. 1You talk
  2. 2A draft map of your own words
  3. 3One question at the pause
  4. 4Recall it later
A TalkGraph map built live from a conversation
Your words become a draft map as you speak
TalkGraph's pop-up asking a question about what was just said
At a pause, it asks you one question
All 26 sources and further reading
  1. Schroeder, N. L., Nesbit, J. C., Anguiano, C. J. & Adesope, O. O. (2018). Studying and constructing concept maps: a meta-analysis. Educational Psychology Review, 30(2), 431-455. View ↗
  2. Bisra, K., Liu, Q., Nesbit, J. C., Salimi, F. & Winne, P. H. (2018). Inducing self-explanation: a meta-analysis. Educational Psychology Review. View ↗
  3. Fox, M. C., Ericsson, K. A. & Best, R. (2011). Do procedures for verbal reporting of thinking have to be reactive? A meta-analysis and recommendations for best reporting methods. Psychological Bulletin. View ↗
  4. Bastani, H., Bastani, O., Sungu, A., Ge, H., Kabakci, O. & Mariman, R. (2025). Generative AI without guardrails can harm learning: evidence from high school mathematics. PNAS. View ↗
  5. Pashler, H., McDaniel, M., Rohrer, D. & Bjork, R. (2008). Learning styles: concepts and evidence. Psychological Science in the Public Interest, 9(3), 105-119. View ↗
  6. Novak, J. D. (1990). Concept mapping: a useful tool for science education. Journal of Research in Science Teaching, 27(10). View ↗
  7. Nesbit, J. C. & Adesope, O. O. (2006). Learning with concept and knowledge maps: a meta-analysis. Review of Educational Research, 76(3). View ↗
  8. Novak, J. D. (1998). Learning, Creating, and Using Knowledge: concept maps as facilitative tools. Routledge. View ↗
  9. Novak, J. D. & Gowin, D. B. Learning How to Learn. Cambridge University Press / Springer. View ↗
  10. Ausubel, D. P. Educational Psychology: A Cognitive View. View ↗
  11. Paivio, A. (1986). Mental Representations: a dual coding approach. Oxford University Press. View ↗
  12. Sweller, J. (1988). Cognitive load during problem solving. Cognitive Science, 12(2). View ↗
  13. Roediger, H. L. & Karpicke, J. D. (2006). Test-enhanced learning: taking memory tests improves long-term retention. Psychological Science, 17(3), 249-255. View ↗
  14. Brown, P., Roediger, H. & McDaniel, M. (2014). Make It Stick. Harvard University Press. View ↗
  15. Sawyer, R. K. (ed.). The Cambridge Handbook of the Learning Sciences. View ↗
  16. Hattie, J. Visible Learning. Routledge. View ↗
  17. National Academies (2000). How People Learn: brain, mind, experience, and school. Free to read. View ↗
  18. The Learning Scientists: evidence-based learning resources. View ↗
  19. IHMC CmapTools: concept mapping research, papers and tools from Novak's group. View ↗
  20. Mueller, P. A. & Oppenheimer, D. M. (2014). The pen is mightier than the keyboard. Psychological Science, 25(6), 1159-1168. View ↗
  21. Morehead, K., Dunlosky, J. & Rawson, K. A. (2019). How much mightier is the pen than the keyboard for note-taking? Educational Psychology Review, 31, 753-780. View ↗
  22. Farrand, P., Hussain, F. & Hennessy, E. (2002). The efficacy of the 'mind map' study technique. Medical Education, 36(5), 426-431. Recall 10% higher after one week; the authors' motivation-adjusted estimate is 15%. View ↗
  23. Cowan, N. (2001). The magical number 4 in short-term memory. Behavioral and Brain Sciences, 24(1). View ↗
  24. MacLeod, C. M., Gopie, N., Hourihan, K. L., Neary, K. R. & Ozubko, J. D. (2010). The production effect: delineation of a phenomenon. Journal of Experimental Psychology: Learning, Memory, and Cognition, 36(3). View ↗
  25. Gallup: Employee Recognition. View ↗
  26. Edmondson, A. Psychological Safety (Harvard Business School). View ↗

See it work on your own thinking.

Try it free, no sign-up

Read: is AI making us worse at thinking? →

These optional services load only after you accept. Closing the banner leaves them off. You can change your choice at any time using “Review privacy choices” on our Privacy Policy page.

Read the Privacy Policy