1 of 25

1

Chat2Learn Suite: a tool to support learning by conversation

Michelle Michelini, University of Chicago

October 2025

2 of 25

The Importance of Skills

Skills are the foundation of both individual well-being (economic and otherwise) and national prosperity.

It is now widely recognized that skills encompass both cognitive (i.e., reading and math test scores) and “non-cognitive” (e.g., persistence, curiosity, self-control) skills.

Many studies correlate childhood skills with adult outcomes. There is some causal evidence on the long-run effect of cognitive skills, but little evidence for the long-run impact on non-cognitive skills.

But good evidence that non-cognitive skills are increasingly valued in the labor market (Deming & Noray, 2018; Deming, 2021; Makridis & Heckman, 2025) and that the wage premium for non-cognitive skills is greater than that for cognitive skills (Noray, 2020)

3 of 25

How to build skills?

Most countries spend a lot on schooling – up to 5% of GDP per capita
Schools are essential to skill-building
But schools cannot do it alone

4 of 25

The problem we aim to solve

Half of children in LMICs fail to reach their developmental potential by age 5. Cognitive and non-cognitive skill gaps based on family wealth emerge early, persist into adolescence, and rarely narrow without intervention (Attanasio et al., 2025).
Playful, responsive, open-ended language interactions with children are most effective for engaging young children in learning and building skills (Weisberg et al., 2016; Golinkoff et al., 2019), but these practices are not accessible to all.
Traditional in-person early childhood interventions are costly and challenging to scale – this applies to classroom and parenting interventions i.e. coaching, home visiting (Dowdall et al., 2020; Kraft et al., 2018).
Promising technologies (including AI integrations) can support playful learning at home and in schools as a complement (not a substitute) for vital adult-student interactions.

4

5 of 25

Talk as Capital

Children learn language through everyday interaction with more knowledgeable others — primarily parents and secondarily teachers (Vygotsky, 1978)

Quantity of talk: More child-directed words predict stronger early oral language skills (Hart & Risley, 1997)
Quality of talk: Higher quality talk more powerfully predicts later vocabulary and reading achievement (Rowe, 2012; Cartmill et al., 2013)

Meta-analysis: input quality (r = .33) is a stronger predictor of child language outcomes than input quantity (r = .20) (Anderson et al., 2021).

Three key dimensions of high-quality input (Rowe & Snow, 2019)

Interactive — open-ended questions, joint attention
Linguistic — lexical diversity, syntactic complexity
Conceptual — decontextualized talk fostering abstract reasoning

6 of 25

We’ve learned that early language exposure, particularly interactive conversation, is crucial for vocabulary development and broader cognitive, social, and emotional growth, with lifelong impacts on learning

(Rowe, 2008 ; Golinkoff et al., 2019 ; Cunha & Heckman, 2007)

7 of 25

What do effective tools for behavior change do?

Motivate action by managing procrastination (i.e., present bias)
Provide in-the-moment, actionable feedback that is relevant, personalized and accurate
Strengthen relevant identity by making sentiment and responsibilities salient in the moment
Boost self-efficacy
Make learning fun, efficient, and habitual

7

8 of 25

8

Chat2Learn Suite:

A practical tool to support parent + teacher engagement in children’s learning

Chat2Learn sparks dynamic, open-ended conversation between children and the adults who care for them through creative, illustrated conversation prompts. Chat2Learn is grounded in behavioral science with an on-demand AI coach adults can use to personalize support for children’s learning through conversation.

Small talk for big ideas

9 of 25

Chat2Learn background and future

9

Origins

Began in Illinois in 2020-2021 as state pandemic response initiative for parents (enrolled 2,000 families)
Original program delivered static, one-way conversation prompts
Experimental test of approach completed with 600 US families showed large positive impacts on child vocabulary and parents’ growth mindset

Current

Multi-modal AI-powered parent program (SMS, mobile app, WhatsApp) launched in March 2025
Classroom app and linkage with parent program
US pilot studies for home and classroom use
Collaboration with UChicago Development Innovation Lab (PI Michael Kremer) for testing in Peru in partnership with Peruvian government

10 of 25

Chat2Learn Suite theory of change

10

Improves PARENT-CHILD conversation

Improves TEACHER-CHILD conversation

Improves PARENT-TEACHER communication

A FAMILY-SCHOOL collaboration to improve children's learning by conversation

11 of 25

Relational tech designed to boost school-readiness

Language Skill.

i.e., Receptive and expressive vocabulary, story-telling, critical thinking.

11

Social Skill.

i.e., Listening, attention, self-regulation, turn-taking.

Curiosity.

i.e., Questioning, probing, exploring novelty, enthusiasm for learning.

12 of 25

How does the Chat2Learn Suite work?

Harnesses technology to boost children’s skills with illustrated conversation prompts that spark two-way adult-child talk
Suggests prompts via app with a custom AI agent controlled by the teacher/parent for more personalized prompts
Creates a log of text- and audio-recorded classroom conversations (audio recording allows hands-free program interaction)
Coming soon: teachers and parents engage in targeted communication (i.e. assign, share, extend conversations)

Example of a Chat2Learn prompt and paired illustration

13 of 25

High value, low cost

Chat2Learn Suite:

requires no resources besides a phone or tablet (iOs, Android)
places minimal time and attention demands on teachers and parents
is very low-cost to implement

Example of a Chat2Learn prompt and paired illustration

Introducing Chat2Learn

14 of 25

Target population

Low-income families with 3-6-year-old children and classrooms serving these children
Available in 50+ languages
No geographic limitations, although low-connectivity areas will require low-tech adaptations
Add-on as a supplement to foundational literacy and numeracy programs possible

14

15 of 25

15m ago

11:11

New Chat

Home

Type a message...

That’s true! Short people can also fit in small spaces like blanket forts or kid slides. If you and Michael Jordan played hide and seek, where would you hide?

11:11

New Chat

Michael Jordan!

Home

Who is the tallest person you know? What are some things that are easy for tall people to do? Text back Chelly’s answer!

Who is the shortest person you know? What are some things that are easy for short people to do?

She said herself! She said her it's easy for her to reach things low to the ground.

Saturday, 25 May

Chat2Learn is especially designed to focus on decontextualized language (DXT), which is a key feature of linguistic input that supports children's cognitive and language development. DXT is defined as talk that references past events, future possibilities, explanations, and abstract ideas, and supports children’s critical thinking, oral language, and reading comprehension abilities (Dickinson & Tabors, 2001; Rowe, 2013; Rowe & Snow, 2020; Snow, 1991; Uccelli et al., 2019). Unlike math talk, DXT lacks fixed linguistic markers and exists along a continuum, from present-focused to highly abstract (Snow, 1983; Van Kleeck, 2014). American parents who use more “decontextualized language” or language beyond the here-and-now, such as talk about the past or the future, or explanations about how things work in the world, have preschoolers with greater vocabulary, syntax, narrative, and academic language skills than parents whose talk is more contextualized. These findings hold when you control for the total amount of talk parents produce (Demir-Lira et al., 2015; Rowe, 2012; Uccelli et al., 2019). American parents who are higher SES, tend to provide more explanations to their children during interactions than lower-SES parents, on average, yet there is no difference in talk about the past for parents from different SES backgrounds (Rowe, 2012).

16 of 25

15m ago

Type a message...

11:11

New Chat

Home

That’s so sweet! 😊 Ask Riley, what’s one adventure you and her should go on together?

11:11

New Chat

I love how she is always happy and likes to explore. She always asks questions and is very curious!

Home

👀 Think about the things you admire about Riley.

Share this with Riley.

Saturday, 25 May

17 of 25

18 of 25

19 of 25

Promising field experiment (RCT) results

We experimentally tested the Chat2Learn approach with 600 low-income, English- and Spanish-speaking families in the US over 6 months and it significantly improved:

↑ Children’s vocabulary

↑ Parents’ beliefs that they have the agency to help their kids learn

(Rury, et al., 2025)

20 of 25

Promising lab experiment results

We experimentally tested Chat2Learn with 63 low-income, English- and Spanish-speaking parent-child dyads in US preschools in a 10-minute session during which we recorded their interactions. Access to Chat2Learn significantly improved:

↑ Number of words, complex sentences and diversity of vocabulary used by parents

↑ Decontextualized talk in conversations (imagination, reasoning, abstract and hypothetical discussion)

↑ Joint engagement (physical and language interaction)

(Lu & Kalil, 2025)

21 of 25

Parental feedback, Chat2learn condition

Chat2Learn provides a different way to talk and play with children
Images are helpful
Viewed as fun, age-appropriate, and interesting
Helps to manage the challenges of child engagement (i.e., sustaining attention)
Recognize need for help generating follow-up questions

22 of 25

All Illinois families with children ____years old can participate.

3-6

How to collaborate

?

Family Enrollment

Community Outreach

Enroll all eligible families
Equitable access
Privacy prioritized
Opt-out option for families

83% Retention Rate

Partners promote C2L
Encourages family engagement
Flexible but low participation

12% Enrollment Rate

Want to see my AI powers in action?

23 of 25

Research Team 2025-26

Susan Mayer

Co-Founder

Co-Director

Michelle Michelini

Executive Director

Ariel Kalil

Co-Founder

Co-Director

Sebastian Gallegos

Affiliate

UAI Chile

Dana McCoy

Affiliate

Harvard

Jorge Garcia

Affiliate

Texas A&M

Mauricio Justino

Pre-doc

Derek Rury

FMR Post-doc

Rohen Shah

FMR Harris PhD

Marlis Schneider

NHH

Laudine Carbuccia

Sciences Po

Elena Ziege

BiB

Noah Liu

Harris PhD

Daniela Bresciani

Harris PhD

Sarah Mogan

Associate Director

Katherine Crosby

Assistant Director

Gigi Pacheco

Research Coordinator

Ritika Sethi

Post-doc

Terri Sabol

Affiliate

Northwestern

Christian Bancayan

Research Coordinator

Linxi Lu

Post-doc

24 of 25

25 of 25

Thank you!

Information and sign-up

https://biplab.uchicago.edu/chat2learn/

Research team

https://biplab.uchicago.edu/

Seeking research partners and funding!

meesh@uchicago.edu

25