AI-Assisted Writing for Teaching Reading and English as a Second Language (ESL)

How to cite

Koutout, Kristine; Athey, Susan; Wang, Yuyan. 2026. "AI-Assisted Writing for Teaching Reading and English as a Second Language (ESL)." AEA RCT Registry. AEARCTR-0017611. https://doi.org/10.1257/rct.17611-2.0

Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.

Kristine Koutout* , Susan Athey , Yuyan Wang

* Primary investigator

Registry ID
AEARCTR-0017611
DOI
https://doi.org/10.1257/rct.17611-2.0
First registered
Last updated
Status
Completed
Trial dates
2026-01-18 – 2026-05-02
Countries
India ()
Registry record
http://www.socialscienceregistry.org/trials/17611

Abstract

Teaching children reading and English as a second language (ESL) skills---skills that provide lifelong returns---requires access to engaging, contextually relevant materials. For children outside of Western contexts, English-language story options often lack familiar settings, norms, problems, and characters. We aim to learn whether stories written with generative AI assistance can fill that gap by creating context-relevant stories that better engage children in India.

Intervention

Our intervention delivers stories written with AI-assistance through an edtech app designed for reading and ESL skills.

Experimental design

An edtech firm will implement an experiment that randomizes users to see different stories on their app for improving reading and ESL skills in two stages, where stages are defined relative to a user's first engagement during the experimental period. In Stage 1, the intervention impacts the stories that users see in the top two "trays'' in the app, where trays are horizontally scrollable story icons from which users choose stories to read. Users will be randomized into three groups: (1) users in the "AI Stories'' group see stories written with AI-assistance; (2) users in the "Matched Stories'' group see stories selected to match the AI Stories on word count (while maintaining similar quality to average stories on the app); and (3) users in the "Baseline'' group continue to see the standard traditionally sourced stories. In Stage 1, users assigned to see AI Stories will only see the Story 1's from each series of AI Stories, so that we can separately identify the role of AI-assisted writing on a single, standalone story versus a series. Stage 1 will answer research questions on whether users engage more with stories generated using AI assistance relative to traditionally sourced stories.

Treatment assignments are persistent in Stage 2; that is, users remain in the same group across both stages. The nature of the intervention changes in Stage 2 for the AI Stories and the Matched Stories groups; in particular, the intervention impacts the stories that are seen throughout the tabs on the app dedicated to stories. Users in the AI Stories group see AI Stories from the full series. Users in the Matched Stories group see traditionally sourced series on the app. Users in the Baseline group continue to see the standard traditionally sourced stories. This stage will answer research questions about how exposure to AI Stories impacts overall engagement with the app, and allow us to differentiate between the impact of the series produced by the AI Stories product from the impact of series more generally, since the AI Stories and Matched Stories groups both see stories that belong to series. Users may increase engagement if they enjoy a series, as they become invested in the premise and know what to expect. On the other hand, the focus on series will decrease the overall diversity of content relative to the baseline which involves a mix of series and standalone stories.

Design details

Randomization method
Randomization done using a computer.
Randomization unit
We separately randomize users who are associated with a school and users who are not. For users with school ids, we randomize users at the school level. Otherwise, we randomize users at the individual level.
Clusters
90
Observations
8100
Arms
AI Stories - 40% or 3,240 users Matched Stories - 40% or 3,240 users Baseline - 20% or 1,620 users
Primary outcomes
Stage 1 1. binary outcome for clicking a story in either of two trays on the edtech app in the one week following first log in during the experimental period 2. binary outcome for completing a story in either of two trays on the edtech app in the one week following first log in during the experimental period 3. binary outcome for completing a story, conditional on clicking a story, in either of two trays on the edtech app in the one week following first log in during the experimental period Stage 2 1. number of stories clicked on the edtech app during a three-week period 2. number of stories completed on the edtech app during a three-week period 3. story engagement index on the edtech app during a three-week period
Secondary outcomes
1. number of stories liked 2. probability of clicking an additional story, conditional on completing a story

Cited by

No citation data available for this registration yet.

Keywords