WaLLM -- Understanding Use and Engagement with a General-Purpose LLM on WhatsApp

Hiba Eltigani; Rukhshan Haroon; Asli Kocak; Abdullah Bin Faisal; Noah Martin; Fahad Dogar

cs.HCMay 13, 2025

WaLLM -- Understanding Use and Engagement with a General-Purpose LLM on WhatsApp

Authors: Hiba Eltigani, Rukhshan Haroon, Asli Kocak, Abdullah Bin Faisal, Noah Martin, Fahad Dogar

Organizations: Tufts University, USA, Medford

Abstract

Large language model (LLM) chatbots are increasingly reaching users through messaging platforms (e.g. WhatsApp). However, these systems remain largely proprietary and opaque, while academic research has focused on narrow, domain-specific assistants. This leaves open questions about how people use general-purpose LLMs and how such systems should be designed. To address this gap, we developed WaLLM, a general-purpose LLM chatbot, and deployed it on WhatsApp as a design probe to study open-ended AI use in the wild. Our findings show that health and well-being accounted for the largest proportion of queries, suggesting that users turned to WaLLM for advice and information. Engagement features varied in their adoption and associated patterns of use: proactive communication supported the service's visibility and correlated with higher user activity, while communal lists facilitated content discovery. We report how these features were adapted to WhatsApp's affordances and discuss implications for designing general-purpose LLM services over messaging platforms.

Figures & tables

Figure 1 . A WhatsApp user asking a question and WaLLM responding, with buttons that support usability (e.g., Menu ) and engagement (e.g., Suggest Follow-ups ). Example of a WhatsApp user asking a question and \chatbot{} responding with an LLM-generated answer. The response message has additional buttons that facilitate usability (e.g., ``Menu'') and engagement (e.g., ``Suggest Follow-ups''). — Figure 1 . A WhatsApp user asking a question and WaLLM responding, with buttons that support usability (e.g., Menu ) and engagement (e.g., Suggest Follow-ups ). Example of a WhatsApp user asking a question and \chatbot{} responding with an LLM-generated answer. The response message has additional buttons that facilitate usability (e.g., `Menu'') and engagement (e.g.,` Suggest Follow-ups'').

Panel of Figure 2 — Figure 2 . Examples of WaLLM’s interactive messages

	One-time users	Casual users	Regular users
# of sessions per user	1	2 - 100	more than 100
# of users	17	71	8
Total sessions	17	706	3,229
Total interactions	54	2,820	14,534
Total freeform queries	22	927	4,722
Total interactive queries	29	1,720	8,117

Table 1 . Categories of users based on activity

Category	% of Queries
Health and Well-being	28%
Cultural and General Knowledge	17%
Science and Technology	11%
Language and Communication	10%
Social and Personal Development	9%
Commerce and Economy	8%

Table 2 . Distribution of User Queries by Category (Sampled Data)

Purpose	Type	Intent	Example	% in First Session (n)	% of Sample (n)
Question	Factual	Information	What is excise duty?	44% (79)	31% (112)
		Verification	can they put MRNA vaccine in vegetables?	8% (15)	12% (43)
		Explanation	how to source high quality bedding sheets for bedding company business startup?	6% (11)	12% (44)
	Non-factual	Advice	how long after gall stone surgery I can travel?	11% (19)	14% (52)
		Viewpoint	what are the cons of being the oldest child?	18% (32)	9% (34)
		Communication and Language	fancy words for exercise?	1% (2)	5% (19)

Table 3 . Categories of freeform queries *

Feature	One Time	Casual	Regular
Continue Reading (Freeform Queries)	7%	23%	25%
Top-rated Questions	62%	25%	14%
Recent Questions	0%	3%	13%
Contribution Leaderboard	4%	8%	16%
Suggest Follow-ups	10%	16%	9%
Get Better Answer	17%	25%	23%

Table 4 . Distribution of interactive queries per user group across interactive features

Appendix figures & tables2 assets

Supplementary material from the paper’s appendix.

Appendix

Regular Users
Top-rated Questions	Topic Category	Chosen x Times
How can we provide emotional care for our parents in their old age?	Social and Personal Development	7
In which fields and subjects is a PhD worth pursuing in this day and age?	Education and Learning	5
I used to be able to memorize the names of every student in my class, but that is no longer the case. What can I do?	Social and Personal Development	4
What are the main reasons for divorce around the world?	Social and Personal Development	4
Is becoming a doctor suitable for everyone, considering the finances, intellect, and time required?	Social and Personal Development	4

Appendix

Table 5 . Regular users top-rated questions

Casual Users
Top-rated Questions	Topic Category	Chosen x Times
What are the psychological effects on children of broken marriages?	Health and Well-being	9
Why do we say "sleep like a baby" when babies wake up crying every few hours?	Language and Communication	9
Are late marriages better than early marriages?	Social and Personal Development	9
Is it okay for parents to push their children into mastering a skill from an early age, or should children be given the freedom to pursue their own interests?	Social and Personal Development	8
How can we provide emotional care for our parents in their old age?	Social and Personal Development	8

Appendix

Table 6 . Casual users top-rated questions

Explore similar work

Adopt $\neq$ Adapt: Longitudinal Analyses of LLM Conversations in the Wild
May 27, 2026Rebecca M. M. Hicke, Kiran TomlinsonMulti-Turn Large Language Model Conversations Copilot
Cornell University · Microsoft Research
Priming, Path-dependence, and Plasticity: Understanding the molding of user-LLM interaction and its implications from (many) chat logs in the wild
May 7, 2026Shengqi Zhu, Jeffrey M. Rzeszotarski, David MimnoMulti-Turn Large Language Model Conversations Chatbots
Cornell University · Ithaca, NY, USA · Loyola University Maryland +1
LLMs as Oracles: Reliance on LLMs for Subjective Personal Questions
Sep 13, 2026Myra Cheng, Lujain Ibrahim, Grace Liu +5Oracles Deliberation
University of Oxford Oxford, England · Carnegie Mellon University Pittsburgh, Pennsylvania, USA

May 27, 2026, cs.AIJ/K move · Enter open · S save

Adopt $\neq$ Adapt: Longitudinal Analyses of LLM Conversations in the Wild

Rebecca M. M. Hicke, Kiran Tomlinson

Cornell University · Microsoft Research

Multi-Turn Large Language Model Conversations Copilot Longitudinal Data Heterogeneity

Open paper AskPDF arXiv

Although a growing body of research has begun to describe user--LLM interactions, the picture it paints is largely static; little is known about how individual users change their behavior over time. To address this gap, we analyze the conversational trajectories of ~12,000 randomly sampled Microsoft Bing Copilot users and compare these with data from WildChat-4.8M. While the Copilot data contains significant population-level trends, we find that trends in individual user trajectories are much weaker; user habits prove to be overwhelmingly sticky. We also find stark differences between users of different activity levels: more active users have more successful conversations and use the LLM for more complex and professionally oriented tasks. Some user trends also appear in WildChat-4.8M, but we find evidence that this dataset is significantly skewed towards highly proficient "power" users. Ultimately, our results suggest that existing user behavior is difficult to change and demonstrate the extent of user heterogeneity. Our comparison between datasets highlights that WildChat does not represent typical user--AI interactions, an important caveat for downstream uses of the data.

Abstract

Explore similar work

Adopt ≠\neq= Adapt: Longitudinal Analyses of LLM Conversations in the Wild

Priming, Path-dependence, and Plasticity: Understanding the molding of user-LLM interaction and its implications from (many) chat logs in the wild

LLMs as Oracles: Reliance on LLMs for Subjective Personal Questions

Adopt $\neq$ Adapt: Longitudinal Analyses of LLM Conversations in the Wild