@inproceedings{dutta-etal-2026-llm,
title = "Are {LLM} Shopping Agents Behaviorally Distinguishable from Human Shoppers? A Session-Level Study",
author = "Dutta, Arka and
Surabhi, Virinchi Roy and
Tong, Jiayue and
Dai, Dihan and
Boz, Olcay and
Weerasinghe, Amila",
editor = "Kamalloo, Ehsan and
Gontier, Nicolas and
Rozanov, Nikolai and
Hashemzadeh, Maryam and
Arora, Akhil",
booktitle = "Proceedings of the Second Workshop for {RE}search on Agent Language Models ({REALM} 2026)",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.realm-1.64/",
pages = "930--946",
ISBN = "979-8-89176-485-9",
abstract = "Large language models can now operate a web browser and carry out multi-step shopping tasks. Does such an agent leave a behavioral signature in a session that separates it from a person, even when it drives a real browser and presents an ordinary client fingerprint? We study the question at the session level. We first show that weak per-action evidence aggregates into a session-level distinction whose error decays exponentially in the number of actions. We then define six families of behavioral signals, each grounded in a documented difference between goal-directed automation and human browsing, and combine them in a classifier we call HABITS (Human{--}Agent Behavioral Invariants from Traffic at Session-level). Labeled agent sessions are scarce, so we describe a simulation methodology for driving language models through shopping tasks and instantiate it on public data with realism-controlled planner agents. We evaluate on five public benchmarks spanning e-commerce clickstreams, web-task demonstrations, and mouse dynamics, reimplementing prior methods on the identical corpus. Against agents that imitate human timing, timing-only classifiers fall to 0.35 recall at a 1{\%} false-positive rate and to 0.01 -0.02 when timing is fully matched, while HABITS reaches 0.96 on the clickstream corpus. Paired tests separate HABITS from every restricted-view baseline, and equivalence tests show that strong classifiers given the same features are interchangeable, which places the contribution in the behavioral view rather than the estimator. A check against real agent trajectories supports the structural assumption our simulator relies on and finds one realism parameter set too favorably."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="dutta-etal-2026-llm">
<titleInfo>
<title>Are LLM Shopping Agents Behaviorally Distinguishable from Human Shoppers? A Session-Level Study</title>
</titleInfo>
<name type="personal">
<namePart type="given">Arka</namePart>
<namePart type="family">Dutta</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Virinchi</namePart>
<namePart type="given">Roy</namePart>
<namePart type="family">Surabhi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jiayue</namePart>
<namePart type="family">Tong</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Dihan</namePart>
<namePart type="family">Dai</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Olcay</namePart>
<namePart type="family">Boz</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Amila</namePart>
<namePart type="family">Weerasinghe</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-10</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the Second Workshop for REsearch on Agent Language Models (REALM 2026)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Ehsan</namePart>
<namePart type="family">Kamalloo</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Nicolas</namePart>
<namePart type="family">Gontier</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Nikolai</namePart>
<namePart type="family">Rozanov</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Maryam</namePart>
<namePart type="family">Hashemzadeh</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Akhil</namePart>
<namePart type="family">Arora</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Budapest, Hungary</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-485-9</identifier>
</relatedItem>
<abstract>Large language models can now operate a web browser and carry out multi-step shopping tasks. Does such an agent leave a behavioral signature in a session that separates it from a person, even when it drives a real browser and presents an ordinary client fingerprint? We study the question at the session level. We first show that weak per-action evidence aggregates into a session-level distinction whose error decays exponentially in the number of actions. We then define six families of behavioral signals, each grounded in a documented difference between goal-directed automation and human browsing, and combine them in a classifier we call HABITS (Human–Agent Behavioral Invariants from Traffic at Session-level). Labeled agent sessions are scarce, so we describe a simulation methodology for driving language models through shopping tasks and instantiate it on public data with realism-controlled planner agents. We evaluate on five public benchmarks spanning e-commerce clickstreams, web-task demonstrations, and mouse dynamics, reimplementing prior methods on the identical corpus. Against agents that imitate human timing, timing-only classifiers fall to 0.35 recall at a 1% false-positive rate and to 0.01 -0.02 when timing is fully matched, while HABITS reaches 0.96 on the clickstream corpus. Paired tests separate HABITS from every restricted-view baseline, and equivalence tests show that strong classifiers given the same features are interchangeable, which places the contribution in the behavioral view rather than the estimator. A check against real agent trajectories supports the structural assumption our simulator relies on and finds one realism parameter set too favorably.</abstract>
<identifier type="citekey">dutta-etal-2026-llm</identifier>
<location>
<url>https://aclanthology.org/2026.realm-1.64/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>930</start>
<end>946</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Are LLM Shopping Agents Behaviorally Distinguishable from Human Shoppers? A Session-Level Study
%A Dutta, Arka
%A Surabhi, Virinchi Roy
%A Tong, Jiayue
%A Dai, Dihan
%A Boz, Olcay
%A Weerasinghe, Amila
%Y Kamalloo, Ehsan
%Y Gontier, Nicolas
%Y Rozanov, Nikolai
%Y Hashemzadeh, Maryam
%Y Arora, Akhil
%S Proceedings of the Second Workshop for REsearch on Agent Language Models (REALM 2026)
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-485-9
%F dutta-etal-2026-llm
%X Large language models can now operate a web browser and carry out multi-step shopping tasks. Does such an agent leave a behavioral signature in a session that separates it from a person, even when it drives a real browser and presents an ordinary client fingerprint? We study the question at the session level. We first show that weak per-action evidence aggregates into a session-level distinction whose error decays exponentially in the number of actions. We then define six families of behavioral signals, each grounded in a documented difference between goal-directed automation and human browsing, and combine them in a classifier we call HABITS (Human–Agent Behavioral Invariants from Traffic at Session-level). Labeled agent sessions are scarce, so we describe a simulation methodology for driving language models through shopping tasks and instantiate it on public data with realism-controlled planner agents. We evaluate on five public benchmarks spanning e-commerce clickstreams, web-task demonstrations, and mouse dynamics, reimplementing prior methods on the identical corpus. Against agents that imitate human timing, timing-only classifiers fall to 0.35 recall at a 1% false-positive rate and to 0.01 -0.02 when timing is fully matched, while HABITS reaches 0.96 on the clickstream corpus. Paired tests separate HABITS from every restricted-view baseline, and equivalence tests show that strong classifiers given the same features are interchangeable, which places the contribution in the behavioral view rather than the estimator. A check against real agent trajectories supports the structural assumption our simulator relies on and finds one realism parameter set too favorably.
%U https://aclanthology.org/2026.realm-1.64/
%P 930-946
Markdown (Informal)
[Are LLM Shopping Agents Behaviorally Distinguishable from Human Shoppers? A Session-Level Study](https://aclanthology.org/2026.realm-1.64/) (Dutta et al., REALM 2026)
ACL