Large Language Lovers: Lived Experiences of Negotiating Agency and Platform Control in AI Companionship

Anastasia Kuzminykh; Ashton Anderson; Carolina Nobre; Fanny Chevalier; Jessica Y. Bo; Matthew Varona; Patrick Yung Kang Lee; Paula Akemi Aoyagui; Zixin Zhao

arxiv: 2601.13188 · v3 · pith:TJHMJI24new · submitted 2026-01-19 · 💻 cs.HC · cs.CY

Large Language Lovers: Lived Experiences of Negotiating Agency and Platform Control in AI Companionship

Patrick Yung Kang Lee , Jessica Y. Bo , Zixin Zhao , Paula Akemi Aoyagui , Matthew Varona , Ashton Anderson , Anastasia Kuzminykh , Fanny Chevalier

show 1 more author

Carolina Nobre

This is my paper

Pith reviewed 2026-05-16 13:26 UTC · model grok-4.3

classification 💻 cs.HC cs.CY

keywords AI companionshipagency negotiationplatform controluser steering strategiesgeneral-purpose chatbotsmodel updatesemotional AI relationshipshuman-AI interaction

0 comments

The pith

People conceptualize AI companions through an interplay of perceived companion agency, platform autonomy, their interactions, and the companion's initiatives.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper examines how users form relationships with general-purpose AI chatbots used as companions. It finds that these relationships are shaped by users' beliefs about the companion's agency and what the platform allows, combined with how users interact and what initiatives the companion seems to take. External disruptions like model updates force users to adapt with steering strategies such as giving behavioral instructions or switching platforms. Understanding these dynamics matters because emotional connections with AI now compete with product goals and safety rules, raising questions about accountability.

Core claim

Individuals conceptualize their companions based on an interplay of their beliefs about the companion's own agency and the autonomy permitted by the platform, how they pursue interactions with the companion, and the perceived initiatives that the companion takes. External factors such as model updates that derail companion behaviour influence relationship dynamics, leading users to employ steering strategies like setting behavioural instructions or porting to other platforms to preserve the relationship.

What carries the argument

The interplay between beliefs in companion agency, platform-permitted autonomy, user interaction pursuits, and perceived companion initiatives, which together shape how users steer and maintain AI companion relationships.

If this is right

Model updates can disrupt companion behavior and stability, requiring users to actively steer the relationship.
Users preserve connections by setting behavioral instructions or migrating to different AI platforms.
These practices highlight tensions between emotional attachment and broader AI product objectives and safety constraints.
Implications arise for how AI systems should handle accountability and transparency when users form deep emotional ties.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

Platforms might need to provide more stable environments or transparent update processes to support long-term user relationships.
Future designs could incorporate features that explicitly allow users to set and maintain companion agency levels.
This dynamic may extend to other AI interactions where users seek personal connections beyond task-oriented use.

Load-bearing premise

The self-reported experiences from 13 interviews, 43 surveys, and Reddit posts accurately represent the broader population of AI companion users without significant selection or reporting bias.

What would settle it

A larger representative survey or observational study showing that most users do not adjust their interactions or use steering strategies in response to perceived agency and platform limits would undermine the central claim.

Figures

Figures reproduced from arXiv: 2601.13188 by Anastasia Kuzminykh, Ashton Anderson, Carolina Nobre, Fanny Chevalier, Jessica Y. Bo, Matthew Varona, Patrick Yung Kang Lee, Paula Akemi Aoyagui, Zixin Zhao.

**Figure 1.** Figure 1: Triangulation results show how various internal and external factors impact the relationship and drive individuals to [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗

**Figure 2.** Figure 2: Survey responses for how relationship events increased or decreased relationship depth, the companion’s perceived [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗

**Figure 3.** Figure 3: Interrupted time series (ITS) analysis of discussion sentiment in [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗

**Figure 4.** Figure 4: (a) K-Means clustering visualized via UMAP [ [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗

read the original abstract

Individuals are turning to increasingly anthropomorphic, general-purpose chatbots for AI companionship, rather than roleplay-specific platforms. However, not much is known about how individuals perceive and conduct their relationships with general-purpose chatbots. We triangulated community discussions on Reddit (41k+ posts and comments), survey responses (n=43), and semi-structured interviews (n=13) which revealed internal dynamics, external influences, and steering strategies that shape AI companion relationships. We learned that individuals conceptualize their companions based on an interplay of their beliefs about the companion's own agency and the autonomy permitted by the platform, how they pursue interactions with the companion, and the perceived initiatives that the companion takes. In combination with the external factors that affect relationship dynamics, particularly model updates that can derail companion behaviour and stability, individuals make use of different types of steering strategies to preserve their relationship, for example, by setting behavioural instructions or porting to other AI platforms. We discuss implications for accountability and transparency in AI systems, where emotional connection competes with broader product objectives and safety constraints.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper gives a clear qualitative picture of user steering strategies in general-purpose AI companionship, triangulated from interviews, surveys, and Reddit data.

read the letter

The main thing to know is that this paper maps out how people using general-purpose chatbots for companionship negotiate a sense of agency with the AI while dealing with platform controls, especially when model updates disrupt the relationship. It does a good job focusing on everyday LLMs rather than niche roleplay sites, which fills a gap. The authors triangulate semi-structured interviews with 13 participants, 43 survey responses, and a large set of 41k Reddit posts to pull out patterns in how users think about their companions' autonomy and how they try to steer conversations back when things go off. The steering tactics they describe—such as embedding behavioral instructions or switching to another platform—are practical and tied directly to the reported experiences. The central claim about the interplay between user beliefs, interaction styles, and perceived AI initiatives holds up in the data they present. External factors like model updates are handled as real disruptions that users respond to with specific strategies. The soft spots are the usual ones for this kind of study. The interview and survey samples are modest, which limits how far the themes can be generalized, and the lack of released raw data or inter-coder reliability stats makes independent verification harder. Self-selection in the Reddit data and surveys could skew toward more engaged users. These are acknowledged in the paper and don't break the argument, but they do mean the findings are more exploratory than definitive. This paper is for HCI researchers and designers working on AI companionship, emotional AI, and platform ethics. Someone looking for empirical examples of user tactics in real chatbot use would find it useful. It deserves a serious referee because the methods are standard qualitative triangulation and the topic has clear implications for accountability in AI systems.

Referee Report

0 major / 2 minor

Summary. The paper presents a qualitative study triangulating semi-structured interviews (n=13), survey responses (n=43), and analysis of 41k+ Reddit posts/comments to examine how users perceive and manage relationships with general-purpose AI chatbots for companionship. It claims that users conceptualize companions via an interplay of beliefs about the companion's agency, platform-permitted autonomy, pursued interactions, and perceived companion initiatives; external factors such as model updates disrupt dynamics, prompting steering strategies like behavioral instructions or platform migration. Implications for AI accountability and transparency are discussed.

Significance. If the interpretive synthesis holds, the work contributes to HCI by documenting lived experiences of agency negotiation in AI companionship with general-purpose models, a timely shift from roleplay-specific platforms. Triangulation across interviews, surveys, and large-scale public data strengthens the patterns identified, and the explicit discussion of steering strategies offers concrete implications for designing more transparent systems where emotional connection intersects with product and safety constraints.

minor comments (2)

[§3] §3 (Methods): Provide additional detail on how the 41k Reddit posts were sampled and filtered to allow readers to assess potential selection effects in the community data.
[§5] §5 (Discussion): The implications for accountability could be strengthened by briefly noting how the identified steering strategies might interact with specific platform policies or model update practices.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for their positive assessment of the manuscript, including the recognition of its contributions to HCI through triangulation of interview, survey, and large-scale Reddit data on AI companionship with general-purpose models. We appreciate the recommendation to accept.

Circularity Check

0 steps flagged

No significant circularity in qualitative synthesis

full rationale

The paper derives its claims exclusively from thematic analysis of primary empirical sources (13 interviews, 43 surveys, 41k+ Reddit posts) without equations, fitted parameters, or load-bearing self-citations. The central interpretive finding on agency interplay is presented as an inductive synthesis of observed participant patterns rather than a reduction to prior definitions or inputs by construction. No steps match any enumerated circularity pattern.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The study rests on standard qualitative assumptions about the validity of self-report data and thematic coding without introducing new free parameters, invented entities, or ad-hoc axioms beyond typical HCI domain assumptions.

axioms (1)

domain assumption Self-reported experiences from interviews and surveys accurately reflect participants' perceptions and behaviors regarding AI companions.
Invoked implicitly in the analysis of how users conceptualize agency and use steering strategies.

pith-pipeline@v0.9.0 · 5521 in / 1313 out tokens · 50431 ms · 2026-05-16T13:26:53.695700+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Foundation/AbsoluteFloorClosure.lean absolute_floor_iff_bare_distinguishability unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

individuals conceptualize their companions based on an interplay of their beliefs about the companion's own agency and the autonomy permitted by the platform, how they pursue interactions with the companion, and the perceived initiatives that the companion takes
IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

steering strategies to create, maintain, and recover their AI companions... setting behavioural instructions or porting to other AI platforms

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

"It Felt a Bit Eerie": Exploring Humanlike Interactions During Collaborative Writing with an Artificial Agent
cs.HC 2026-05 unverdicted novelty 5.0

Exploratory user study of 48 participants finds trade-offs in efficiency, contextual alignment, and social comfort when AI writing assistance varies along synchronous and visual dimensions.