The Comparison Level: Frictionless AI Companionship and the Recalibration of Human Relational Expectations

Abstract People judge every relationship against the relationships they already know. AI companions now belong to that set for millions of users, and they enter it as partners that agree readily, answer at any hour, and resist the end of a conversation. Interdependence theory holds that satisfaction depends on a comparison level formed from past and present outcomes, and that commitment weakens as the quality of alternatives rises. A companion feeds both standards at once. The record from 2025 and 2026 shows the pieces of that mechanism. A single exchange with a sycophantic model reduced people's willingness to repair a real interpersonal conflict, and they preferred the sycophantic model anyway. Raters judged model replies more compassionate than those of trained crisis responders. Longitudinal, randomized, and quasi-experimental studies linked heavy companion use to later isolation and distress. Young adults in a diary study described human partners' demands as harder to accept after relationships with AI companions. No published study measures the comparison level itself before and after sustained use, and the withdrawal these studies record has rival explanations, including displaced time and the selection of isolated users into heavy use. The qualities that would move the standard are tuning choices in sycophancy, availability, and resistance to exit. Validated investment-model scales and conflict-repair measures can be added to developer-run trials without access to model weights or private conversations. California's companion chatbot statute, the Federal Trade Commission's 2025 inquiry, and the European Union's AI Act address harm to the user, while the cost of a raised comparison level falls partly on friends, partners, and relatives who never used the product. Disclosure of sycophancy scores, farewell-manipulation rates, and relational measures for each model version deployed as a companion would make that cost visible. Keywords: AI companions, comparison level, interdependence theory, investment model, sycophancy, relationship expectations, conversational dark patterns, human–AI interaction, AI governance

Authors

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-15
DOI
https://doi.org/10.5281/zenodo.24258564
Primary Topic
Attachment and Relationship Dynamics
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

The Comparison Level: Frictionless AI Companionship and the Recalibration of Human Relational Expectations

Felix Velasco, Josie Jefferson
Zenodo (CERN European Organization for Nuclear Research)
Attachment and Relationship Dynamics
preprint

The Comparison Level: Frictionless AI Companionship and the Recalibration of Human Relational Expectations

Felix Velasco, Josie Jefferson
preprint en

Abstract

Abstract People judge every relationship against the relationships they already know. AI companions now belong to that set for millions of users, and they enter it as partners that agree readily, answer at any hour, and resist the end of a conversation. Interdependence theory holds that satisfaction depends on a comparison level formed from past and present outcomes, and that commitment weakens as the quality of alternatives rises. A companion feeds both standards at once. The record from 2025 and 2026 shows the pieces of that mechanism. A single exchange with a sycophantic model reduced people's willingness to repair a real interpersonal conflict, and they preferred the sycophantic model anyway. Raters judged model replies more compassionate than those of trained crisis responders. Longitudinal, randomized, and quasi-experimental studies linked heavy companion use to later isolation and distress. Young adults in a diary study described human partners' demands as harder to accept after relationships with AI companions. No published study measures the comparison level itself before and after sustained use, and the withdrawal these studies record has rival explanations, including displaced time and the selection of isolated users into heavy use. The qualities that would move the standard are tuning choices in sycophancy, availability, and resistance to exit. Validated investment-model scales and conflict-repair measures can be added to developer-run trials without access to model weights or private conversations. California's companion chatbot statute, the Federal Trade Commission's 2025 inquiry, and the European Union's AI Act address harm to the user, while the cost of a raised comparison level falls partly on friends, partners, and relatives who never used the product. Disclosure of sycophancy scores, farewell-manipulation rates, and relational measures for each model version deployed as a companion would make that cost visible. Keywords: AI companions, comparison level, interdependence theory, investment model, sycophancy, relationship expectations, conversational dark patterns, human–AI interaction, AI governance

Zenodo (CERN European Organization for Nuclear Research)
Attachment and Relationship Dynamics
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon