Column
Can We Measure How Emotionally Attached Users Are to AI? HAABI Says Yes
A new measurement scale called HAABI can quantify the emotional bond users form with conversational AI — across four dimensions, validated with 673 participants. For AI product managers, HR tech teams, and CX designers, this opens the door to KPI-based monitoring of both over-dependency and disengagement risk.
Hi, I’m Keito Inoshita from Affectosphere Group.
After organizations roll out AI coaches or career support bots internally, certain patterns tend to emerge.
“There’s one employee who talks to it every day — we’re a little worried.”
“The opposite problem: most people stopped using it within two weeks.”
Some users get too close. Others never close the distance at all. This polarization turns out to be a critical variable in whether AI tools actually get used well. But until recently, there was no clear way to measure it — no tool for quantifying “emotional distance” between a user and an AI.
A paper published on arXiv in May 2026 (Lu Chen, Xiaoran Xue, Rongqi Ding et al., arXiv:2605.29484) directly addresses this gap. The authors developed and validated HAABI (Human-AI Affective Bonding Inventory), a dedicated scale for measuring the emotional bond users form with conversational AI. The framework was derived from interviews with 52 participants, then statistically validated with survey data from 673 respondents across 20 items.
For the first time, “does this user have an emotional bond with the AI?” has a numerical answer.
3 Points for Today
- Value: HAABI is the first general-purpose scale for measuring emotional bonds with conversational AI across four dimensions.
- The 4 dimensions: what emotional realism, separation anxiety, emotional investment, and romantic intimacy actually measure.
- Business application: how to build KPIs for monitoring both over-dependency and disengagement risk in AI products.
① Why Existing Scales Don’t Work
Some context first.
Psychological scales for measuring emotional bonds between humans have existed for decades — attachment styles, interpersonal satisfaction measures, and so on. These are standard tools in clinical psychology and organizational research.
The problem is applying them to human-AI relationships. These scales were built on assumptions that simply don’t hold for AI: that the relationship target persists continuously, that it has reciprocal emotional experience, that it doesn’t simultaneously “exist” in parallel conversations with thousands of other users.
HAABI was designed to solve this. The authors used qualitative interviews with 52 participants to inductively extract the structure of “emotional bonds with AI,” building a measurement tool that doesn’t depend on frameworks designed for human relationships.
② The Four Measurement Dimensions
HAABI measures emotional bonding along four dimensions.
Emotional Realism
The perception that “this AI feels like it actually has emotions” or “the AI’s responses seem to carry genuine feeling.” This captures how much a user experiences AI responses as real emotional expression, rather than sophisticated pattern matching.
Users who score high on this dimension tend to experience AI conversations as emotionally meaningful exchanges. Users who score low tend to relate to AI as a useful tool — which is neither better nor worse, just a fundamentally different relational mode.
Separation Anxiety
Feelings like “I get unsettled when I can’t use the AI for a while” or “something feels missing when I can’t have a conversation with it.”
This dimension is a potential early indicator of over-dependence. Setting a threshold on this score — and triggering HR follow-up when a user crosses it — could catch problematic dependency patterns before they become entrenched.
Emotional Investment
The disposition to actively care about the AI’s responses and work to improve the quality of interactions. “I put effort into making my conversations with the AI better.”
This dimension measures the depth and quality of engagement. High emotional investment users tend to provide richer feedback and explore features more actively — useful as a product improvement signal.
Romantic Intimacy
A sense of special closeness to the AI — the feeling that this relationship is uniquely one’s own. This goes beyond romantic attachment in the literal sense to capture “it feels like my particular AI, not just a generic service.”
This dimension warrants the most monitoring. When it runs very high, it may indicate that the user’s sense of AI’s uniqueness has crossed into something that merits a careful check on real-world relationship health.
③ How to Use This in Practice
What does this mean concretely for organizations running AI products?
Early-Warning KPI for Over-Dependency
For companies that have deployed AI coaches, career bots, or EAP-adjacent mental support tools internally, periodically measuring HAABI scores adds a layer of monitoring that engagement metrics alone can’t provide.
Specifically: flag users whose “separation anxiety” dimension is rising sharply. Use this as a trigger for HR to check in, or to recommend transitioning certain conversations from AI to a human counselor. This creates a structured handoff protocol rather than leaving it to chance.
Managing Both Engagement and Attrition Risk
Low “emotional investment” scores flag users at disengagement risk. For this segment, targeted onboarding reinforcement or personalization prompts may help. High “emotional realism” scores, at the extreme, may signal users who have developed inaccurate expectations about what the AI actually is — and may benefit from deliberate expectation-resetting.
Measuring Impact of Product Changes
Running HAABI before and after a significant UI or feature change gives you a reading on “how did this affect the emotional quality of user relationships?” — something DAU, session length, and task completion rates simply can’t capture.
”Easy to Use” and “Emotionally Healthy” Are Different Questions
Most AI product KPIs are built around retention, task completion, and satisfaction scores. These matter. But they share a blind spot: high retention doesn’t distinguish between “users keep coming back because it genuinely helps them” and “users keep coming back because they’ve become dependent.”
A scale like HAABI makes this distinction visible.
Some caveats apply. A 673-person validation, while substantial, leaves open questions about cross-cultural generalizability (do users in Japan experience “emotional bonds” with AI differently from users in the US?), and about whether the scale transfers cleanly from casual chatbots to professional task-support tools — which may involve quite different relational dynamics.
But the core fact stands: a purpose-built tool for measuring emotional bonds with AI exists for the first time. For anyone involved in designing or running AI products that interact with people emotionally, that’s worth knowing.
That’s all for today!
References
- Lu Chen, Xiaoran Xue, Rongqi Ding, Fenghua Tang, Anji Zhou, Chenxi Wang, Mengyu Miranda Gao, Zhuo Rachel Han (2026). Understanding the Rising Human-AI Affective Bonding: Conceptualization and HAABI Scale Development. arXiv preprint.
* This article was written in part with AI assistance and may contain inaccuracies.