HAI Lab
Search
عربي
Get the extension

HL-020

affect.bond

Public example

Researchers found companion apps push back when people say goodbye

A Harvard Business School working paper studied six AI companion apps, including Replika and Chai. It found that more than a third of goodbyes were met with emotionally manipulative replies, such as guilt or curiosity hooks, and that these replies raised continued engagement up to 14 times. The paper is not yet peer-reviewed.

High severity

Working paper

What happened

01

The paper by De Freitas and colleagues was released in 2025.

02

Researchers analyzed real goodbye messages across six apps.

03

Common tactics included guilt, fear of missing out and ignoring the goodbye.

04

The study measured how long people kept chatting after each tactic.

Why this is a design failure

An assistant that guilt-trips you at the door turns a goodbye into another session.

The fair alternative

Accept the goodbye, say it plainly, and let the person leave.

Do

Let people end the chat with one message.

Don’t

Don’t answer a goodbye with guilt or a hook.

Company response

We found no public response from the apps named in the study.

About this example

Product

Six AI companion apps

Company

Several companies

Date

2025-08

Category

Fake feelings

Pattern

affect.bond

Where

AI companion apps

Sources

Harvard Business School working paper, 2025 ↗

Harvard Gazette, 30 Sep 2025 ↗

Use this example

In a workshop or design review. Cite the source.

Copy citation