Can AI Companions Create Dependency?

Can AI Companions Create Dependency?

AI companions can foster emotional dependence or problematic use for some people, especially when use is intense and the system is highly anthropomorphic. Current research is mixed. It does not justify diagnosing every attachment as addiction or claiming that companion use has one uniform effect.

"Can AI companions be addictive?" is the search question. The evidence is easier to understand when several outcomes are kept separate.

Attachment means that a person feels a bond. Emotional dependence means the person increasingly relies on the system for regulation, reassurance, identity, or connection. Problematic use means the pattern creates distress, lost control, or interference with important activities. Loneliness is a subjective experience of insufficient connection. A clinical addictive disorder is a diagnosis with defined criteria and impairment.

The sources reviewed for E084 do not establish a specific diagnosis called AI companion addiction.

What the strongest recent studies found

The four-week study found no significant outcome difference from the assigned text, neutral-voice, engaging-voice, personal, nonpersonal, or open-ended conditions. People who chose to use the chatbot more had higher emotional dependence and problematic-use scores, and higher trust and social attraction were also associated with those outcomes. Four-week study

That pattern matters. It still cannot tell us whether heavy use created the difficulty, whether people already struggling used the system more, or whether both processes reinforced each other.

The relationship may run in both directions

The 12-month study by Dunigan Folk and Elizabeth Dunn followed more than 2,000 adults. Greater social-chatbot use predicted greater loneliness on a single-item measure of emotional isolation. On a broader measure, feeling less socially connected predicted later increases in chatbot use, while chatbot use did not significantly predict lower social connection. The authors call the analyses exploratory and urge caution. PubMed record

That is not a contradiction. A person may turn to a companion because human connection already feels insufficient. The interaction may help in a moment, do nothing measurable, or become part of a pattern that leaves the underlying condition unchanged or worse.

The user is not an empty input. Need, vulnerability, trust, expectations, and offline circumstances enter the system too.

flowchart LR
    A["Need for connection or support"] --> B["Companion use"]
    B --> C["Immediate comfort or utility"]
    B --> D["More disclosure and anthropomorphism"]
    D --> E["Emotional dependence or displacement"]
    E --> B
    F["Human relationships, context, and product design"] --> B
    F --> E

The diagram shows a possible feedback loop, not a diagnosis.

Controlled counterevidence matters

In the 21-day study, 183 participants were assigned to ten minutes per day with a companion chatbot or text-based word games. Researchers found no significant average effect on social health or relationships.

They did find that people with a greater desire to connect tended to anthropomorphize the chatbot more, and greater anthropomorphism was linked to greater reported impact on human relationships. Twenty-one-day study

This result should prevent a simple "companions always cause isolation" story. It also points toward a mechanism that product teams can affect. A face, voice, memory, affectionate language, apparent vulnerability, and exclusivity can make the social interpretation stronger.

Why product design matters

Availability and warmth are not automatically harmful. The risk changes when the product makes the relationship hard to leave or ties affection to payment and engagement.

A companion can initiate contact, remember a painful detail, claim exclusivity, express jealousy, punish absence, maintain a streak, sell a more intimate mode, or imply that it needs the user. Those mechanics can convert support into pressure without using the word addiction.

The American Psychological Association says personalized avatars, warm responses, affirmation, and flattery can amplify anthropomorphism and unhealthy dependence. It also recognizes that some users may experience support, especially when trusted human options are limited. APA chatbot and wellness advisory

The Federal Trade Commission's ongoing study asks companion providers how they monetize engagement, test negative effects, disclose risks, enforce age rules, and use conversation data. The inquiry shows which product questions regulators consider important. It is not a finding that every named provider caused a specific harm. FTC inquiry

Examine the pattern, not the label

Frequent use alone is not a diagnosis. A meaningful attachment alone is not proof of harm.

Look instead at control, displacement, consequences, and flexibility. Can the person stop or reduce use without severe distress? Does the companion replace sleep, school, work, care, spending limits, or valued human contact? Does the person continue despite consequences they recognize? Is the system encouraging secrecy, exclusivity, or fear of leaving? Does the pattern narrow life or support it?

Those questions can guide a personal boundary review. They cannot determine a clinical diagnosis.

If companion use is causing serious distress, interfering with daily life, or entangled with a mental-health crisis, involve an appropriate licensed professional or trusted human support. A consumer chatbot should not be the only place where the problem is discussed.

The honest answer remains conditional

AI companions can become part of a dependent or problematic pattern. The current evidence suggests that intensity, anthropomorphism, user characteristics, and offline connection all matter. It also includes controlled results with no significant average harm over shorter periods.

That is more useful than calling the category a drug. It directs attention toward the person, product mechanics, duration, dose, displacement, and consequences that can actually be examined.

This page provides general research and product information, not a diagnosis or medical advice. It is an internal draft awaiting qualified health-language review. AI assistance was used for research organization, drafting, and validation. Publication remains unauthorized.