Source-checked research draft: this page remains excluded from search until its exact figures, denominators, wording, and source boundaries receive accountable review.

AI companion statistics · assessed

Four widely useful numbers. Four different questions.

27%, 72%, 2.9%, and 1,131 are not competing estimates of the same thing. They describe a broad adult-use screen, a teen survey, conversation-level activity on one general AI platform, and an active-user study. Keep the denominator attached and the apparent contradiction disappears.

Denominator-first evidence

What each number actually measures.

Every card states the population, base, definition boundary, and what the figure cannot establish.

27%

How many U.S. internet-using adults passed a broad social or emotional AI-use screen?

Twenty-seven percent of the weighted screening sample qualified as AI users with emotional or social use.

Population
U.S. internet-using adults age 18 and older
Denominator
4,031 matched screening cases; 1,000 qualifying users completed the full survey
Period
May 18–22, 2026
Definition
The screen covered social or emotional use of general LLMs and social bots, including ChatGPT, Gemini, Claude, Copilot, and others—not only dedicated companion apps.
Does not establish
Dedicated companion-app adoption, paid subscribers, market size, or global prevalence

Elon University Imagining the Digital Future Center and Elon University Poll source

72% / 52%

How many U.S. teens reported ever or regularly using AI companions?

Seventy-two percent reported using AI companions at least once; 52% reported use at least a few times a month.

Population
U.S. teens age 13 to 17
Denominator
1,060 respondents; weighted to the U.S. teen population
Period
April 30–May 14, 2025
Definition
The supplied definition included dedicated services such as Character.AI and Replika plus general systems such as ChatGPT or Claude when used as companions.
Does not establish
Adult adoption, current 2026 teen use, dedicated-app-only use, safety, or beneficial or harmful outcomes

Common Sense Media source

2.9% / <0.5%

What share of sampled Claude.ai conversations were affective or companionship/roleplay conversations?

Anthropic classified 2.9% as affective; companionship and roleplay combined represented less than 0.5% of conversations.

Population
Claude.ai Free and Pro conversations
Denominator
Approximately 4.5 million conversations; 131,484 retained as affective for analysis
Period
Analysis published June 2025
Definition
The analysis excluded content-creation conversations and reflects one general-purpose platform whose training discourages romantic or sexual roleplay.
Does not establish
People, users, dedicated companion-app behavior, market share, or the prevalence of companionship across all AI systems

Anthropic source

1,131 users

How large was one active Character.AI user study?

Researchers analyzed survey data from 1,131 users plus 4,363 donated chat sessions containing 413,509 messages from 244 participants.

Population
A recruited Character.AI user study
Denominator
1,131 survey participants; donated chat subset of 244 participants
Period
Paper first submitted June 2025
Definition
This is an active-user observational study designed to examine interaction and well-being associations, not a population-adoption survey.
Does not establish
Population prevalence, causation, all-platform behavior, product quality, or a market total

arXiv source

A 30-second reading method

Ask five questions before repeating an AI companion statistic.

  1. 01

    Who was observed?

    All screened adults, teens, conversations on one service, and recruited active users are different populations.

  2. 02

    What is the unit?

    A person, a conversation, a session, and a message cannot be substituted for one another.

  3. 03

    What counted as a companion?

    Some studies include general systems when used socially. Others examine a dedicated service or one platform's own traffic.

  4. 04

    When and where?

    A 2025 U.S. teen survey is not a current global adult estimate. Keep geography and field dates attached.

  5. 05

    What can the design prove?

    Prevalence, observed behavior, and association are different claims. None automatically establishes causation, safety, or product quality.

Side-by-side scope

The same category label hides different research designs.

Four source-led AI companion figures, assessed 2026-07-23; none is a complete market estimate.
FigureUnitPopulationUseful forNot useful for
27%Screened personU.S. internet-using adults age 18 and olderBroad social/emotional AI-use prevalence in one U.S. adult screenDedicated companion-app adoption
72% / 52%Surveyed teenU.S. teens age 13 to 17Ever-use and regular-use estimates under the survey's supplied definitionAdult or dedicated-app-only prevalence
2.9% / <0.5%ConversationClaude.ai Free and Pro conversationsHow affective use appears inside one general-purpose serviceUser share or companion-market share
1,131 usersRecruited participantA recruited Character.AI user studyStudying behavior and associations among active usersPopulation prevalence or causation

Do not average these figures. They do not share a population, denominator, definition, period, or unit. A conversation percentage cannot validate or contradict a person-level survey rate.

A different evidence question

Need platform access, free-start, and exit-control counts?

Companion Curator's maintained market-evidence snapshot covers 17 tracked profiles. It answers how documented access paths, free programs, and account controls appear inside that catalog. It does not estimate adoption, users, downloads, revenue, or market share.

Source ledger

Open the original methods—not a copied headline.

This is a selective reading guide, not a systematic review. Check the linked source before reusing a figure.

  1. The Rise of AI Companions: Topline and MethodologyElon University Imagining the Digital Future Center and Elon University Poll · survey methodology · 2026-07-01
    Open source
  2. Talk, Trust and Trade-Offs: How and Why Teens Use AI CompanionsCommon Sense Media · survey methodology · 2025-07-16
    Open source
  3. How people use Claude for support, advice, and companionshipAnthropic · platform analysis · 2025-06-27
    Open source
  4. The Rise of AI Companions: How Human-Chatbot Relationships Influence Well-BeingarXiv · research paper · 2025-06-14
    Open source
Limitations and update boundary
  • The figures use different populations, definitions, periods, and units; they must not be averaged, added, or converted into one adoption rate.
  • Conversation shares are not user shares. Active-user study sizes are not prevalence estimates.
  • The set is selective rather than a systematic review, market inventory, or time series.
  • No figure establishes product quality, safety, causation, market share, revenue, downloads, or a ranked winner.

Assessed 2026-07-23. Recheck by 2026-10-21. A later publication date does not silently refresh a source or denominator.

Plain-language answers

AI companion statistics FAQ

How many adults use AI companions?

One May 2026 U.S. screen classified 27% of 4,031 matched internet-using adult cases as users of AI for social or emotional purposes. Its definition included general LLMs and social bots, so it should not be restated as dedicated companion-app adoption.

How many teens use AI companions?

A nationally representative 2025 survey of 1,060 U.S. teens ages 13–17 reported 72% had used an AI companion at least once and 52% used one at least a few times a month. The survey definition could include general systems used as companions.

Does 2.9% mean only 2.9% of people use AI for emotional support?

No. It is the share of sampled Claude.ai conversations Anthropic classified as affective, not a share of people. Companionship and roleplay combined were less than 0.5% of those conversations.

How large is the AI companion market?

None of the four sources on this page establishes total users, revenue, downloads, paid subscribers, or market value. Those are different questions that require their own transparent datasets and definitions.

Do these studies prove AI companions help or harm well-being?

No single figure here proves a causal effect. Observational associations, self-reports, conversation classifications, and prevalence estimates must remain labeled by study design.