One in three people aged 15 to 22 across 16 countries told Gallup they use AI for companionship, according to the Friendship in an Online-Offline World report summarized by Wired and the Gallup wellbeing release with the Boston Children's Digital Wellness Lab. Among daily AI users, that share rises to 50%.
The finding is not proof that chatbots replace friends. Authors warn that loneliness links are correlational. Still, for anyone shipping consumer AI, the numbers show companionship is a primary use case, not an edge case.
Headline numbers from Gallup
| Metric | Value | Notes |
|---|---|---|
| Sample | 15,925 youth 15–22 | Fielded May 28–Aug. 21, 2026 |
| Countries | 16 | Mix of web and face-to-face samples |
| AI companionship (median) | 33% | Across countries |
| Daily AI users using AI as companion | 50% | Subpopulation |
| Youth with close in-person friend | 93% median | Per Gallup poll page |
| Close online-only friend | 37% median | Same release |
Wired's November 18 report by Jason Parham popularized the 33% figure for US readers. The Boston Children's analytics brief adds wellbeing framing: 45% rate in-person friendships highest for strength and support versus 9% for online-only friendships.
What questions the study actually asked
Gallup examined three friendship modes:
- In-person close friends.
- IRL-started friendships sustained online.
- Online-only friends never met in person.
It layered AI companionship as a separate behavior: 82% used AI to look up information, 68% for advice, and 33% as a companion, per 404 Media's report on the Gallup study and Wired.
Researchers emphasized coexistence: "In-person and digitally mediated friendships coexist within the same social life," a line Wired quotes from the report. AI has not, in their wording, replaced IRL bonds at population scale.
Loneliness: association, not verdict
The sensitive claim: youth who rely on online-only friends or AI companionship may have higher odds of loneliness than peers without those patterns.
Gallup principal researcher Andrew Dugan told Wired: "It wouldn't be surprising if AI companionship becomes more common over the next five years," but whether young people turn to AI because they feel lonely, or AI reinforces loneliness, remains an important question for future research."
That sentence should gate product policy and parental guidance:
- Do not market companions as therapy replacements.
- Do label synthetic peers clearly — relate to Replika and Character.AI coverage.
- Do study retention separately from wellbeing outcomes.
Country and platform differences
Wired notes Brazil and Japan show higher daily AI use than the US, UK, and Canada, where nearly half used AI no more than a few times per month.
For staying in touch with online-only friends, Instagram and Discord ranked above TikTok in the US sample — a detail product teams guessing "short video = Gen Z friendship" should note.
Implications for builders and parents
| Stakeholder | Practical takeaway |
|---|---|
| App developers | Companion features need minor safety, crisis routing, and session limits — see also Grok proactive check-ins debates. |
| Educators | South Korea exam cheating reforms show schools treating AI as cheating vector; Gallup shows the same cohort uses AI for social bonding. Policies must separate uses. |
| Policy | Americans negative AI sentiment coexists with youth adoption — regulate harms, not existence. |
Youth digital wellbeing and friendship online versus in person in the Gallup 2026 companionship study
Image: Library reference — Gallup ties in-person ties to higher wellbeing ratings than online-only friendships.
How this differs from viral companion app stories
Prior explainx.ai posts tracked Replika, Character.AI, and Meta Muse incidents. Gallup is cross-country, vendor-neutral, and survey-based. It supports demand forecasting for teen-facing assistants without naming a single SKU.
Study limitations to cite honestly
- Non-probability web samples in 14 countries — not every country uses probability sampling.
- Self-report on loneliness and friendship quality.
- AI companion definition may include general chatbots, not only romance bots.
Methodology: how Gallup fielded the poll
According to the Gallup poll page, 14 countries used non-probability web samples while Kenya and South Africa used face-to-face probability samples designed to be nationally representative within those countries. Each country targeted about 1,000 completes except the Netherlands (820).
That design trades speed and cross-country comparability for strict census-style precision in wealthier markets. When you cite 33% companionship, you are citing a cross-national median pattern, not a US-only teens statistic — though US respondents are inside the sample.
The study is phase one of a broader Belonging by Design program referenced in secondary coverage. Expect follow-up waves on whether AI companionship rises as models integrate into messaging apps.
Daily AI use versus occasional experimentation
Wired highlights a split: in the US, UK, and Canada, nearly half of youth used AI no more than a few times in the past month, including never users. Meanwhile Brazil approaches 40% daily use and Japan and Italy sit near 35%.
Product teams interpreting "Gen Z loves AI friends" must segment DAU companions from monthly homework helpers. The 50% companionship rate among daily AI users is the cohort most likely to anchor habits — and most likely to trigger parental concern.
Platforms youth use to maintain online-only ties
Researchers asked where online-only friendships stay warm. In the US, Wired reports Instagram and Discord outranked TikTok — surprising if your marketing plan assumes short video equals Gen Z social graph.
For builders integrating AI into social products:
| Platform type | Implication |
|---|---|
| DM-centric (Discord, Instagram) | Companion bots may need thread-native presence, not standalone apps |
| Feed-centric (short video) | Less cited for maintaining close ties in this sample |
| General assistants | Used for lookup and advice more often than companionship, but overlap exists |
Designing safer companion defaults
Boston Children's Digital Wellness Lab partnership signals clinical scrutiny. Even without new laws, expect hospital systems and schools to cite this poll when arguing for:
- Age gates and parental dashboards
- Crisis hotline routing when chats mention self-harm
- Transparent memory policies so teens know what persists
Connect to MAI Cyber and youth safety models only where security tooling applies; this story is wellbeing, not CVEs.
How marketers should talk about the stat
Do:
- Cite 33% median and 50% among daily users together — otherwise you cherry-pick.
- Say loneliness association is not causal in the same paragraph as the headline number.
Do not:
- Claim "AI causes loneliness epidemics" — the study explicitly refuses that jump.
- Present companionship use as only Replika-like products; survey wording is broader.
Pinterest Impact Fund context
The Boston Children's analytics page notes Pinterest Impact Fund support. That funding line matters for disclosure hygiene when journalists rewrite the poll. Pinterest's corporate interest is digital wellbeing branding, not a particular chatbot vendor.
FAQ alignment with search queries
Parents search "is AI safe for teens" alongside companionship headlines. This study's honest answer: many teens already use AI socially, most still have in-person friends, and we do not know long-term mental health effects yet. Point families to vendor policies and OpenAI education programs for school context, not this poll alone.
Comparing companionship to advice and lookup use
The 82% information lookup and 68% advice figures show most youth still treat AI as a tool first. Companionship at 33% is substantial but not universal. Product roadmaps should assume multi-modal usage per user: the same teen may ask for homework help, draft a message to a friend, and vent to a bot in one week.
Enterprise builders selling into schools should separate tutoring copilots from social chat in policy docs — regulators will read Gallup medians as proof students already blur the lines.
What Gallup said about AI rating quality
Secondary coverage notes youth often rate AI companionship as medium quality rather than high or low, with Australia showing higher high-quality ratings than Poland in one breakout. Cross-country variance warns against US-centric apps globalizing companion UX without local mental health resources.
Research agenda the authors want next
Andrew Dugan's five-year forecast — that companionship may become more common — is not a product mandate. It is a research funding argument. Expect longitudinal studies tying companionship to sleep, grades, and offline club participation. Until those land, builders should log opt-in wellbeing surveys sparingly and transparently.
Tie-in to workplace and education policy
Teachers already report AI freeing time in a separate Gallup poll explainx.ai covered. Students using AI as friends while teachers use AI as grading assistants describes one school day, not two unrelated worlds. Districts need one acceptable-use policy covering companionship bots and essay generators together. Until those policies exist, treat this poll as baseline prevalence data, not a license to ship unlabeled friend bots to minors.
Related reading
- AI relationships and companion apps
- Gallup teachers and AI time savings
- Americans negative AI sentiment polls
- South Korea exam AI cheating reforms
- Grok bot proactive check-ins
- OpenAI education guidance programs
Companion percentages are from Gallup's November 2026 public release unless noted as Wired paraphrase.
