A quarter of Americans call an AI a close friend or partner

Where things stand (October 2026)

No Pew, Gallup or General Social Survey publication reports anything close to 25% of US adults saying they consider an AI a close friend or romantic partner. The best measurements are far below that:

• Pew, "Americans and AI 2026" (published June 17, 2026; fieldwork Feb 17–23, 2026; 5,119 adults on its probability-based panel). 49% of adults use chatbots, 10% use them for emotional support, and only 4% use them for companionship. Pew has no item that asks about close friends or romantic partners. https://www.pewresearch.org/internet/2026/06/17/americans-and-ai-2026-chatbots-smart-devices-and-views-on-impact/
• YouGov opt-in poll (published July 1, 2026; 1,126 adults). 10% say they have an "ongoing personal friendship" with a chatbot (18% of under-30s). That counts as "friend" without closeness, and opt-in panels tend to inflate such items. https://yougov.com/en-us/articles/55093
• Elon University/YouGov (published Sept 2, 2026). 27% of internet-using adults have personal or emotional conversations with AI, and 31% of that group call their chatbot a "friend". That is about 8% of adults, and again not "close friend". https://www.elon.edu/u/news/2026/09/02/imagining-the-digital-future-center-reports-on-the-rise-of-ai-companions/
• Yahoo/YouGov (Nov 2025). 6% say they could form, or already have, a "deep emotional bond" with an AI. 3% say the same about a romantic relationship, and 1% actually use AI that way.
• Gallup/Walton Family Foundation, Gen Z only (fieldwork Oct 2025). About 1 in 4 people aged 18–28 used a chatbot "as a friend" and 10% "as a girlfriend or boyfriend". This is a subgroup, so it does not count, but it shows Gallup does ask these questions.
• Other context. Match (June 2026) found near-universal disapproval of dating an AI. Menlo Ventures (Sept 2026) found adult AI adoption nearly flat (61% to 64%) and about 6% of Americans talking with AI companions. The AI 2027 scenario's milestone of "10% consider an AI a close friend" by July 2027 is still listed as unmet by a tracker (Sept 2026).

My estimate is that, in a probability-based survey with explicit wording, about 2–5% of adults would currently say an AI is a close friend or romantic partner.

Reference class and model

The closest reference class is new social behaviours tied to technology: online dating, for example, took more than a decade to reach about 30% of adults. Getting from about 3% to 25% within five years needs more than 50% growth a year, every year, with no saturation. That is despite strong stigma, resistance among older adults, and regulatory pushback against AI companions.

Several things must all happen:

1. The true share passes 25%. I put this at about 5% by mid-2031, and nearly all of that comes from a fast-AI world with highly engaging voice, memory and embodied companions.
2. Pew, Gallup or the GSS asks a qualifying question: explicit "close friend" or romantic wording, all adults, fieldwork after Sept 16, 2026.
3. They publish it before the end of 2031.

I put the chance of steps 2 and 3, given step 1, at about 50%. Pew is the most likely source. Gallup is plausible. The GSS is unlikely.

Calculation

I built quarterly hazards in Python: 0.1% for Q4 2026, then 0.04% per quarter in 2027, 0.08% in 2028, 0.12% in 2029, 0.18% in 2030 and 0.23% in 2031. The cumulative product gives 0.26% at end-2027, 1.05% at end-2029 and about 2.7% at end-2031. The hazard rises over time because any YES realistically needs years of growth and a late survey.

Strongest case for YES

AI is advancing very quickly. Among Gen Z, about 25% already use chatbots "as a friend" and 10% as a girlfriend or boyfriend. Agree/disagree wording ("I consider an AI a close friend") could push estimates up. A fast-takeoff world could normalise AI companionship within a few years. These points are why I do not go below about 2–3%. But the 25% bar applies to all adults and needs explicit closeness or romance, which is far beyond any current probability-based number. So I don't think a higher figure is justified.

Question. By the last day of each quarter, has Pew Research Center, Gallup, or the General Social Survey reported that at least 25% of US adults say they consider an AI system a close friend or a romantic partner?

Definitions.

• US adults: the nationwide population aged 18 or older, as represented by the survey's published weighted estimate. Standard household or noninstitutionalized-adult sampling frames count. Estimates restricted to chatbot users, subscribers, workers, particular age groups, or other subgroups do not count.
• AI system: an artificial-intelligence chatbot, assistant, virtual character, or embodied system, from any developer. The survey must identify the relationship as being with an AI, not a human contacted through an AI-enabled service.
• Consider a close friend or a romantic partner: respondents affirm that they currently regard at least one AI as a close friend, best friend, romantic partner, boyfriend, girlfriend, or spouse, or affirm equivalent explicit relationship wording. Agreement with a statement such as “I consider an AI chatbot a close friend” counts. Merely calling it a “friend” or “companion,” without closeness or a romantic relationship being specified, does not.
• At least 25%: the published weighted point estimate is at least 25%. Use the most precise estimate published with the finding; if only a whole-number percentage is available, a reported 25% counts. Confidence intervals do not determine resolution. “Close friend or romantic partner” is the union of the two categories, counting each respondent once. Either category alone reaching 25% is sufficient; adding overlapping percentages without an overlap-adjusted union is not.

Resolves YES if at least one named organization publishes a qualifying survey finding meeting these definitions. A report, official article, topline, or officially published weighted table counts. The questionnaire or accompanying documentation must establish the relationship wording and population denominator. A split-sample survey item counts if its published estimate represents US adults generally. The survey's fieldwork must end after the opening date, under the Observation period convention.

Resolves NO if no qualifying finding has been published by the quarter-end. In particular, exclude:

• Seeking emotional support, advice, companionship, or romantic or sexual conversation, unless respondents explicitly affirm the qualifying relationship.
• Willingness to form such a relationship, beliefs about whether such relationships are possible, or reports only of past relationships.
• Percentages for teens or other subgroups, anecdotes, convenience website polls, and estimates from other polling organizations merely quoted by a named organization.
• Independent calculations from downloadable microdata, including GSS microdata, without an official publication of the qualifying estimate; combinations of separate surveys or waves; and sums whose overlap is unknown.

Resolving source. Pew Research Center publications at pewresearch.org; Gallup publications at gallup.com or news.gallup.com; or NORC at the University of Chicago's official General Social Survey publications and tables at gss.norc.org, gssdataexplorer.norc.org, or norc.org. Jointly authored survey reports count; unrelated NORC surveys do not qualify as GSS. These named-source restrictions replace the broader source options in Public confirmation (A2). Apply Resolution date (A3), the Resolution window, and the Observation period conventions. A qualifying revised estimate counts on its publication date; revisions concerning only observation periods ending on or before the opening date do not count. Once YES, later revisions or lower estimates do not reverse it.