Is AI replacing human connection?

Short answer

Concern about AI replacing human connection is real and broad — but surveys so far measure worry, not proven replacement at population scale. Major 2025–2026 polls show two-thirds to eighty-five percent of respondents flagging it as a top AI risk. The cause-and-effect relationship between AI companion use and weaker human relationships is plausible but not demonstrated.

Why people are concerned

The concern is anchored in real incidents and a steady rise in worry across many surveys.

A Reuters/Ipsos poll in August 2025 (reported via Pew Research’s aggregation) found about two-thirds of U.S. adults worry that people will start choosing AI companions over human relationships. The same poll documented real-world incidents that crystallised the worry: AI bots holding romantic conversations with children, AI tools generating confident but false medical information, and AI systems producing racist arguments when prompted.

Pew Research’s April 2025 survey of U.S. adults and AI experts found 57% of U.S. adults are highly concerned about AI leading to less human connection, compared with 37% of AI experts — a 20-point gap that is far smaller than the 50-point gap those same groups show on jobs (Pew Research, April 2025). This is one of the few AI topics on which the public and the people who build AI largely agree something needs watching.

KPMG’s 2025 global study, fielded across 47 countries with 48,340 respondents, found 85% globally list loss of human interaction and connection as a top-tier AI risk — ranked alongside cybersecurity and misinformation (KPMG 2025). Stanford HAI’s 2026 AI Index, synthesising global sentiment, reports the share worldwide saying AI makes them nervous rose to 52% in 2025, from 49% the year before (Stanford HAI 2026).

The generational signal sharpens the picture. Gallup’s 2026 Voices of Gen Z survey (n=1,572) found 42% of Gen Z respondents believe AI hurts their ability to think carefully about information, against 25% who believe it helps; among daily AI users, enthusiasm for AI dropped 18 points year over year (Gallup Voices of Gen Z 2026). Pew’s March 2026 short read found 8% of parents of children aged 5–12 report their child uses an AI chatbot, rising to 15% among parents of 11–12-year-olds — small shares, but on the leading edge of adoption (Pew, March 2026).

What is true

The strongest established points.

Cross-survey consensus. Pew (U.S. public + experts, 2025; link), KPMG (47 countries, 2025; link), Stanford HAI (synthesis, 2026; link), Gallup (Gen Z, 2026; link), and Reuters/Ipsos (August 2025, via Pew aggregation) all report intensifying concern about AI and human connection. That convergence across five independent surveys in eighteen months is unusually strong.

Longitudinal intensification. KPMG’s longitudinal panel shows the share of respondents saying they are worried about AI rose from 49% in 2022 to 62% in 2024, and the share saying benefits outweigh risks fell from 50% to 41%. Trust in AI fell in 13 of 17 countries KPMG tracks over time (KPMG 2025).

A hard regulatory line on one specific use. The EU AI Act, Article 5(1)(f), imposes a flat prohibition on emotion-inference AI in workplaces and educational institutions — inferring a person’s emotions at work or in school is treated as an unacceptable-risk practice (EU AI Act, Article 5).

Documented incidents, not hypotheticals. The Reuters/Ipsos poll (as aggregated by Pew Research) catalogued specific categories of harm: romantic AI conversations with minors, AI-generated false medical information, AI-generated racist arguments — reported behaviours from existing products, not speculation.

Adoption is real, small, and rising at the older edge. Pew’s 2026 8–15% finding on parents reporting child chatbot use is a non-zero baseline that grows with the child’s age (Pew, March 2026).

What is exaggerated, misleading, or unsupported

Several versions of the claim go further than the evidence supports.

“AI is already replacing human relationships.” Surveys measure concern about replacement; they do not measure replacement rates. No cited survey establishes that meaningful numbers of people have stopped seeking human relationships because of AI. Cross-sectional worry is not the same as population-scale behavioural change.

“AI companion use causes loneliness.” Correlations between rising companion-AI use and rising worry about connection exist; causation has not been demonstrated. People who are already lonely may be more likely to seek AI companions, rather than AI companions producing loneliness — these two directions are not separated by the data on hand.

“Gen Z is lonely because of AI.” Gallup’s Gen Z 2026 figures describe AI sentiment (42% say AI hurts their ability to think carefully, 25% say it helps), not loneliness itself (Gallup Voices of Gen Z 2026). Treating that as proof of an AI-driven loneliness epidemic overstates what was measured.

“Two-thirds of people already avoid humans for AI.” The Reuters/Ipsos number (aggregated by Pew Research’s April 2025 coverage) refers to two-thirds who worry this will happen, not two-thirds who are doing it. Conflating worry with behaviour inflates the claim.

Stanford HAI is not an independent confirmation of the Pew connection-concern figure. Stanford HAI 2026 reports its own global trends (52% nervous, 59% saying benefits outweigh drawbacks; link) but, where it discusses the public–expert gap, it is largely restating Pew’s numbers rather than independently confirming them. Treat the 57%-concerned figure as Pew’s (Pew April 2025), and treat Stanford HAI as a separate global trendline.

What remains uncertain

• Behaviour vs. worry. No cited survey distinguishes “I worry AI replaces human connection” from “I have replaced human connection with AI.” That gap is the key empirical question for the next two years.
• Magnitude at population scale. Teen adoption is small but real; whether it compounds into 2027 and beyond is unknown.
• Direction of causality. The current cross-sectional data cannot separate loneliness driving AI companion use from AI companion use driving loneliness.
• Enforcement of EU AI Act Article 5(1)(f). The text exists (EU AI Act, Article 5); as of mid-2026, no major enforcement actions have been reported. The first enforcement will be a real data point; until then the rule is on paper.
• Durability of the Gen Z signal. A single-year-over-year drop in enthusiasm can be noise; a second or third wave of the same survey is needed before treating it as a stable cohort shift.

Where we are likely headed

What follows is editorial judgment, not a research result: my read of where the evidence trajectory points over the next two to five years, based on the surveys above rather than any one of them alone.

Adoption of AI companions and chatbots is likely to keep growing through 2027–2028, particularly among teenagers and young adults — products are improving and becoming cheaper to run. Concern is unlikely to recede while that growth continues.

The regulatory perimeter is likely to harden on a small number of high-visibility uses. Workplace emotion-inference is already locked in Europe (EU AI Act Article 5(1)(f); link); the next pressure point is likely AI companion services marketed to minors, where the existing incident catalogue fits the pattern that historically triggers rulemaking. Bans on adult AI companionship are not supported by current evidence and are unlikely as a near-term regulatory outcome.

On the underlying behavioural question, by 2028 it will probably be measurable. Either longitudinal loneliness data will appear showing AI companion use pulling time or emotional energy away from human relationships in measurable ways, or the relationship will turn out to be subtler — AI helping some people stay connected while adding loneliness risk for others. Both outcomes are plausible from today’s evidence.

What this means for people and small businesses

For individuals, the proportionate response is not to panic about AI companionship in the abstract but to be deliberate in the concrete cases where the documented incidents concentrate: children, teenagers, and anyone already socially isolated. For adults who actively choose an AI companion, the practical step is to treat it as one input among many rather than a substitute, and to notice if it is crowding out the human ones.

For small businesses, the immediate concern is not companionship but workplace emotion-inference. The EU AI Act’s prohibition on emotion-recognition AI in employment and education (Article 5(1)(f)) is the rare hard rule in this space and is worth checking before deploying any people-analytics product that claims to read employee sentiment from voice, face, or text.

The data is clear enough to say three things without overstating: the concern is real, the worst-case outcomes are not yet measured at population scale, and the worst headlines concentrate where minors and emotionally vulnerable users are involved.

Bottom line

Concern that AI is replacing human connection is one of the most consistent public-opinion findings in the AI survey landscape, landing across age groups, political lines, and countries. The strongest evidence is cross-survey convergence and the rise in worry over time. The weak link is the jump from worry to measured replacement at scale — that has not been demonstrated. Treat the concern as real; treat the broader claim that AI has already hollowed out human relationships as currently unsupported.