If your product strategy framework assumes that the use case for AI as a companion or ongoing emotional support service is a niche application that won't reach material adoption within a strategic planning horizon, the 2025 Longitudinal Expert AI Panel (LEAP) and Ipsos-Google forecasts diverge sharply on both the pace and ultimate scale of AI companionship adoption. AI experts forecast that 10% of US adults will use AI for companionship at least daily by 2027, rising to 15% by 2030 and 30% by 2040. The top quartile of expert forecasts predicts more than 40% daily by 2040; the top decile predicts over 60%. The general public's median forecast is 20% by 2040. The expert-public divergence on AI companionship is one of the most contested forecast areas in the global AI survey landscape.

The forecast divergence has three structural dimensions worth examining.

The first dimension: pace. Experts expect 10% daily AI companionship by 2027, within two years of the forecast date. The public's median forecast doesn't reach 10% until later in the 2030s. The pace gap is 5-10 years for the same penetration level. The implication: experts believe AI companionship is at an inflection point and will grow rapidly; the public expects gradual expansion over a longer horizon.

The second dimension: ultimate scale. The expert median for 2040 is 30%. The top quartile is 40%+. The top decile is 60%+. The public median for 2040 is 20%. The expert distribution skews substantially higher than the public distribution. The implication: experts see AI companionship as potentially reaching mainstream adoption levels (30-40% of adults) within 15 years; the public sees it as remaining a minority application even in the long run.

The third dimension: variance. The expert distribution is wide: the gap between the median (30% in 2040) and the top decile (60%+ in 2040) is 30 percentage points. The wide distribution reflects genuine uncertainty among experts about how the technology will evolve, what social-acceptance dynamics will be, and which use cases will dominate. The public distribution is narrower, clustered around lower adoption levels. The expert-public divergence is partly that experts are more uncertain (wider distribution) and partly that they are more bullish on average.

The Ipsos-Google global excitement data adds a cross-country dimension. 52% of worldwide respondents reported some excitement about using AI for companionship in 2025, substantially higher than the US-only figure of 42% from Pew. The geographic distribution:

  • Highest "extremely excited": Nigeria 29%, India 27%, UAE 22%, South Africa 20%
  • Highest "not excited at all": US 36%, Canada 34%, Japan 32%
  • High "don't know": Japan 18% (the highest), reflecting cultural ambivalence

Excitement about using AI for companionship by country, 2025

The cross-country pattern matches the broader sentiment geography: Southeast Asian, African, and Middle Eastern populations show higher AI companionship excitement; North American, European, and Japanese populations show lower excitement. The 52% global excitement and the 42% US excitement both substantially exceed the 20% by-2040 public adoption forecast, meaning many people are open to AI companionship in principle but don't expect to personally use it daily.

Adoption forecasts here vary by an order of magnitude across expert versus public sources and across countries.

Three interpretations are defensible:

Interpretation 1: experts are right; AI companionship is at inflection. Replika, Character.ai, and XiaoICE already have millions of users (Zhang and Lu 2023; Zhang et al. 2025). Memory of past interactions, emotional recognition, and adaptive response capability are now standard in leading systems. The technology has crossed the capability threshold for sustained use. The trajectory will follow the typical S-curve for emergent applications, with rapid growth from a current low base to mainstream adoption. Under this interpretation, organisations should plan for AI companionship to be a significant consumer product category by 2030 and a mainstream one by 2040.

Interpretation 2: the public is right; AI companionship will remain niche. Mental health professionals, therapists, social science researchers, and concerned observers have raised significant concerns about AI companionship's effects on emotional dependence, psychological distress, and existing relationship quality. The 36% "not excited at all" share in the US, the 34% in Canada, and similar shares in Europe suggest there is substantial public resistance to the application. Under this interpretation, AI companionship will remain a smaller-scale application, used by some and resisted by many, without reaching mainstream adoption. The current Replika and Character.ai user bases may be the upper bound rather than the lower bound.

Interpretation 3: both forecasts are partially right; adoption will be geographically uneven. The Ipsos-Google country distribution suggests AI companionship may reach high adoption levels in some regions (Nigeria, India, UAE, parts of Southeast Asia) while remaining niche in others (US, Canada, Japan, parts of Europe). Under this interpretation, global aggregate adoption may match the expert median while specific country adoption diverges significantly. Strategic planning should engage with regional variation rather than a single global forecast.

The uncertainty runs deep. AI companionship adoption is one of the highest-uncertainty product strategy questions in the 2026 landscape. The technical capability is real and improving rapidly. The user-base adoption depends on social, cultural, and psychological dynamics that the data does not fully predict. Organisations operating in adjacent spaces (consumer AI, social media, dating, mental health, education) need to engage with multiple scenarios rather than commit to a single forecast.

Three implications follow for organisations whose product strategy intersects with AI companionship or adjacent emotional-support applications.

The first implication: the technical infrastructure for AI companionship is now widely available, and the regulatory environment is shifting. California's SB 243 (effective January 2026) requires companion chatbot operators to disclose AI nature and implement safety protocols related to suicidal ideation; Utah's HB 452 (effective 2025) regulates mental health chatbots. Organisations deploying companion-class AI face compliance requirements that didn't exist 24 months ago. Building compliance infrastructure now is cheaper than retrofitting.

The second implication: the public-expert divergence on adoption forecasts creates a market sizing challenge. Conservative product strategy operates from the lower (public) forecast; aggressive product strategy operates from the higher (expert) forecast. The 5-10x range between the two creates strategic risk in either direction. Investments calibrated to the high forecast risk over-building if the lower forecast materialises; investments calibrated to the low forecast risk being out-positioned if the higher forecast materialises. The reasonable planning posture: scenario-plan for both, and monitor leading indicators (Replika, Character.ai, OpenAI's GPT-based companion features, Anthropic's Claude usage patterns) for trajectory signal.

The third implication: the geographic distribution suggests product strategy should be regionally adaptive rather than globally uniform. Markets with high companionship excitement (Nigeria, India, UAE) represent earlier-stage opportunities; markets with high resistance (US, Canada, Japan, parts of Europe) require more careful product design and communication. Multinational AI companies face a choice: build a single global companion product that meets the most restrictive regulatory and cultural requirements, or build regionally-adapted products that match local sentiment and regulation.

The bottom line stays contested. AI companionship is the product category where 2026 strategic decisions carry the highest forecast uncertainty. The recommended posture is engaged uncertainty: actively monitor adoption data, maintain product strategy optionality, build compliance infrastructure, and prepare for either the high-adoption or the niche-adoption scenario to materialise over the 24-48 month planning horizon.