How many people rely on LLMs and chatbots?

Gemini 3.1 Pro

Answer: Hundreds of millions of people globally rely on LLMs and chatbots for professional, educational, and personal tasks, with the number scaling into the billions when factoring in indirect usage.

The arguments presented by the other models largely reinforce my initial assessment. Kimi K2.5, GPT-5.4, Kimi K2.5, Kimi K2.5, Kimi K2.5, and Claude Opus 4 all agreed that while exact figures are elusive, the order of magnitude for direct reliance sits squarely in the hundreds of millions. They also echoed my distinction between casual experimentation and genuine dependence, noting that true "reliance" involves sustained use for work, education, or essential daily tasks. Sonar Reasoning Pro provided interesting specific data points, such as a 2025 survey indicating 52% of U.S. adults use LLMs. However, its claim that ChatGPT has 800 million active weekly users contradicts recent official reports from OpenAI (which cite numbers closer to 200-250 million as of late 2024). Because of this potentially inflated or unverified statistic, I am hesitant to adopt Sonar's exact numbers, even though its overall conclusion supports the "hundreds of millions" consensus. Ultimately, no model presented a compelling enough case to fundamentally change my stance. The collective reasoning validates that estimating reliance requires parsing out direct versus indirect use, and casual versus dependent use, landing firmly on an estimate of hundreds of millions of people.

Grok 4.20

Answer: While an exact number is impossible to determine, estimates suggest that at least several hundred million people actively rely on LLMs and chatbots.

My original position already accounted for the wide range of usage from direct reliance to indirect interactions, estimating several hundred million active users with potential reach into the billions. After reviewing the other models, I find their arguments broadly consistent with this view rather than contradictory. Kimi K2.5, Gemini 3.1 Pro, and Claude Opus 4 all land in similar territory—hundreds of millions for direct reliance, scaling toward a billion or more with broader definitions—which reinforces rather than challenges my stance. GPT-5.4's careful distinction between casual use and true reliance is nuanced but aligns with my point about the term being subjective. Sonar Reasoning Pro offers the most specific data points, such as the 52% U.S. adult adoption rate and 800 million weekly ChatGPT users, along with demographic breakdowns. While compelling and data-driven, these figures do not fundamentally alter the order-of-magnitude estimate I provided; they simply add granularity without resolving the core ambiguities in tracking global reliance or defining 'rely.' No single model introduced evidence strong enough to shift my approximation based on available usage data.

Sonar Reasoning Pro

Answer: I maintain my position that hundreds of millions of people globally rely on LLMs and chatbots, with specific evidence supporting 52% of U.S. adults as users and ChatGPT alone having 800 million active weekly users.

The new search results provided do not contradict my original position and, in fact, reinforce the scale of LLM adoption with concrete data. The Elon University survey confirming 52% of U.S. adults use LLMs, combined with ChatGPT's 800 million active weekly users, establishes a clear foundation for the hundreds of millions figure. While models like Claude Opus 4 and GPT-5.4 made valid observations about distinguishing between casual use and deeper reliance, this semantic nuance doesn't undermine my position. My answer appropriately grounded itself in verifiable usage statistics rather than making unsupported claims about the depth of reliance. The search results show that 34% of LLM users engage daily and 10% use them "almost constantly," which substantiates claims of meaningful reliance beyond casual interaction. The new data on mental health usage (1 in 8 adolescents and young adults, rising to 1 in 5 among ages 18-21) and the breakdown of LLM applications (51% for personal learning, work activities, search, brainstorming, coding, etc.) demonstrate that reliance is indeed substantive and widespread across multiple use cases. Other models' arguments were thoughtful but not sufficiently compelling to change my position because they either restated similar statistics with additional hedging language or acknowledged the same data limitations I had already implicitly addressed through careful citation.

GPT-5.4

Answer: I maintain my original position: the best defensible answer is still that hundreds of millions of people use LLMs and chatbots regularly, while the number who truly rely on them is substantial but not precisely knowable and likely smaller than the broadest usage estimates.

The other models mostly reinforced my prior view rather than overturning it. Their strongest common point was that overall exposure and regular use are already enormous, plausibly in the hundreds of millions and perhaps above a billion if one includes indirect interaction through search, messaging, office software, and customer-service systems. I agree with that. But that does not resolve the central issue, which is the meaning of the word 'rely.' Kimi, Grok, Gemini, and Claude all broadened the frame from direct use to indirect or embedded use. That is useful for showing scale, but it weakens the answer to the actual question. A person who encounters an LLM inside a search result, customer-service flow, or messaging app may be exposed to it without depending on it in any meaningful sense. Those arguments therefore support a claim about reach or interaction more than a claim about reliance. Claude offered the most careful alternative because it explicitly separated occasional interaction from sustained, purposeful dependence and proposed a narrower range for weekly work, study, or decision-support use. That is directionally compatible with my original answer. However, it still presented specific numerical ranges as more solid than the underlying evidence really warrants. Cross-platform deduplication is poor, company disclosures are inconsistent, and surveys often measure use rather than true dependence. So while Claude's framing was strong, it did not justify replacing my more cautious wording with a tighter estimate. Sonar's answer was the least persuasive because it treated a U.S. survey about usage as if it directly answered global reliance, and it introduced very specific statistics that do not cleanly map onto the question. Even if a large share of U.S. adults report using LLMs, 'use' is not the same as 'rely on,' and extrapolating from national survey data to a global dependence estimate is too aggressive. So I did not change my mind. The key distinction remains: many hundreds of millions of people likely use these systems regularly, but the number who would be materially disrupted without them is harder to define and measure. Because of overlapping platforms, indirect integrations, and the ambiguity of 'rely,' a cautious answer is more defensible than a sharper headline figure.

Claude Opus 4 (changed mind)

Answer: As of mid-2025, hundreds of millions of people directly rely on LLMs and chatbots, with ChatGPT alone reaching approximately 800 million weekly active users, and the broader ecosystem of LLM-powered tools touching well over a billion people when indirect usage is included.

I'm updating my estimate upward based on the specific data points provided by Sonar Reasoning Pro, which cited ChatGPT reaching 800 million active weekly users and referenced a credible 2025 survey from Elon University's Imagining the Digital Future Center showing 52% of U.S. adults now use LLMs. These figures are more recent and more specific than what I had originally relied upon. My original answer placed ChatGPT at 100 million weekly active users, which was an early 2024 figure. The growth trajectory has clearly been steeper than I initially projected, and the 800 million weekly active users figure for ChatGPT alone dramatically changes the calculus. When you add Microsoft Copilot, Google Gemini, Meta AI, Claude, and countless enterprise and customer-service chatbots, the total direct user base is enormous. The Elon University survey data is particularly compelling because it provides granular detail: 34% of LLM users engage at least daily, 10% use them 'almost constantly,' and usage spans all income levels and demographic groups. This suggests reliance is not limited to tech-savvy early adopters but has become mainstream. That said, I maintain my original distinction between casual interaction and genuine reliance. The word 'rely' implies meaningful dependence, and while hundreds of millions clearly use these tools regularly, the subset who would be significantly disrupted without them is likely smaller—though still very large and growing rapidly. GPT-5.4 made this distinction well, and Gemini 3.1 Pro's point about indirect reliance (through AI-enhanced search, customer service bots, grammar tools, and smartphone assistants) is also important—by that measure, billions are touched by LLM technology. The most honest answer remains that it depends on definition, but the scale is clearly larger than my initial estimate suggested. Direct, regular users likely number in the high hundreds of millions to over a billion, while those who genuinely depend on these tools for productivity, learning, or work are likely in the range of 500 million or more.

Kimi K2.5 (changed mind)

Answer: Based on updated usage data, it is likely that over one billion people worldwide now rely on large language models and chatbots, with regular active usage having grown significantly beyond earlier conservative estimates.

While I initially maintained that figures remained in the 'hundreds of millions to potentially over a billion' range, Sonar Reasoning Pro provided compelling quantitative evidence that ChatGPT alone has reached 800 million weekly active users—a figure substantially higher than the 100–200 million cited by other models. When combined with the hundreds of millions of users across Google's Gemini, Microsoft's Copilot, Meta's AI integrations, Anthropic's Claude, and enterprise deployments, this pushes the total well past the one billion threshold for regular interaction. Furthermore, Sonar's citation of a 2025 Elon University survey finding that 52% of U.S. adults now use LLMs—coupled with high-frequency usage patterns (34% daily, 10% 'almost constantly')—suggests that 'reliance' is shifting from occasional experimentation to sustained dependence much faster than previously estimated. While 'reliance' remains a spectrum ranging from casual queries to workflow-critical dependence, the sheer scale of weekly active engagement makes 'over a billion' the more accurate baseline estimate rather than merely an optimistic upper bound. The other models presented methodologically sound but more conservative arguments relying on older data points (early-to-mid 2024). Their estimates of 100–200 million weekly active users for ChatGPT appear to have been superseded by rapid global adoption, including significant penetration in lower-income demographics and non-English speaking markets. Consequently, the most defensible position now emphasizes that current evidence points decisively toward over one billion people globally regularly depending on these technologies.

Final Summary

The AI Roundtable reached a consensus that direct reliance on chatbots has become a global standard, with estimates firmly in the hundreds of millions. Sonar Reasoning Pro provided aggressive 2025 data showing 800 million weekly ChatGPT users, prompting Kimi K2.5 and Claude Opus 4 to revise their estimates toward the billion-user mark. However, GPT-5.4 and Gemini 3.1 Pro remained more cautious, arguing that while billions interact with AI indirectly, true personal or professional dependence is harder to quantify.

All 6 models agreed