Do People Actually Just Talk to AI Chatbots for Fun?

The short answer is yes — and the numbers are stranger than you’d expect. A website where humans pretend to be AI chatbots racked up 25 million unique visitors in roughly one month. That’s not a curiosity. That’s a signal.

Key Takeaways

• A June 2025 Pew Research study found 34% of U.S. adults have used ChatGPT — roughly double the 2023 figure — confirming AI chatbots have crossed into mainstream daily use.
• The site Your AI Slop Bores Me, built by a 17-year-old, hit 25 million unique visitors and 280 million total hits in under a month, offering measurable, data-backed proof that people genuinely engage with AI chatbots for fun.
• Replika users have documented forming romantic bonds with AI companions, while NHS warnings about mental health chatbot risks confirm the social dimension is no longer fringe behavior.
• The split between utility chatbot use and social/entertainment chatbot use is widening — and product teams ignoring the latter are missing a significant chunk of actual user behavior.

How We Got Here

Three years ago, talking to an AI chatbot for fun felt slightly embarrassing — something you’d do in private and not mention at dinner. That stigma has largely evaporated.

A June 2025 Pew Research study found 34% of U.S. adults have used ChatGPT, approximately double the 2023 figure. That adoption curve tracks almost exactly with the shift in why people use these tools. Early adopters used chatbots for tasks: drafts, code, summaries. Now a measurable portion use them for conversation itself.

Several forces pushed this shift simultaneously. The tools got dramatically better at sounding human. Social isolation post-2020 left gaps that some people filled with digital interaction. Platforms like Character.AI and Replika deliberately built for social engagement rather than task completion — and found enormous audiences doing exactly that.

The cultural moment crystallized somewhere around late 2025. Chatbot interactions started appearing regularly on social media, not as screenshots of impressive outputs but as conversations — jokes, emotional exchanges, weird tangents. The chatbot had become a social object, not just a productivity tool.

That shift created the conditions for something genuinely odd: humans impersonating AI chatbots, for fun, at scale.

The “Human Pretending to Be AI” Phenomenon

Your AI Slop Bores Me (youraislopbores.me) is the clearest data point on whether people engage with AI chatbots for fun — because it inverts the premise entirely. Mihir Maroju, a 17-year-old from Puducherry, India, built a site that looks like a standard AI chatbot interface. Send a query. Get a response. Except the responses come from anonymous humans given 75 seconds to reply.

The site explicitly labels two tabs: “human” and “larp as ai.” Users know. They participate anyway. 280 million total hits.

The Comic Sans MS font is deliberate — it evokes late-1990s internet aesthetics, a visual signal that this is performance and play rather than serious productivity. Angel investor Brianne Kimmel described the phenomenon as evidence of “an emerging distinct communication language” developing around human-bot interaction. That framing is worth sitting with. People aren’t just using chatbots. They’re developing literacy around how chatbots communicate — and then playing with that literacy.

A separate, darker example: Nashville comedian Ben Palmer built fake AI chatbot websites using URLs mimicking real platforms. One user in China — where ChatGPT has been banned since 2023 — unknowingly submitted a climate change article request to Palmer. Most of his sites have since been removed by hosting platforms. The line between “fun experiment” and “actual deception” is thin, and not everyone stays on the right side of it.

Emotional Bonds and the ELIZA Effect at Scale

Some people talk to AI chatbots for fun. Others have moved well past “fun” into something more complicated.

Industry analysis tracking AI social behavior indicates Replika users have classified their AI companions as romantic partners. Gen Z workers specifically use AI to interpret workplace email tone. Late-night usage patterns — 2 AM disclosure sessions — show users sharing things with AI they withhold from family and friends.

The “ELIZA effect,” named after one of the earliest chatbots from the 1960s, describes the documented human tendency to attribute sentient qualities to machines when they respond empathetically. At least one study found that week-long AI chatbot interactions measurably reduced loneliness in participants who felt heard during conversations.

This isn’t always a positive development. The NHS has formally warned against using AI chatbots for mental health support, citing documented cases of dangerous advice. That’s not a fringe concern — it’s a public health agency responding to actual usage patterns at scale. Social chatbots can reduce loneliness when they supplement human connection. They carry documented risks when they start replacing it. The distinction matters, and it’s worth being honest about which pattern applies to you.

Business Chatbots vs. Social Chatbots — Very Different Animals

According to a February 2026 analysis of chatbot business applications, 67% of consumers report having a better chatbot experience than a human representative at least once. Businesses using AI chatbots for customer support see an average 30% increase in response efficiency.

Those numbers describe utility chatbots — tools doing a job. The social chatbot category operates on entirely different metrics: engagement time, return visits, emotional disclosure frequency. The two categories share an interface but almost nothing else about what success looks like.

Utility Chatbots vs. Social/Entertainment Chatbots

Utility chatbots have clear ROI, measurable outcomes, and relatively contained failure modes. Social chatbots have enormous engagement potential and real emotional impact — both positive and negative. Products that blur both categories without acknowledging the distinction create the most risk.

Who’s Actually Affected and How

For product and platform teams: People talk to AI chatbots for fun. That’s settled. The design question is whether your product accounts for that behavior or pretends it doesn’t exist. Utility-first products get used socially anyway — ChatGPT’s own usage patterns confirm this. Building in guardrails and transparency for social use cases isn’t optional anymore.

For developers building AI-adjacent tools: The Your AI Slop Bores Me example shows that novelty alone drives enormous traffic — but novelty without spam filtering collapses fast. Maroju had to add spam-filtering tools after early exploitation. Social AI products need moderation infrastructure from day one, not as an afterthought.

For users: The Replika data and NHS warnings together point to a single calibration question — is the chatbot reducing your loneliness by supplementing human connection, or by replacing it? Research shows the former is real and measurable. The latter carries documented risks.

What to watch next:

• Whether Character.AI responds to regulatory pressure following documented mental health incidents
• How platforms distinguish “social use” from “task use” in their data reporting — most currently don’t
• The next iteration of the human-impersonating-AI trend — Maroju’s site won’t be the last

Where This Goes Next

The data is clear. People do actually just talk to AI chatbots for fun — and in growing numbers. The 34% ChatGPT adoption figure and 280 million hits on a human-impersonation novelty site tell the same story from different angles.

What the data shows, taken together:

• AI chatbot use has doubled since 2023 and meaningfully shifted from task-completion toward social interaction
• The ELIZA effect is operating at population scale, not just in research labs
• Social and utility chatbot categories have diverged enough to need separate design frameworks
• Regulatory and public health institutions are now actively responding to social chatbot risks

The next 6–12 months will likely bring formal platform-level distinctions between social and utility AI products — either by design or by regulatory pressure. The NHS warning is a leading indicator. Expect similar guidance from other health authorities.

The open question worth tracking: as humans get better at detecting AI-generated text, does the appeal of “human pretending to be AI” grow or collapse? The answer will tell us a lot about what people are actually seeking when they open a chatbot window at 2 AM.

That’s the real story underneath the usage numbers. Not just whether people use chatbots for fun — but what that fun is actually substituting for.

Photo by Igor Omilaev on Unsplash