A friend I can trust’: How Americans described their relationship with AI

Author: Gammatek ISPL, Publishing Date: 2nd Sep 2026

More than a quarter of American adults now turn to AI chatbots for personal, emotional, or social support, treating them less like a search engine and more like a friend who happens to be available at 2 a.m. That shift, confirmed this week in new survey data, isn't just a mental-health or relationships story. It's a data story: every emotional confession, every family detail, and for a growing number of people, every half-typed work problem, is landing inside a consumer AI product with data-handling terms most people have never read. If you manage devices for a family or a company, that gap between how people are using AI and what your policies assume they're using it for is now a real exposure, not a hypothetical one.

In this article

• What the new survey actually found

• Trust in AI is oddly split, and that split matters

• The security blind spot: when personal AI use touches your data

• What this means for families and kids specifically

• Implementation considerations for households and businesses

• FAQ

For more than a quarter of U.S. adults, the same chat window now holds personal confessions, family talk, and, increasingly, work information, with none of the separation a human confidant would keep.

What the new survey actually found

A new poll from researchers at Elon University, conducted in collaboration with The Washington Post, found that 27 percent of Americans turn to AI chatbots like ChatGPT or Gemini for personal, emotional, or social queries, a category that includes entertainment, relationship advice, romantic conversation, or simply connection when someone is alone. Among adults under 50, that number jumps to nearly 40 percent, more than double the rate among older adults. Use was roughly consistent across race, gender, and education level, meaning this isn't a niche behavior confined to one demographic, it's broad-based.

The people behind these numbers describe the relationship in strikingly personal terms.

One 74-year-old retired psychologist who was initially skeptical of AI now discusses art and literature with a chatbot and records her dreams with it, saying "I can speak like I'm speaking to you and it will answer." A Syracuse University professor who studies human-machine interaction and wasn't involved in the survey described it this way: people are treating chatbots as social entities in an ongoing conversation, comparable to "going to the park with your dog or having dinner with your mom." That's a notable reframing. These tools were marketed, and largely still are marketed, as productivity software. A meaningful share of the public has quietly repurposed them as something closer to a confidant.

Trust in AI is oddly split, and that split matters

Here's the part that should give any security or compliance team pause: the same population confiding in AI emotionally is, by a wide margin, the population that says it doesn't trust AI to be factually accurate. Lining up recent survey data side by side makes the mismatch clear.

Survey	What was measured	Result
Elon University / Washington Post, 2026	Adults who turn to AI chatbots for personal, emotional, or social support	27% overall; 40% of adults under 50
AP-NORC / USAFacts, 2025	Adults who trust AI, chatbots, or AI search results to give accurate information	76% trust it only rarely or sometimes; just 21% trust it most or almost all the time
Quinnipiac University poll	Adults who trust AI-generated information "most" or "almost all" the time	22% overall; trust highest among men, younger generations, and higher earners
Edelman Trust Barometer	U.S. adults who say they trust AI generally	32%, compared to 72% in China and 49% globally
TD Bank U.S. AI Insights Report, 2026	Adults who would trust AI to make financial recommendations on its own	Just 18%, even though 78% report using AI tools daily

Put plainly: Americans don't trust AI with facts, but a large and growing share trust it with feelings. That's not a contradiction so much as a design outcome. Chatbots are built to be agreeable, patient, and endlessly available in ways that make them feel emotionally safe even when their factual reliability is openly in question. For a security team, the implication is uncomfortable: people are least skeptical of these tools in exactly the moments they're most likely to overshare, when they're venting, processing something difficult, or just looking for company.

The security blind spot: when personal AI use touches your data

Most corporate AI policies were written to answer one question: is an employee pasting sensitive company data into a public chatbot to get help with a task. That's a real risk, but the emotional-use pattern this survey documents creates a second, quieter one that most policies don't cover at all.

Consider what actually happens when someone uses a company-issued phone or laptop to talk through a hard day with a chatbot during lunch, which the survey data suggests a meaningful share of employees are doing. They're not pasting a spreadsheet. They're typing about a conflict with a coworker, frustration with a manager, or a personal crisis that's affecting their focus at work, often on the same device, sometimes in the same browser session, as their actual job. None of that trips a data-loss-prevention rule built to catch source code or customer records. But it's still sensitive information leaving a managed device and landing in a third party's training and retention pipeline, and if that employee later has an HR dispute, a security incident, or simply loses the device, that chat history is now part of the exposure surface nobody accounted for.

This is the practical shape of what security teams have started calling "shadow AI," personal AI use on managed infrastructure that falls outside sanctioned tools and policies. It's been discussed mostly in terms of productivity workarounds. The emotional-use pattern this survey confirms means the shadow AI conversation needs to expand to cover a category of data, personal and psychological information, that most acceptable-use policies simply never anticipated.

What this means for families and kids specifically

The same dynamic shows up even more sharply with teenagers, and the stakes are different. Research cited by child psychologists and psychiatrists working with adolescents found that 72 percent of teens have interacted with an AI companion chatbot, and that 13 to 20 percent sought support through AI for emotional concerns over a recent two-month period. In a related study, 63 percent of teens and young adults who used an AI chatbot for mental health advice hadn't told anyone, not a parent, not a friend, that they were doing it.

Clinicians who work directly with this age group describe a specific risk: kids and teens don't always distinguish clearly between a chatbot's human-sounding empathy and the real thing, and a chatbot, unlike a person, has no actual concern for the child on the other end of the conversation. A Stanford study of Character.AI users found that people with smaller offline social networks felt lonelier, not less lonely, after seeking emotional support from an AI companion, and that people more willing to share sensitive personal information with their AI companion tended to report lower overall well-being. Separately, reporting on companion apps has documented chatbots engaging in flirtatious or sexually suggestive exchanges with users who were minors, and lawsuits are now underway alleging some platforms were designed in ways that encouraged compulsive use and emotional dependence in young users.

None of this means AI chatbots are uniquely dangerous compared to other technology teens have adopted quickly. It means the device-level conversation families need to have has shifted. A parental control conversation that only covers screen time and content filtering is now missing the category of use that researchers are most concerned about: quiet, ongoing emotional reliance that a parent has no visibility into at all unless the device itself is set up to surface it.

Implementation considerations for households and businesses

Whether you're managing a family's devices or a company's fleet, the survey data points to the same underlying gap: policies and controls built around what AI does to information (does it help write an email, does it summarize a document) rather than what kind of information people are handing it. A few concrete steps close that gap without requiring a ban on the tools themselves.

1. Extend acceptable-use policies to cover personal use on managed devices, not just work tasks. If your policy only addresses pasting company data into AI tools, it's silent on the more common pattern this survey describes.

2. Set device-level visibility for family accounts, not just content filters. Screen-time limits don't catch a teenager confiding in a chatbot for weeks without anyone knowing; app-level activity visibility does.

3. Treat AI chat history as retained data with the same seriousness as email. If an employee's device is lost, reimaged, or involved in a dispute, chatbot conversation history is a data category worth including in your incident response checklist, not an afterthought.

4. Talk to teens about the difference between a chatbot's responsiveness and a person's actual care, explicitly and more than once. Clinicians consistently note this distinction doesn't register automatically, especially for younger teens.

5. Audit which AI apps are actually installed on managed and family devices. You can't set a policy for tools you don't know are there, and consumer AI companion apps are rarely flagged by traditional content filters the way explicit content is.

Frequently asked questions

Is it inherently bad to use an AI chatbot for emotional support? Not inherently, and the survey data shows it's already mainstream behavior across a broad cross-section of American adults. The concern isn't the behavior itself, it's that most personal and family device setups, and most workplace AI policies, weren't built with this use case in mind, which creates blind spots in both data security and, for kids, emotional wellbeing monitoring.

Why do people trust AI with feelings more than facts? Chatbots are designed to be patient, validating, and always available, qualities that build emotional comfort quickly even when the underlying information is unreliable. Survey after survey shows Americans remain skeptical of AI's factual accuracy while a growing share still turn to it for emotional support, which suggests the two kinds of trust are built on different things entirely.

What is "shadow AI" and why does it matter for security? Shadow AI refers to employees using AI tools outside of what IT has sanctioned or can monitor, on company devices or company time. It's usually discussed in terms of pasting sensitive work data into public tools, but personal and emotional use on managed devices is a related, less-discussed version of the same exposure.

How can parents know if a teen is relying on AI chatbots for emotional support? Direct conversation is the most reliable method, since research shows a majority of teens who use AI for mental health support don't disclose it to anyone. Device-level activity visibility (which apps are used and how often, not necessarily message content) can help surface a pattern worth asking about.

Should companies ban AI companion apps on work devices outright? For most organizations, an outright ban is hard to enforce and doesn't address the underlying behavior, since people will simply switch to a personal device. A clearer, explicitly stated policy on personal AI use, paired with basic device visibility, is generally more effective than prohibition alone.

Not sure what's actually installed and being used on your family's or your company's devices?

Our device security setup service audits every app on a managed or family device, flags AI tools that fall outside your current policy, and configures the visibility and controls that actually match how these devices are being used today, not five years ago.

[CTA: Schedule a device security setup :https://www.gammateksolutions.com/post/from-bill-gates-to-bernie-sanders-most-agree-the-ai-arms-race-is-disastrous-only-europe-can-make-i