Companies have spent years building mental-health benefits, employee-assistance programs, wellness portals, manager trainings, HR processes and increasingly polished language about psychological safety. Then a chatbot arrived with a blank text box and no visible place on the org chart.
A new workplace survey suggests a surprising number of employees prefer the blank text box.
Founder Reports surveyed 2,000 employed U.S. adults in July 2026 about work-related strain, mental-health resources, concealment and AI use. The resulting Workplace Wellbeing Report found that 42.6% had used an AI chatbot to talk through work stress or frustration. By comparison, just 10.7% said they had ever used mental-health support provided by their employer. Another 19.7% had considered using AI for work stress, meaning nearly two-thirds of respondents had either done it or thought about doing it.
Those numbers are easy to turn into a story about artificial intelligence replacing human support. That is probably the less interesting reading. The better one is about trust architecture inside the workplace: who gets access to an employee’s unedited account of what work is doing to them, and why.
The same survey found that 75.2% of workers had done at least one thing in the past year to conceal work-related distress. Only 10.8% said they had told a manager or HR representative what they were going through. Nearly 39% had called in sick and given a reason other than the real one. The important pattern is not simply that employees are stressed. It is that a large share are actively managing who is allowed to know.
That management becomes more revealing as stress rises. Among workers who said their jobs never hurt their well-being, 26.6% said they could be completely honest with their manager, compared with 9.2% who said the same about an AI chatbot. Among workers experiencing daily strain, the relationship flipped: 12.3% said they could be completely honest with a manager, while 28.9% said they could be completely honest with AI. Founder Reports calls this the “trust crossover.” The people who most need to say something about work appear to become less comfortable saying it to the person who represents work.
That is not a technology problem in the usual sense. It is an institutional-design problem with a technology-shaped escape hatch.
The chatbot occupies a peculiar position. It is not a therapist, even when a user talks to it in therapeutic language. It is not a coworker because it does not share the politics of the office. It is not HR because it does not represent the employer. It is not a friend because it has no stake in the employee’s social life. What it offers, at least from the user’s perspective, is a conversational space with unusually low social friction. There is no awkward pause after admitting you cannot keep up. No manager suddenly recalculating whether you are “resilient.” No concern that a private confession will quietly become part of somebody’s understanding of your promotability.
The machine may be wrong. It may give mediocre advice. Depending on the product and account being used, the conversation may not be as private as the employee imagines. But those are separate questions from why the employee opened the window in the first place. The immediate attraction can be much simpler: the chatbot does not look disappointed.
That distinction matters because employers often treat mental-health support as an availability problem. Offer the benefit. Publish the EAP number. Add a wellness section to the intranet. Train managers. Remind employees during Mental Health Awareness Month. Those things can be useful, but availability and usability are not the same thing, and usability includes social risk.
The 2026 NAMI-Ipsos Workplace Mental Health Poll points in the same direction. Seventy-four percent of employees said discussing mental health at work is appropriate, yet only 61% said they were comfortable sharing their own mental health at work. Comfort fell further when the audience was HR or senior leadership. Nearly half worried they would be judged for sharing mental-health struggles, and among people uncomfortable discussing the issue at work, fear of retaliation remained one of the reasons cited. Only 15% said they had told their manager that their mental health was suffering because of work.
In other words, workplaces may have succeeded at normalizing the abstract principle that mental health matters without fully solving the practical question of what happens to a person after they admit theirs is deteriorating.
That gap is where the chatbot fits.
The Founder Reports data becomes especially useful because it does not support the simple idea that AI users have abandoned human help. Workers who felt comfortable confiding in AI were also more likely to trust therapists and more likely to have used employer-provided mental-health benefits than workers who were not comfortable with AI. Only 2.1% of the sample trusted an AI chatbot while trusting no human source at all. The people turning to AI were, in many cases, more help-seeking rather than less.
Their notable exception was the workplace itself.
That changes the managerial interpretation. If employees were rejecting every form of human support in favor of a machine, the obvious concern would be social withdrawal. But if employees are willing to speak with therapists, friends or other human supports while becoming less willing to speak honestly with managers as work stress increases, the machine is not necessarily replacing human connection. It may be routing around a specific trust failure.
This is why “we already offer mental-health benefits” can be a technically correct answer to the wrong question. A benefit can exist and still sit behind too much uncertainty, stigma, administrative friction or perceived exposure to become the first place an employee turns. Founder Reports found 38.8% of respondents said their employer offered no mental-health support, while another 11.2% were not sure whether any support existed. Among workers who did have benefits available, utilization remained relatively flat as stress increased, even while AI use climbed.
NAMI’s 2026 findings similarly show that only six in ten employees know how to access mental-health care through employer-sponsored insurance, and a majority said they had not received training about their workplace’s mental-health resources. The infrastructure can be present without feeling reachable.
AI changes that equation because reachability is almost the entire product. Open a window. Type the thought. Get an answer. No appointment. No explanation of eligibility. No need to decide whether the problem is severe enough to count as a mental-health issue. No requirement to translate “my boss is making me miserable” into the language of a benefits portal.
That convenience should not be confused with clinical competence. A chatbot can help someone organize thoughts, rehearse a conversation or put language around a frustrating situation, but those functions are different from professional mental-health care. The workplace lesson is not that employers should replace benefits with bots. It is that employees have discovered a channel with radically lower disclosure friction, and organizations should be curious about why their own channels feel harder to use.
There is another layer employers will eventually have to confront: workers may perceive AI as a private room even when the technology does not guarantee privacy in the way they assume. Consumer chatbots, enterprise AI products and employer-managed accounts can have very different data-handling rules. A person venting about a manager may think primarily about the absence of social judgment, not about retention settings, account ownership or organizational policy. The emotional feeling of privacy and the technical reality of privacy are not automatically the same.
That makes the phenomenon structurally strange. Employees may be choosing AI because they do not want the organization to know what the organization is doing to them, while sometimes using tools that the organization itself licensed, approved or placed on their laptop.
The workplace has built a new room without fully deciding whether it is a confessional, a productivity tool, a coaching layer, a searchable record or all four at once.
For managers, the temptation will be to make this an AI-governance issue. Some of it is. Companies need sensible policies about what employees should and should not put into workplace AI systems, particularly when conversations include sensitive personal information, confidential business details or identifiable information about coworkers. But governance alone does not answer the harder organizational question raised by the survey: why did the employee prefer typing it there?
The answer may differ by company. Some workers may fear judgment. Some may distrust HR’s independence. Some may not know what benefits exist. Some may be trying to think through a problem before involving anyone else. Some may simply want to complain at 11:30 at night without turning the complaint into an official event. A chatbot collapses all those motivations into the same easy interface.
That is what makes it such a revealing workplace instrument. Not because it necessarily understands the employee better than a human manager does, but because employees may behave around it differently. The things a person tells a chatbot can expose the distance between the organization’s stated culture and the employee’s experienced culture.
A company may say, “Talk to your manager.” The employee may hear, “Create a record that you are struggling.” A company may say, “Use the benefits we provide.” The employee may hear, “Decide whether this is serious enough to become a formal mental-health matter.” The chatbot asks for almost none of that interpretation before the first sentence.
The AI did not invent those fears. It simply created somewhere else for the sentence to go.
That may be the more consequential workplace story emerging from the survey. Artificial intelligence is developing a role inside organizations that nobody exactly designed: a private-seeming intermediary between an employee’s internal experience and the institution around them. It can become the rehearsal room before a difficult meeting, the place where someone tests whether a complaint sounds reasonable, the screen that receives the first unedited version of a problem before the employee converts it into something professionally survivable.
For employers, the uncomfortable metric is not how many workers are using chatbots. It is how many are telling the chatbot something they believe they cannot safely tell you.
The AI did not replace HR. It became the room people go into before deciding whether HR gets to know.
SOURCE NOTES
• Founder Reports — 2026 Workplace Wellbeing Report
• Founder Reports — “Workers Are Skipping Their Company’s Mental Health Benefits and Talking to AI Instead,” Aug. 20, 2026
• NAMI — 2026 NAMI-Ipsos Workplace Mental Health Poll
• Ipsos — “Most Employees Say Mental Health Talk is Work-Appropriate; Stigma Persists,” 2026
• SHRM — 2026 Mental Health Snapshot
• Inc. — “Workers Trust a Chatbot With the Stress They Won’t Tell Their Boss, New Survey Finds,” Sept. 8, 2026
Survey findings attributed to Founder Reports, NAMI/Ipsos, SHRM and Inc. coverage cited in SOURCE NOTES. A chatbot is not a therapist; felt privacy is not the same as technical privacy. Cultural framing is RMN's.