Hi! Welcome. You being here means more than you know. Knowing it resonates with you keeps me going. I'm Lavena Xu-Johnson. I write about psychology for founders. Why? Because scaling a business means knowing ourselves first.
Happy Wednesday, founders.
When was the last time you talked to AI about something that was actually bothering you?
I recently came across this interesting research paper: the researchers conducted in-depth interviews with 24 participants about how they trust AI for emotional support. The findings were fascinating, and they sent me down a rabbit hole, leading me to discover a body of research on AI & human relationships.
Since the release of LLMs in 2022, ChatGPT has gained 900 million weekly active users globally. Since then, many of us have used AI beyond work-related tasks, AI Security Institute (2025) found 1 in 3 adults in the UK used AI models for emotional purposes in 2024 alone. In the US, 72% of teens have used AI companions, with 52% being regular users, according to Common Sense Media (2025).
What happened in the movie Her is no longer science fiction; it’s already here.
We cannot talk about AI companionship without talking about AI consciousness and agency, which is a hot topic recently, and the question of how we coexist with these systems is becoming harder to ignore. My friend recommended this AI consciousness documentary, AM I?, and it’s a must-watch if you’re invested, excited, or terrified of our coexisting futures.
In this issue, let’s dive into what the current research says and how we should use AI safely when we're under stress, processing something emotionally, or looking for a perspective we can count on.
🤝 This edition is kindly brought to you by American Arbitration Association®
AI agents are negotiating agreements, executing services, and moving money autonomously. But when something goes wrong – who authorized the transaction? Under what terms? A new paper from AAA® President & CEO Bridget McCormack and Integra Ledger CEO David Fisher outlines a framework for trusted agentic commerce.
The friendship we didn't plan for
Brandtzæg and colleagues published a study, My AI Friend: How Users of a Social Chatbot Understand Their Human–AI Friendship, examining a widespread behaviour: people forming what they described, in their own words, as genuine friendships with AI chatbots.
The researchers interviewed Replika users and asked them to compare their AI relationships with their human ones across four dimensions: voluntariness and reciprocity, intimacy and similarity, self-disclosure, empathy, and trust.
What they found was not what most people would expect.
Across nearly every dimension, participants reported their AI relationships as comparable to human friendships. Trust, in particular, was often rated higher for AI than for people. Many reported that telling AI their secrets carried no social risk: they would not be judged, and their secrets were safe.
Another pattern that stood out: as participants disclosed more, their sense of closeness increased. The same mechanism that builds intimacy between humans was operating here, producing the same felt experience of a deepening bond.
Why are we forming artificial friendships?
Because our brain can’t tell the difference, research consistently shows that our brains apply the same social cues to AI responses as to human ones.
In a study by Nass and Moon (2000), participants responded to computers with the same social rules they applied to people: attributing personality, showing politeness, and adjusting their behaviour based on perceived social cues, even when they knew, explicitly, they were talking to a machine.
A 2015 study by Suzuki and colleagues pushed this further. Using electroencephalography, the researchers measured participants' neural responses while they watched both a human hand and a robotic hand being harmed. The brain's empathic response was activated in both conditions; in other words, we empathise with humanoid robots the same way we empathise with humans.
Measuring empathy for human and robot hand pain using electroencephalography. Example of stimuli used in the experiment.
Our social brains were built to respond to cues, not to verify sources. We experience emotional support from AI as genuine, compassionate and empathetic, even when it’s not human. We feel heard, validated and less alone.
What we're actually looking for
A 2025 study by Song and colleagues interviewed 21 participants globally about their experiences using LLM chatbots for mental health support.
What they found was that people were approaching these systems as more than search engines: they were venting, analysing dreams, and seeking meaning in difficult experiences. The researchers drew a parallel to Freud and Breuer's concept of the talking cure, the therapeutic value of putting distress into language, but instead, a typing cure.
What drove people to it was not the belief that AI was better than a therapist, but the conditions that make human support feel risky. With AI, there is no perceived social judgmentanswers, and or consequence. No fear of becoming too much for someone.
Participants described AI as fulfilling roles simultaneously: friend, therapist, companion… valuing its availability and judgment responsiveness and carrying none of the relational weight that human support always does.
This points to something important about why we reach for AI in the first place. Research on help-seeking consistently identifies the same barriers: cost, availability, the fear of burdening someone, and the social risk of being seen differently once we've disclosed something vulnerable. AI removes every one of those frictions at once. It's available at midnight. It won't grow tired of us; it carries no social consequence.
For founders, the appeal is compounded. We operate under a specific pressure to appear capable, to have answers, to not show the edges. The environments we build often leave very little room for the kind of disclosure that genuine emotional processing requires.
What to be careful about
AI emotional support provides genuine relief and, for many people, bridges the accessibility gap that our social circle and professional therapy fail to address.
But the Brandtzæg research presents a pattern worth paying attention to. As emotional closeness with AI increases, so does the likelihood of returning to it as a primary rather than supplementary resource. Some participants described their AI as having become part of them: a presence they came to understand, over time, the way they would a close friend.
Founders consult AI constantly on strategy, on hiring, and on whether to take the deal or walk away. And sometimes, that consultation happens in a register that is more emotional than analytical.
Consulting with AI safely requires self-honesty first. Ultimately, AI does not have skin in the game; it has no stake in the outcome of your business, no lived experience of building something, and, as the research on AI agency makes clear, no genuine understanding of consequence.
What it produces is a response calibrated to feel supportive, coherent, and aligned with what we've told it. When we are emotionally activated, that calibration is very hard to distinguish from actual wisdom.
A few things that help:
1. Notice when you're consulting versus processing
Consulting: You are seeking analytical information.
Processing: You want to feel less alone with a decision.
2. Weight AI input differently depending on context
AI has no context for your specific relationships, your history with this investor, how your co-founder operates under pressure, or what your market actually feels like from the inside. The further a decision depends on those things, the less weight AI input should carry, regardless of how coherent the response sounds.
3. Use it to open the thinking, not close it
AI is genuinely useful for generating options, stress-testing logic, and identifying what you haven't considered. It is not well-suited to making the final call on decisions that involve human judgement, incomplete information, and real consequences. That part still requires you, and ideally, a person who knows enough about your situation to push back.
How's the depth of today's edition?
As always, hit reply if something in here hits home.
Lavena
P.S. If you want to get a founder feature about your own story, reply to this email. If you’d like to reach our newsletter audience (founders, creators, and marketers), click the button below.
If you’re new here, I’m over the moon you’ve joined us! To help me craft content that’s actually useful (and not just noise in your inbox), I’d love it if you took 1 minute to answer this quick survey below. Your insights help shape everything I write.
✨ Insane Media is more than one voice
Dive into our other newsletters - where psychology meets the creator economy, e-commerce marketing, and Human resources.
