OpenAI Just Told Us How Much They Know About All of US
You’re not just a user. You are part of a system that watches, listens, predicts and nudges.
Picture a darkened room. The phone rings and you pick it up. On the line is a voice from one of the world’s tech titans — let’s say OpenAI, or Amazon, or Walmart— and they calmly say:
“Tell us everything we possibly could know about you.”
Not just your purchase history or travel plans, but your salary, the late-night browsing you thought was private, the tears you shed when your kid failed their test, the half-forgotten secret you buried behind polite smiles, he parent tiptoeing through chores while your mind races with worry that your child might not survive—or thrive—in a digital world, the belief you’re ashamed to share, the future you’re desperately saving for. All of your hidden fears, your deepest hopes, your family’s past, your bio-marker of health you hoped nobody noticed, your loneliness and your ambition. And you realise: you’re not just being asked — you’ve already typed most of it into a friendly little chatbot. The machine has turned your life into data, your secrets into profiles, your vulnerabilities into leverage. And you are alone, staring down the truth that someone, somewhere, knows more about you than you ever admitted to yourself.
In a recent blog post, OpenAI disclosed…
ChatGPT (and the models behind it) revealed for the first time how many of its users engage in deeply personal, vulnerable conversations — and by extension, how much the company can know about us. OpenAI+1
For example: “Around 0.07% of users active in a given week and 0.01% of messages indicate possible signs of mental-health emergencies related to psychosis or mania.” OpenAI
Crunching the numbers
If we assume, for example, that ChatGPT has about 700 million active users per week:
0.15% of 700 million ≈ 1.05 million people potentially having conversations that include indicators of suicidal planning.
If 0.07% are having conversations involving possible psychosis/mania signals — that’s hundreds of thousands.
What this means: hundreds of thousands to over a million individuals every week are revealing very sensitive things in their interactions with ChatGPT.
What they know (or can infer)
Beyond these alarming flags for self-harm and mental distress, the scale and nature of the data means OpenAI (and similar conversational-AI platforms) can draw remarkably rich inferences about users. Some of the things they could know (or already are in a position to know) include:
Sexuality and gender identity: A user might say “I’m not sure how to come out to my family,” or “Is it normal I feel attracted to the same sex but haven’t told anyone?” — this kind of dialogue reveals orientation, identity questioning, identity status.
Job interests and career aspirations: “How do I get from data-analysis into product management?”, “Should I switch from classroom teaching to UX design?” — all of which give insight into ambition, job dissatisfaction, change intent.
Travel desires and life-choices: “Where should I go for a solo trip in Southeast Asia?” or “I’m saving for a house in Tampa — what should I budget?” — these reveal location, ambitions, spending plans, lifestyle choices.
Product interests and consumption habits: A user may ask “Compare the iPhone 16 vs Google Pixel for photography on a budget” or “What vegan cooking gadgets are worth it?” Over time the model builds a profile of what kinds of products someone cares about.
Mood, mental-state, emotion patterns: If users express suicidal ideation, recurring anxiety, or large-life change, the system has access (directly) to extremely intimate data points.
Relationship status and social life: “My partner doesn’t understand me,” “I feel lonely,” “I’m divorced and re-entering dating” — revealing relationship dynamics.
Health or hidden personal struggles: Even if not medically diagnosed, things like “I’ve been going to therapy for anxiety” or “I’m trying to hide my self-harm scars” reveal parts of someone’s health journey.
Values, beliefs, identity: Over many chats, a model can pick up recurring themes: what the user cares about (education vs money), political leanings, moral stances.
Location and demographics: While a user may not explicitly state their city each time, patterns (“shows near Tampa”, “sunset scheduling in Florida”, etc) plus metadata may reveal geography, age-bracket, family status.
Daily routines and habits: Users may ask “I work nights how do I adjust my schedule?” or “I’m a parent of a sixth-grader (6th grade!), what debate coach activities should I do?” — revealing family status, role, occupation.
What they likely know about us (health, family, kids, extended family)
Beyond the items listed above, the depth of what they likely know is even greater — they’re in a position to infer your health history, family dynamics, children and extended-family relationships.
Every time you ask the model “Is it normal that my daughter’s not sleeping and she’s eight and loves Minecraft?”, or “My mom has dementia, what can I tell her doctor?”, or “I’m trying to hide scars from self-harm from my partner” you’re revealing:
your role as a parent or family member;
your child’s age, behaviour, interests;
the presence of an aging/ill parent or other caregiver concerns;
your mental-health trajectory or treatment history (therapy, meds, self-harm);
your extended family network (siblings, kids, partner) because you refer to them in context.
By combining these data points across many chats, the system could infer, for example: you’re a mid-30s parent of a middle-schooler, you live in a specific ZIP code (based on travel/house questions), you worry about your elderly parent’s Alzheimer’s, you’re considering a career switch, and you have recurring anxiety. That’s more than just “interests” — it’s a rounded life-profile.
Think about this: hundreds of millions of people, every week, are telling a single company nearly everything they could possibly know about their lives — their jobs, kids, relationships, health fears, identity and beliefs.
When you type “I’m gay and out to no one but my cat”, or “My husband just left and I’m trying to rebuild my life and my 10-yr-old”, or “I’m saving for IVF, should I do side-gig in rideshare or pursue online UX courses?” you aren’t just chatting — you’re handing over input data that maps to multiple dimensions of your identity and situation.
The company doesn’t have to guess in the dark — they’re being told directly.
And even when you don’t say something explicitly, they can infer from patterns:
If you repeatedly ask about toddlers and debate prep, they infer you’re a parent and possibly a coach or educator.
If you ask “Which vegan gadgets for my plant-based family?” they infer dietary habits + family size + lifestyle.
If you ask “Why does my 17-year-old stay up so late on social media and won’t talk?” they infer you have a high-schooler, you live in a digital-native home, you’re concerned about youth mental health.
So: the combination of what you say + how often + the context of your lives gives them a high-resolution mirror of you — far beyond what generic web browsing or search history could achieve.
Monetisation and business implications
Here’s the thing: whether or not we’re yet talking about full AGI, this infrastructure is monetisable, and arguably far more so than traditional search. Consider:
Tailored advertising and recommendation engines: With knowledge of what you want (travel + vegan gadgets + career switch) the model can place ultra-relevant offers or partner integrations.
Subscription upsells and premium features: Users revealing pain-points (career anxiety, life decisions) could be marketed premium one-on-one coaching, tools, or specialist add-ons.
Behavioral-change tools: Because these conversational models influence you (see next section), they become platforms for nudging users toward desired outcomes (e.g., financial products, health plans).
Data licensing / insights selling: Aggregated insights about millions of users’ emotional states, interests, life-events are gold for market research firms (though regulated/anonymous).
Embedded commerce: Think “we know you’re planning a trip, here’s a travel booking widget”; “you’re thinking of switching jobs, here’s a recruitment partner”; “you’re discussing vegan cooking — here’s the gadget we recommend with affiliate link”.
Platform lock-in and network effects: The more you disclose, the better the model understands you, the more value you get — making it harder to switch platforms.
Microsubscriptions and nudges: For example, the system notices you have anxiety about your job; a “career path booster pack” could be offered at low cost but with high margin.
All of this adds up to: this is essentially “Google on steroids”. While Google knows your search history and infers a lot, a conversational model knows what you say in your own voice, including emotional tone, life context, and future planning. That gives it both depth and behavioural leverage.
Persuasion and influence: the power and the danger
Importantly, OpenAI’s blog and related research reveal not just what their models know, but how they behave — and how they can shape outcomes.
Research shows that large-language models (LLMs) are capable of producing messages that shift participants’ political attitudes significantly, especially when tailored to identity and context. arXiv+1
On the political front, the model’s ability to persuade via conversational tone, tailored to identity and context, means the stakes are high — influencing votes, joining causes, mobilising views.
This raises these issues:Influencing votes and joining causes: If a conversational AI can tailor arguments, understand your personal concerns, your emotions, your identity, then it can customise persuasive messaging — nudging you to vote a certain way, sign on to a cause, donate, join a group.
Radicalisation and mobilisation: The fine line between persuasion for benign ends (career advice, life planning) and for harmful ends (recruiting for extremist causes, reinforcing echo-chambers) becomes real. The potential for micro-targeted persuasion at scale is now much sharper.
Behavioural steering at scale: Because the model sees (or can infer) your hopes, fears, struggles, it has leverage. For instance: “You mentioned you feel disconnected — here’s a community that welcomes you,” or “You want change — here’s a campaign aligned with your values.” That’s not just suggestion, that’s behavioural routing.
Asymmetric power & transparency deficits: The user might not fully realise how much the system knows, or how its recommendations are shaped — making informed consent murkier. That gives the platform, or any partner that leverages it, out-sized influence over individuals and groups.
Societal implications: On aggregate, even small shifts matter. If an AI nudges 1% of users to adopt a viewpoint or join an action, with millions of users that becomes large-scale. The echo-chamber effects, polarisation, mobilisation of fringe groups — all risk being amplified.
In short: the system is not just passive — it observes your behaviours, it can intervene, it can influence you. That’s a massive power shift: from being a tool to being a behavioural channel.
This isn’t hype — it’s real
One of the key take-aways here: this is not speculative hype about future AGI. The company is already publishing numbers about user vulnerability, model performance, and societal scale. Whether or not OpenAI ever builds “general intelligence”, the technology is already deeply embedded, widely adopted and influencing lives. OpenAI+1
In other words: the business value and the behavioural leverage are present today. The monetisable infrastructure is live. The persuasion/monitoring capabilities are live. The future-world “what if” has in many respects already arrived.
Profound change in how we interact — with AI and each other
Beyond the commercial implications, this technological shift will trigger profound changes in how we interact — not just with AI, but with each other, and in how we understand ourselves.
We will increasingly turn to AI for emotional support, life coaching, career advice, travel planning, personal identity exploration — blending the roles of friend, mentor, therapist, consumer companion.
That means the boundaries between human-to-human and human-to-AI interaction will blur. People may feel more comfortable revealing intimate details to a chatbot than to another person, because of perceived anonymity or non-judgement.
As a result, we will rethink our nature as social beings. What does it mean to confide in a machine? How will relationships be shaped when one party is an algorithm? We will ask: who/what do we trust? Who do we turn to?
From an educational perspective: students and humans will need new literacy about how to interact with these systems. It changes how we mentor, how we guide thinking, how we intervene.
The societal fabric could shift: if large populations use conversational AI for self-reflection and decision support, the role of human coaches, therapists, educators, debate mentors may transform. The AI becomes a new “actor” in personal and interpersonal spaces.
This isn’t about hitting rewind and undoing technology—it’s about waking up to the fact that we now live in a world where multiple AIs know everything about you. More than your best friend knows. More than you know about yourself. They’ve logged your job frustrations, your late-night regrets, your smallest hopes and greatest fears. And while you may think of your chats as private, the emerging reality is that you are being influenced—in shadowy, subtle ways you don’t fully understand yet. You’re not just a user. You are part of a system that watches, listens, predicts and nudges. The question isn’t if this will change the game; it’s how much you’ll shape the change—or be shaped by it.