350 Handwritten Confessions Map How People Really Use AI

Artist Olivia Tai has spent a year gathering more than 350 anonymous handwritten confessions for her project "What We Tell AI," documenting how people quietly use chatbots to handle dating, work and daily life.

A year of collecting scraps of paper has produced something the consumer technology industry rarely gets: an unfiltered account of what people actually do with artificial intelligence when nobody is watching, and what they refuse to admit to.

Artist Olivia Tai has gathered more than 350 handwritten secrets for her project "What We Tell AI," anonymous confessions about how people lean on chatbots to navigate dating, work and the ordinary business of living. The premise, as she frames it, is blunt: everyone has an AI secret. The collection was reported by CNBC.

Handwriting as a workaround for the survey problem

The method matters as much as the material. Almost everything the market knows about consumer AI usage arrives through channels that reward flattering answers: corporate telemetry that counts sessions but not motives, developer-run surveys, app store reviews written under a username. People misreport habits they are embarrassed by. That is a well-worn limitation of self-reported research, and it bites hardest exactly where AI is most interesting — the tasks users consider slightly shameful to outsource.

A pen, a slip of paper and no name attached removes most of that friction. The confessions are not a statistically representative sample and should not be read as one; they are 350-plus volunteered disclosures from people who chose to participate. But qualitative material of this kind has a long history of arriving ahead of the quantitative picture, because it captures the reason behind a behavior rather than just its frequency.

Dating, work and life: three very different disclosure problems

The three domains the project spans are not equivalent, and the difference is commercially significant.

• Dating. Using a model to draft a message to a romantic prospect touches authenticity directly. The concealment is aimed at one specific person, and the perceived offense is impersonation of one's own voice.
• Work. Here the hiding is institutional. Employees weigh policy risk, the appearance of diminished effort, and the awkward possibility that disclosing how fast a task was completed invites more of it. Concealment is rational self-interest, not shame.
• Life. The broadest and least legible category — decisions, worries, questions a person might once have taken to a friend or a professional. Concealment here is about not wanting to look like someone who needed to ask a machine.

For anyone building or funding consumer AI products, that taxonomy is the useful part. A product used in secret still gets used. But secrecy suppresses the two things that have historically driven cheap growth in consumer software: word of mouth and visible social proof. A tool nobody admits to using does not go viral in the ordinary way.

What hidden usage does to the growth math

Consumer adoption curves are usually powered by conspicuous use. People saw the phone in someone's hand, the app icon on a friend's screen, the shared playlist. AI in the confession mode inverts that. Usage may be deep and daily while remaining almost entirely invisible in the social layer, which means acquisition has to be paid for or bundled rather than earned through advocacy.

It also complicates measurement of engagement quality. A user who quietly relies on a model for emotionally weighted tasks is, on any retention metric, a valuable user — but not one who will appear in a testimonial, complete a referral, or answer a satisfaction survey honestly. Companies reading their own dashboards may be systematically underestimating both the intimacy and the stickiness of the relationship they have built.

There is a product design consequence. If a meaningful share of high-value usage is usage people would rather not have logged, then privacy features stop being a compliance checkbox and become a growth lever. Local processing, deletable histories, incognito-style modes and clear data-retention promises speak directly to the person writing an anonymous note on a piece of paper rather than telling a colleague.

A culture project arriving in a market priced for AI

The confessions land at a moment when equity markets have committed enormous capital to the assumption that AI use will keep broadening. Index levels give a rough read on the temperature. At the last close on Sept. 4, 2026, the S&P 500 tracker (NYSEARCA: SPY) finished at $770.19, down 0.39% on the day from a previous close of $773.17. The tech-heavy Nasdaq 100 fund (NASDAQ: QQQ) closed at $718.96, up 0.18%, with the Dow 30 vehicle (NYSEARCA: DIA) at $534.08, off 0.53%. The divergence — technology holding while the broader and more industrial gauges slipped — is the shape the market has worn for much of this cycle.

None of that is caused by an art project, and no one should trade on 350 pieces of paper. But the gap between what people tell pollsters about AI and what they write down anonymously is the kind of gap that eventually shows up in revenue lines, either as upside that models missed or as backlash that arrived without warning.

What the project suggests to watch next

Three threads are worth following from here.

• Whether disclosure norms harden or soften. Workplace policies on AI use are still being written. If employers move toward mandatory disclosure, a large body of currently hidden usage becomes either sanctioned or punished — and behavior will shift sharply either way.
• Whether privacy becomes a marketed feature. Watch which providers start advertising what they do not keep. Secrecy-driven usage rewards that positioning.
• Whether the emotional use cases get built for deliberately. Companionship, advice and reassurance are showing up in the confessions category. They are also the uses most likely to draw regulatory attention and the hardest to price.

Tai's archive is a piece of cultural documentation, not market research, and it should be read as such. Its value is that it captured the gap most instruments cannot: the distance between how people describe their relationship with these tools and how they actually conduct it. A year of collecting handwriting suggests that gap is wide, and that a great deal of AI adoption is happening in a register no dashboard is currently built to see.

Key facts

• Confessions collected: More than 350 handwritten anonymous secrets
• Project: "What We Tell AI," run by artist Olivia Tai
• Duration: About one year of collecting and sharing
• Market context (last close, Sept. 4, 2026, 20:00 GMT): QQQ $718.96 (+0.18%); SPY $770.19 (-0.39%)

Frequently asked questions

What is "What We Tell AI"?

"What We Tell AI" is a project by artist Olivia Tai that collects and shares anonymous, handwritten confessions about how people use artificial intelligence. Over roughly a year she has gathered more than 350 of these secrets, which describe the ways individuals rely on AI tools to handle dating, work and everyday personal decisions.

How many confessions has the project collected?

More than 350 handwritten secrets have been collected over about a year. Each is anonymous, written by hand rather than typed or submitted through a form, which removes the identity signals that normally discourage people from admitting to behavior they consider embarrassing or professionally risky.

Why do people hide their AI use?

The project spans three areas with distinct motives. In dating, concealment is about authenticity — not wanting a specific person to know a message was drafted with help. At work, it is self-protection against policy risk or the appearance of reduced effort. In personal life, it is largely about not wanting to look like someone who needed to ask a machine.

Is this a representative survey of AI users?

No. The collection is qualitative and self-selected: it captures confessions from people who chose to participate, not a statistically drawn sample. Its value is in explaining motives behind behavior rather than measuring how common that behavior is across the general population. It should not be treated as market research or polling data.

Why does hidden usage matter to AI companies?

Consumer software has historically grown through visible use — recommendations, shared screens, social proof. Usage that people conceal generates deep engagement but little word of mouth, which pushes acquisition costs onto paid channels. It also means internal surveys and testimonials may understate how heavily and how intimately customers actually rely on a product.

What were the major indexes at the most recent close?

As of the last trade on Sept. 4, 2026 at 20:00 GMT, the S&P 500 tracker SPY closed at $770.19, down 0.39% from a previous close of $773.17. The Nasdaq 100 fund QQQ closed at $718.96, up 0.18%, and the Dow 30 vehicle DIA finished at $534.08, down 0.53%.

Sources

• 'Everyone has an AI secret': This artist has spent a year collecting and sharing hundreds of anonymous AI confessions — CNBC Top News

Photo: rao qingwei · Pexels Licence — source