Everything the Internet Knows About You Is Worse Than Nothing

Ask an AI assistant for a book recommendation. A genuinely personalized one, not just "what are popular books in your genre."

Now imagine that AI has access to everything you've ever made public: your LinkedIn profile, your published articles, your Amazon reviews, your GitHub commits, your conference talks. It has scraped years of your professional life. It knows what you work on, what you've said publicly about your field, how you present yourself to colleagues and employers.

It gives you a recommendation. It's exactly wrong. Not "not quite right" — exactly wrong. The book it suggests is one you would have rolled your eyes at five years ago. It doesn't know that you read literary fiction for pleasure, that your actual taste in nonfiction runs to history rather than business, that you're suspicious of any book with a subtitle containing the words "thinking" and "better." It knows your front stage. It doesn't know you.

This is worse than starting cold. A blank AI would make a guess. This one makes a confident guess, grounded in evidence, and the evidence is systematically misleading.

The front-stage problem

In 1959, the sociologist Erving Goffman published The Presentation of Self in Everyday Life, which argued that social interaction is essentially theatrical. We manage our "performances" for different audiences — behaving differently at a job interview than at dinner with close friends. Goffman called the public performance the "front stage" and the unguarded version the "back stage."

What the internet has captured about most people is an almost perfect record of their front stage. Your professional history. Your considered opinions. The reviews you wrote after you were satisfied with a purchase. The tweets you chose not to delete.

That record is not false — it's just radically incomplete. And the incompleteness isn't random noise. It's systematically skewed toward the version of yourself you perform for others. Your professional persona is optimized to communicate capability and credibility. Your public opinions are filtered through the lens of how you want to be perceived. Your Amazon reviews reflect what you were willing to say publicly after the fact, not what you actually felt in the moment.

Seventy-three percent of people in a recent study agreed that social platforms increase the gap between what people say they care about and how they actually behave. That's not hypocrisy — it's just the nature of public output. We all know the gap is there. The AI doesn't.

Confidently wrong is worse than agnostic

Here's the part that actually matters: an AI without context will hedge. It will ask. It will generalize from broad patterns and present those generalizations loosely, because it knows it doesn't know you.

An AI that has ingested your public footprint has evidence. It will form confident priors. And confident wrong priors are genuinely harder to dislodge than admitted ignorance.

If an AI thinks you're "a technology professional interested in productivity and systems thinking" — which is exactly what your LinkedIn and published writing would suggest — it will keep applying that lens. Every recommendation, every summary, every suggestion will be slightly warped by that frame. And you won't always notice, because the frame isn't entirely wrong. It just misses everything that matters to you privately.

There's a related problem in recommendation systems. Netflix famously found that users' actual watching behavior diverged sharply from their stated preferences — people would add serious documentaries to their queues and watch reality TV instead. The queue was a public-facing artifact, optimized for self-presentation. The watching behavior was private. The algorithm that modeled "you" from one would get the other wrong consistently.

Your public outputs are your queue. Your actual preferences are what you watch at 11pm when no one is looking.

What private context actually changes

The fix isn't giving AI access to more public data. It's giving it access to a categorically different kind of data.

When I write a note that says "I find most productivity frameworks condescending" — not for anyone to read, just to capture a feeling — that's back-stage information. It's not optimized for any audience. It's just true, at least at the moment I wrote it. When I record that I prefer direct feedback over diplomatic hedging, or that I'm currently skeptical of a technology I used to be enthusiastic about, those are preferences that would never surface in my public outputs.

This is what a private knowledge base can do that public data can't: give AI something honest to work from. Not curated. Not managed. Just recorded.

The difference isn't just accuracy. It's that private notes, when you actually write them, tend to capture the things you care about enough to write down but not enough to perform. The texture of real preferences rather than stated ones. The gap between the book you tell people you loved and the one you actually stayed up to finish.

There's a reason Goffman called the back stage a "backstage" and not a "hidden stage." It's not a secret — it's just the part that isn't performed. It's where the actual person is, when they're not working at being someone.

Any AI that only knows your front stage doesn't know you. It knows your resume.

---

Asgeir Albretsen is the founder of Harbor.