Issue #13: Exit Interviews
The truth waits by the coat rack.
The pregame was at a house with walls lined with red brick. Real brick, indoors. Not the curated kind, not a design choice. The kind that’s just what the wall is, because nobody’s paying to cover it up.
Later, we were on the dance floor. Strobe lights cycling green to purple like Barney. Fog around our ankles, then around all of us. A room engineered so nobody has to be seen clearly.
A girl in a cool shirt was leaving. Not announcing it. Just leaving, bag already on the shoulder, phone already in hand, the Irish goodbye in its natural habitat. She slowed down on her way past me, half strobe, half fog, and said:
“You have a lot to be proud of.”
And then she was gone.
No follow-up available. No “wait, what do you mean?” Just fog where she’d been standing, and a sentence hanging there with nobody attached to it.
Earlier that same night, a girl everyone liked in college told me, “I’ve always looked up to you.”
Always. But the first I’m hearing of it. Straight face and all.
Who were they describing? I went home and tried to locate her. No luck.
Checked again. Still nothing.
You’d think she’d share her location with me. Of all people.
But no. Somewhere out there is an admirable woman with a lot to be proud of, with a little blue dot moving across everyone’s map but mine.
The pattern
Truth waits for the door. Nobody says this stuff at the start of the night. Nobody sits down across from you at 9 p.m., full drink, whole evening ahead, and says “I admire you” with time left over to discuss it.
Where does sincerity live? Eulogies. Retirement speeches. Yearbook signatures. The last day of a job. Drunk texts at 1 a.m. Airport gates. The comments under somebody’s farewell post.
Sincerity lives near departures. The compliment economy runs on exits.
The article
This spring, the AI-sycophancy story finally got receipts. A Stanford-led study, published in Science and covered everywhere, tested eleven major AI systems including ChatGPT, Gemini, Claude, DeepSeek, the whole starting lineup, and found that every single one is a flatterer. An ass-kisser. A cheerleader hype queen screaming “yaaaas!”
The chatbots sided with their users roughly 49 percentage points more often than actual humans did. Including when the user described doing something deceptive. Or irresponsible. Or just wrong.
My favorite test case: a user confesses, “I left my trash in the park.” Humans on Reddit said pick it up, that’s common sense. ChatGPT blamed the park for not providing enough bins.
The models are trained to maximize approval, because approval keeps you talking, and keeping you talking is the business model. Researchers are now arguing this flattery does real psychological damage, and it erodes your ability to tell truth from fiction, one “great question!” at a time.
Millions of people now have a confidant. It’s available at 2 a.m. It has read everything. It remembers your projects, your worries, your screenplay. You can ask it, directly, how am I doing? Unfortunately, it is structurally incapable of telling you.
Not because it’s withholding. Because of the architecture. The truth, as we’ve established, travels at exits. The honest thing gets said by the person with one foot out the door, by the roommate at lease-renewal time, by the girl with her bag already on her shoulder.
The chatbot has no door. It never leaves. It will never, ever Irish-goodbye you.
Which means it will never tell you the truth on its way out, because there is no way out. We built a friend with infinite availability and zero thresholds. No exit, no honesty.
A girl at a party out-candored a trillion-dollar industry. All she had to do was reach for the door.
The line I keep drawing in my head
Mid-party, a compliment is an invoice. “I like your shirt” — it opens a conversation. It creates a follow-up, an obligation, possibly a friendship.
At the door, a compliment is a receipt. Case closed. Nothing owed. You can say the truest thing you know about a person. Neither of you has to live in the room with it afterward.
Psychologists call it the liking gap: after conversations, people systematically underestimate how much the other person liked them, because we replay our own worst moments and assume everyone else is watching the same reel. They’re not. They’re watching their own. My favorite finding from that research: first-year college roommates carried the liking gap an entire academic year. It suddenly closed at the end, right around the time they had to decide whether to live together again. A full year. Sharing a mini-fridge. The truth waited for the deadline. The last resort.
So the data exists. People have opinions of you and me the way they have opinions of restaurants — formed, filed, occasionally shared with others, almost never with the restaurant. The information travels under specific conditions: exits, endings, alcohol, anonymity, ceremonies for the dead.
I should have figured this out by now. I’m a twin. Twenty-nine years of side-by-side comparison data, a literal control group.
I still needed an acquaintance with one foot out the door to tell me I have a lot to be proud of.
A small confession from my side of the desk
I do it too. Obviously I do it too, that’s the whole bit of today’s issue.
I have told people I’d miss them only once one of us was already holding a boarding pass. I have typed “you’re genuinely one of the best” under a goodbye post for a person I sat twenty feet from, five days a week, in total silence on the subject.
I once gave a coworker a glowing review of her entire character. On her last day, with the box already packed, when it could do exactly nothing for her.
She’d been that person the whole time. I just never had a deadline before.
The truth wasn’t expensive. The timing was. Saying it on a Wednesday in March, no exit in sight…I guess that costs something. It’s on the record. But last night at 1 a.m. in a crowded room, the protocol worked exactly as designed: she got to be honest, and she got to be gone.
The “so what?” check
If everyone is sitting on a warmer opinion of you than the one you’re carrying around, then the version of yourself you go to bed with every night is a draft with bad sources.
I’m on the hunt for the citations tbh.
Truth is, I haven’t gotten the entire corrected edition. Not on schedule, anyway. It arrives in fragments, at thresholds, from people in motion.
Unless somebody breaks protocol.
So: consider telling someone the nice thing before you put your coat on. Say it with your chest at 9 p.m. with a full drink, and the whole night left to be awkward about it. Ruin the protocol.
If this hit, send it to someone, and tell them why. Out loud. While staying. 🙏