A friend recommends a doctor. Before you've left the conversation, you've already asked — not the internet, not yet a website, but the machine. "Is she any good?" You type it into an AI the way you'd once have muttered it to yourself. It answers. Something in you decides. You believe it, or you quietly don't. You haven't opened a single page. You may never open one. The verdict already landed, and no one watched it land — not the doctor, not your friend, not even you.
I do this a dozen times a day now without noticing. So, increasingly, do you. Someone mentions a company, a restaurant, a freelancer, a homestay in the hills, and before the sentence is finished a machine has been asked to vouch for them. Not to find them — we've just been told they exist. To tell us whether to believe it.
The eye that opens after first contact
There are two moments, and we only ever talk about the first.
The first is the glance — the eye barely open: discovery, how someone comes across you at all. Almost everyone pours their energy here: rank higher, get listed, buy the ad, be findable.
The second is the look that opens the instant after first contact, when they turn to decide whether you're real and whether you're worth it. Call it the second eye — the one that opens after the glance and reads you properly. The first search gets you noticed. The second gets you chosen. Same moment, two names for it — the second search, the second eye.
And the eye has started asking a different question than it used to. Not "does this person exist" — being told about them settled that. Something harder: who are you, really, and is that worth my money, my time, my trust? Until now, a person asked it — silently, on your behalf. Increasingly, the person doesn't ask it. A machine does, and hands them the answer.
So I asked the machine about my own work
Being an engineer, I did the obvious thing: I asked it about me.
I opened the AI I use every day to build software, and I pointed it at things I'd made — our own homestay, a handful of small operations I know personally up in the hills, run by people whose food and views and years of work I can vouch for with my own eyes. Then I asked it the second eye's question about each one. Is this place any good? Who runs it? Would you send someone there?
The answers split cleanly, and the split has stayed with me since.
Where someone had said who they were — clearly, specifically, in their own words — the machine was a startlingly good spokesperson. Sometimes better than I'd have managed. It had read the whole thing and said it back, warm and accurate, like a friend who'd actually stayed there.
Where the self-presentation was thin, the eye had nothing to hold. Sometimes it shrugged. Worse, sometimes it invented — it described one place in loving, confident detail that simply wasn't true, a homestay assembled out of plausible guesses. And one genuinely lovely little operation I know well, it couldn't find at all. So for the traveller asking, that place did not exist. The food, the view, the years — invisible, because the owner never even knew the question was being asked.
That is the whole argument, and I found it by accident, asking about my own backyard.
Then I caught myself doing it. Weighing a place I didn't know, I skipped the check I always run — Google it, read the site — and asked the machine if it was any good, half-believing before I'd looked. It was confidently wrong about one thing, and I nearly didn't catch it — the deciding I'd always done by hand passing to the machine.
What's actually happening under the hood
I build with these tools all day, so let me say plainly what the machine is doing, because it isn't magic and it isn't an oracle.
It reads back whatever words about you it can find and trust — yours first. When your own page says something specific and true, the model has something solid to stand on; it grounds its answer in your words and repeats them. When your page says nothing — "premium quality, best-in-class service, customer-first," the mush that could describe anyone — it has nothing to ground on. And a language model asked a question it can't ground will not go quiet. It fills the silence. Confidently. That confident fill has a name: a hallucination. The invented homestay wasn't the machine malfunctioning. It was the machine doing exactly what it does when you've given it nothing real to say.
So the thing that protects you is almost embarrassingly simple: specific, first-party, honest text. The one true detail. The sentence only you would write. Not because it reads well to humans — though it does — but because it's the only thing the machine can safely repeat.
What an AI can and can't repeat about you
This is where the old marketing advice quietly inverts.
For years the counsel was: be there. Have the listing, the profile, the page. But presence answers none of the eye's question, because the eye doesn't reward existence — it reads, and now it reads on someone else's behalf. It cross-references, too: it holds your words and everyone else's words about you side by side, and the gap between them is what it's measuring. Four in five Indian shoppers already check reviews before they buy; the machine checks all of them at once, in a second, and reports the verdict as fact. You can't out-adjective it. You can only give it something true enough to carry.
What this actually asks of you
So the lesson isn't the tired "build a website." Plenty of people own a page that says nothing, and it protects them from nothing.
It's sharper, and a little stranger, than that. You're no longer writing for a reader who will weigh your words and make up their own mind. You're writing the source text a machine will quote about you to someone you'll never meet — and it can only quote what's actually on the page. So the job has changed: not copy to be admired, but sentences a machine can repeat without inventing — the specific, first-party, checkable ones only you would write. Give it a true detail and it stands on yours; give it adjectives and it stands on a guess. There's a colder version, counted in money rather than trust — when a place can't say who it is, people reach it through a platform instead, quietly costing it real income: the rupee cost of not owning your channel. Your page stopped being a brochure the moment the thing reading it became the thing answering for you. And more and more, when the eye is a machine, it literally is the thing that answers — whether or not you gave it the truth to work with.
The verdict you never see
The first search is loud — ads, algorithms, going viral. The second is silent, and now it's silent in a new way. It happens in a question typed into an AI at midnight, a machine reading you on a stranger's behalf and handing back the sentence that decides them, in a conversation you'll never know took place.
You are never in the room for it. You never see the verdict, or the invention that stood in for it when you'd left the machine nothing true to say. The only thing you get to decide — long before, on the one surface you own — is how well you've already said who you are.
Now — what do you see?
- When did you last let a machine vouch for someone before you trusted them — and did you check its answer, or just believe it?
- Look yourself up the way a stranger would, then ask an AI to describe you. Would you recognise the answer — and could it find you at all?
- What's on your page that only you could have written, and what's just filler an AI could never safely repeat?