What 44 strangers taught us about trusting AI

Before we measured a single brand, we measured something harder: people. Through 2026 we've been stopping strangers on the street — Toronto first, with New York, Seoul, and Busan filming in progress — and asking them, on camera, how much they trust what AI tells them.

The number

Across 44+ unscripted interviews so far, the average sits at 4.9 out of 10. Not rejection. Not faith. Almost exactly the middle — which turns out to be the most interesting place a number can land.

What the middle actually means

A 4.9 isn't ambivalence, and it isn't a coin flip. Listen to the interviews and the single number decomposes into two very different scores: people trust AI highly for facts they can check, and far less for judgments they can't. "What time does the pharmacy close" — sure. "Which lawyer should handle my mother's case" — pause.

The pause is where every business now lives — because the engines answer both kinds of questions with the same confident voice.

Your category almost certainly sits on the judgment side of that line. Nobody's life changes if AI gets a pharmacy's hours wrong. Choosing a lawyer, an orthodontist, a rehab clinic — that's exactly where people hesitate, verify, and look for reasons to believe. Which means the recommendation alone was never going to close the deal. It has to survive the checking.

Three things strangers taught us

1. People verify with AI, not instead of it. The most common behavior we heard wasn't blind trust or refusal — it was triangulation. Ask ChatGPT, then check a review site, then ask a friend. AI has joined the jury; it hasn't replaced it. The practical consequence is brutal for inconsistent brands: a flattering AI answer that doesn't match your reviews, your listings, or what a friend half-remembers fails the triangulation — and the prospect quietly moves on without telling you why.

2. Trust is borrowed, not granted. When people did trust an AI recommendation, they usually explained it by pointing somewhere else: "it's pulling from reviews anyway." The machine's credibility is assembled from the sources it cites — and the public intuitively knows it. That's the whole game for brands: you don't win the model's trust directly. You win the reviews, directories, press, and professional listings it borrows trust from, and the model's confidence in you is the compound interest.

3. Nobody knows why the machine says what it says — and everyone assumes someone does. Several people guessed businesses could "pay to be the answer." They can't — but the suspicion tells you how the public reads this era: whoever shows up is assumed to have earned it or bought it. There is no neutral. Showing up accurately, with a story that survives checking, is the only durable position — because the moment your story doesn't check out, the suspicion has somewhere to land.

What would move a 4.9

Our read, watching the interviews back: the number doesn't rise when the machines get smarter. It rises when the checking gets easier — when the answer and the evidence point the same way. That's not something the engines can do for a brand. It's something a brand does for the engines: one consistent fact pattern, corroborated everywhere a skeptical human might look next.

Why we do this

The Trust Index exists because both halves of the question matter: what AI says about brands, and whether humans believe it. The street interviews are the second half — the human baseline under every visibility score we produce. The 4.9 never appears in any client's score; it's the context those scores live inside.

The first half — what the engines actually say about you today, and whether it survives the checking — is measurable right now. → Get your Trust Index

The research continues — 44+ interviews and counting, one city at a time. Scores and scope are stated on everything we publish, this page included.