The premise of this research was simple: if AI agents are becoming a way people discover local businesses, someone should measure who the agents can actually see. So we did it properly — not "ask ChatGPT and screenshot the answer," but building a verified master list of every business in a vertical (cross-referenced against the Apple Maps POI layer), then sweeping four engine families with the queries a real customer would use, and checking which of those real businesses ever appeared.
The headline: pooled across our six exhaustive verticals, 321 of 537 businesses (60%) were invisible to every AI engine we tested — no mention, no citation, nothing, across every query and every model tier. If anything that's an undercount: our name-matching is deliberately generous, so a business named after its category ("Sushi Takeaway") can be credited as seen when the answer merely used those words. Each query ran once per tier — a zero across four or five queries and every model is meaningful, though it isn't infinite sampling. And a business missing from the Maps layer entirely never made our master list in the first place.
| Vertical | Businesses | Invisible to every engine |
|---|---|---|
| Japanese restaurants | 122 | 79 |
| Sushi | 111 | 66 |
| Indian restaurants | 108 | 74 |
| Barbers | 104 | 61 |
| Italian restaurants | 73 | 39 |
| Board-game shops | 19 | 2 |
Two readings of that table matter. First, the wall is consistent: every food and service vertical loses half or more of its real businesses. Second, the one exception is instructive — board-game shops are a tiny, well-documented market (19 shops, a passionate community writing about them), and almost all of them surface. The wall isn't about AI disliking small business. It's about how much of the web talks about you.
The long tail exists only on the map
Most of the invisible businesses do exist somewhere an agent could theoretically reach: they're in the Apple Maps layer, which is why our master lists could include them. In the restaurant verticals, Apple Maps (the layer Siri draws on) actually surfaced more individual venues than any language-model engine — 25 to 30 per vertical in our runs. But being a pin on a map is not the same as being recommendable: a pin gets you found when someone searches your name, not chosen when someone asks "where should we eat Friday?"
And before you assume the invisible 60% are blocking AI crawlers or have broken websites: they aren't, and mostly they don't. Only 28 of all 679 probed businesses deny AI crawlers in robots.txt — we wrote that up separately in Nobody is blocking the agents. The wall is an authority problem, not an access problem: nobody independent writes about these businesses, so engines that assemble answers from written sources have nothing to assemble from.
Check which side of the wall you're on. Ask an AI assistant the question your customers would ask — not your name, the need ("best Japanese in Auckland for a date"). If you never come up, you're in the 60%.
Then fix authority before aesthetics. The invisible businesses aren't invisible because their sites are ugly — they're invisible because no list, review platform, or community thread mentions them. Getting written about is the way through the wall. Where those citations live per vertical is the subject of Every vertical has a kingmaker.