Field research · Auckland · June–July 2026

Schema markup won't
save you. Being written
about might.

The standard AI-readiness advice is "add structured data." In our Auckland field data, schema.org markup was rare, and having it didn't predict whether AI engines ever surfaced you. What predicted it: whether humans had written about you somewhere an engine trusts.

SCOPE — Auckland, New Zealand · June–July 2026 snapshot. 7 verticals, 679 probed businesses: Italian, Japanese, sushi and Indian restaurants, barbers, plumbers, and board-game shops. Engines queried via their APIs — Gemini with Google Search grounding, OpenAI web_search, Claude with live search, and Apple Maps (MKLocalSearch, the layer Siri uses for local queries). APIs are clean, reproducible instruments and proxies for the consumer products, not the apps themselves. One city, one snapshot — treat this as field data, not a global law.

If you read SEO-industry advice about AI visibility, you'll be told to add schema.org JSON-LD to your site — machine-readable name, hours, menu, reviews. It's reasonable advice and we still recommend it (see the end of this post, honestly). But our field data keeps saying the same uncomfortable thing: markup is not what gets you onto the AI shelf.

  • Across 537 businesses in six exhaustive verticals, only 72 (13%) had usable schema.org structured data on their own site at all.
  • 34 of those 72 schema-equipped businesses were still invisible to every engine. Roughly half did the technically-correct thing and got nothing for it.
  • Meanwhile, consensus winners — businesses every engine agreed on — were often schema-less. The starkest case: King of Cards, a board-game shop whose site is a JavaScript shell a plain crawler reads as nearly empty. Every language-model engine surfaced it anyway, because the board-game community writes about it constantly.

To be precise about what we measured: this is an observed absence of correlation in one city's snapshot — with only 72 schema-equipped sites to compare, the statistical power is limited, and we measured presence in answers, not ranking position or bookings, where markup may matter more. It is not proof that schema does nothing. But the direction is consistent across all seven runs, and the mechanism is visible in the citations the engines actually produced.

What the engines actually cite

When an engine names a business, it cites where it learned about it. Pooling citations across our runs, the top sources are not business websites — they're places where someone else wrote about the business:

SourceTypeCitationsRuns seen in
The Urban ListEditorial1065 of 7
RedditCommunity917 of 7
Heart of the CityEditorial / local756 of 7
DenizenEditorial725 of 7

Reddit appearing in every single run is worth sitting with. A four-year-old thread titled "best sushi in Auckland?" is doing more work for the businesses named in it than any amount of on-site markup. The engines are, in a very real sense, outsourcing their taste to whoever already wrote things down.

What to do about it

Keep your structured data — it's cheap and it helps agents read you correctly once they've found you. Wrong hours or an unreadable menu still hurt you at the point of decision.

But spend your real effort on being written about. Local editorial lists, the review platforms your vertical actually uses, and the community threads where your customers ask for recommendations. One genuine mention in a source the engines trust outperformed perfect markup on an unmentioned site in every vertical we measured.

Want to know what the engines cite about you?

A shelf check shows you — verbatim — what each engine says, and which sources it learned it from.

Request a shelf check →