One wedding website in 140 blocks AI crawlers. 108 give them nothing to read.

The industry is guarding the wrong door

Every major site builder now ships a switch for blocking AI, and a good deal of wedding-business advice this year has been about whether to flip it. In this sample the question is already settled. 139 of 140 sites let GPTBot, ClaudeBot, PerplexityBot, Google-Extended and CCBot through without restriction. The one exception names four of them outright — GPTBot, ClaudeBot, Google-Extended and CCBot.

The door is open almost everywhere. The problem is what is on the other side of it.

77% publish nothing a machine can turn into a fact

Schema.org JSON-LD is the part of a page that states, without prose, what kind of business this is and where it operates. We sorted every site into three groups.

46 sites — no structured data at all

No JSON-LD block anywhere on the homepage.

62 sites — structured data that says nothing about the business

The largest group, and the one a simpler audit would miss.

32 sites — a business type with an address, phone, or service area

23% of the sample. Mostly LocalBusiness, sometimes Organization.

The middle group is the finding. Those 62 sites are not missing markup — they are carrying their platform’s markup. WebSite on 56, BreadcrumbList on 29, WebPage on 27, Organization on 16. That is a theme describing the page it sits on. Of the 62, 32 carry no business-type entity of any kind; the other 30 declare one and then omit the address, the phone number, and the service area.

A checker that asks “does this site have schema?” passes all 62. A model asked which videographers work near Beaufort gets a page title and a breadcrumb trail.

These are not neglected websites

123 of the 140 sites pass every human-facing basic we check: HTTPS, a descriptive title tag, a mobile viewport, and complete Open Graph tags so the link previews correctly when someone pastes it into a group chat. They were built by people who cared how they looked.

93 of those 123 — 76% — still publish no machine-readable business facts. The gap is not between good sites and bad ones. It is between what a designer optimizes and what a parser needs.

Contact details follow the same shape. 41 homepages carry no phone number or email as text in the HTML. The check reads the homepage only, so a number set in an image, or reachable through a contact form a click away, does not count — which is exactly the position a crawler is in. 36 sites have neither a business fact in their markup nor a readable phone or email on the homepage. Asked to name someone, an assistant reading those pages has nothing to copy down.

Signal by signal

All 13 signals, ordered by how badly the sample performs. The one at the top is the one we trust least — see the caveat on llms.txt below before quoting it.

Pass Partial credit Fail

Two failures that cost nothing to fix

42 homepages have no H1 heading at all — not a wrong one, none. On a further 31 sites the page declares several, which comes to the same thing: nothing is marked as the page’s subject. And on 16 sites, neither the word “wedding” nor any location appears in the title, the H1, or the meta description. A reader works out what those businesses do from the photographs. Nothing else can.

By category

DJs came out best on structured data at 6 of 15, and bands worst at 3 of 21. The spread in overall score is narrower than the spread in any single signal.

By region

Four of the five South Carolina wedding markets sit within a point of each other. Columbia & the Midlands sits 7 to 8 points below each of them, on 24 sites — a gap worth noting and too small a sample to explain.

How the scores fall

Median 81 out of 100, mean 78.4, range 27 to 100. 5 sites passed every signal.

What this study does not show

It measures whether a machine can read a site, not what any assistant currently says about the business. Nobody in this sample was queried against ChatGPT, Claude, Perplexity, or Google’s AI results, so nothing here demonstrates that better markup produces more citations. Readability is a precondition, and a precondition is all we tested.

The sample is not random. These are the 140 businesses our editors had already selected into a curated directory, which skews toward the established end of each market. If anything, that makes the result conservative: this is what the better-run half of a state’s wedding industry looks like under the hood.

One signal deserves a caveat of its own. 100 of 140 sites have no llms.txt file. We report that because we measured it, but llms.txt is a proposed convention with no confirmed major consumer, and we weight it at 4 points out of 100 for that reason. Treat the 71% as a fact about adoption, not as a defect.

Method

On September 4, 2026 we fetched each business’s homepage, robots.txt, sitemap.xml, and llms.txt, then evaluated 13 signals against the response. Each signal passes, earns half credit, or fails, and the weighted result is a score out of 100. The sample is every active business in our South Carolina directory with a working website: 47 photographers, 38 venues, 21 bands, 19 videographers, 15 DJs, across 5 regions.

The live check runs 14 signals. The extra one asks whether a business is listed in an editor-reviewed directory — and every site in this sample is listed in ours, so it passed 140 out of 140 and added its full weight to every score. We dropped it and recomputed over the remaining 13. That correction moves the median from 83 down to 81. Every figure on this page is the corrected one.

The checker is the same one we publish free at /ai-visibility, so any site in the sample can be re-run against the identical logic. No business is named here, and no per-site score is published.

Use this data

Every figure on this page is free to republish, quote, chart, or teach from, including commercially, with attribution and a link. No permission needed. If you are writing about it and want a cut we did not publish, ask and we will run it.