GEO for Wix: Six Sites, Six llms.txt Files, Zero AI Crawler Rules

Wix is the only major website platform where the machine-readable file ships by default. We fetched six Wix-built sites on October 12, 2026 the same way a text-only AI crawler reads them — plain HTTP, no JavaScript, no browser — and every single one returned a working llms.txt file. Zero of the six robots.txt files names a single AI crawler. And the files the platform generates for free advertise something no hand-built file in our earlier tests had: a live MCP endpoint an agent can query directly.

Six Wix sites, six llms.txt files — platform defaults beat opt-in effort

Site

llms.txt

robots.txt (AI crawlers)

Homepage for a text crawler

Structured data

wix.com

200 — 36.7 KB curated file

2.8 KB — none named

200 — 1.29 MB

Organization, 2× FAQPage (8 Q&A), HowTo, VideoObject

Paul Hollywood

200 — 3.2 KB auto-generated

495 B — none named

200 — 541 KB

WebSite only

Eterna

200 — 3.7 KB auto-generated (French)

488 B — none named

200 — 461 KB

WebSite, LocalBusiness, PostalAddress

Izzy Wheels

200 — 3.2 KB with MCP

492 B — none named

200 — 988 KB

WebSite, LocalBusiness, PostalAddress

Black Sheep Bikes

200 — 3.3 KB with MCP

497 B — none named

200 — 744 KB

WebSite, LocalBusiness, PostalAddress

Hair Comes the Bride

200 — 3.2 KB with MCP

499 B — none named

200 — 1.29 MB

WebSite only

Six for six. Our travel sector test found three llms.txt files among eight booking sites, and our hospitality test found one among eight hotel and restaurant brands. Wix hits six for six because the platform generates the file. Wix's help center article "Understanding Your Site's llms.txt File" says the company automatically generates and maintains llms.txt, and site owners can view, edit, or opt out from the "SEO & GEO" section of the Wix dashboard. What costs other industries an afternoon of manual work is a default checkbox here.

The gap sits right behind that default. Every auto-generated file is roughly 3.2 KB of boilerplate: one brand paragraph, a homepage link, and an "AI Agent Access" section. wix.com's own 36.7 KB file shows what a curated version looks like — named sections and links for its AI features, pricing plans, templates, and the Symphony agent platform. The platform that sells AI readiness invested in its own file. The businesses that inherit the default did not.

Every Wix customer's llms.txt advertises a live MCP endpoint

The auto-generated files carry a section none of the hand-built llms.txt files in our travel or hospitality tests had: a Model Context Protocol endpoint. Black Sheep, Izzy Wheels, Hair Comes the Bride, Paul Hollywood, and Eterna all point to /_api/mcp and list two tools — GetBusinessDetails for contact, location, and hours, plus SearchInSite for site search. The file's own notes say no authentication is required to connect and content stays live.

We poked the endpoint with a plain HTTP request. It answered 401 — a raw GET is not an MCP session, so the server asked for the handshake. The plumbing exists and a machine can find it: an agent that implements MCP can pull live business details straight from a Wix site without scraping anything. That is the most agent-first setup we have measured on any platform, and it costs a Wix owner nothing.

Odd footnote: wix.com's own curated file skips the MCP section entirely. The platform wrote a richer marketing file for itself and left its own endpoint out of it.

Zero AI crawler policy and pages that weigh a megabyte

The robots.txt files tell the other half of the story. All six are stubs — a sitemap line, generic User-agent rules, 488 to 499 bytes each. Not one names GPTBot, ClaudeBot, PerplexityBot, anthropic-ai, or Google-Extended. Wix solved llms.txt at the platform level and left crawler policy to each site owner, who never writes it. Airbnb in our travel test wrote 248 Disallow lines for GPTBot alone; the entire Wix group we tested wrote zero.

The pages those files describe are heavy. Homepage HTML: 1.29 MB for wix.com and Hair Comes the Bride, 988 KB for Izzy Wheels, 744 KB for Black Sheep, 541 KB for Paul Hollywood, 461 KB for Eterna. A model's crawler pays the bandwidth and the parse cost on every one of them.

Schema stops at identity. All six sites carry WebSite; three add LocalBusiness plus PostalAddress. None carries Product, Offer, or Review — Eterna sells Swiss watches, Black Sheep sells custom bikes, Izzy Wheels sells wheelchair wheel covers, and none of their pages proves the product exists to a machine. wix.com is the only site with FAQPage or Organization markup. The platform's own file and schema are first-class citizens. The businesses that build on it get the default and nothing more.

What to copy from the platform that got one half right

Real talk: Wix did the hard work so its users do not have to, and six-for-six llms.txt beats every industry we have tested. But the default file is a summary, not a strategy, and the rest of the machine layer is untouched. If you run a Wix site:

1. Edit the auto-generated file. Dashboard → SEO & GEO → llms.txt → edit. Once you edit it, automatic updates stop, so write the version you want to keep: product pages, the help center, the five pages support already routes people to. Validate it with the llms.txt checker and skim our llms.txt guide first.
2. Add Product + Offer schema to every page that sells something. Eterna and Izzy Wheels already carry LocalBusiness, so the leap is small; the schema markup for AI citations guide shows the block shapes. Wix lets you drop custom code or pick an app to do it.
3. Write one AI policy line in robots.txt. Do you want GPTBot and ClaudeBot reading your site? Say so out loud in the file. Wix will not decide for you.
4. Test from a server, not a browser. The AI crawler check replays your pages the way GPTBot does, and the free GEO audit scores all six dimensions with an ordered fix list.

The interesting wrinkle is the piece nobody else has: the MCP endpoint Wix gives every site. Point something at /_api/mcp and read what SearchInSite returns for your own brand — then fix what the file gets wrong, because that is what the models will read.