Shopify, Wix, Webflow, Framer: what your platform actually lets you do for AI

Your platform controls part of what AI assistants can read on your site, because it owns the root of your domain. During 2026, Shopify, Wix, Webflow and Framer all shipped a way to publish an llms.txt file, a restriction that still held a year earlier. The agent layer, however, remains uneven from one platform to the next, and that is now where the real gap sits.

Why does your platform change what you are able to do?

Some of the signals AI systems read do not live in your pages. They live in files served from the root of your domain: robots.txt, llms.txt, and the .well-known folder that holds the files aimed at AI agents. On a custom-coded site or on WordPress, you own that root and you put whatever you want there. On a hosted website builder, the root belongs to the vendor, and you can only serve what the vendor decided to let you serve.

This is a technical constraint, not a mistake on your part. A business owner who picked Shopify in 2022 did not make an AI visibility error, because the question did not exist yet. What matters is what that platform allows today.

What changed across the four major platforms in 2026?

The four leading website builders each shipped, at their own pace, the ability to serve an llms.txt file at the domain root. That file gives AI systems a condensed, structured view of what you offer. What belongs in it, and why it matters commercially, is covered in our article on llms.txt as the business card AI systems read.

• Shopify now serves /llms.txt, /llms-full.txt and /agents.md natively on every store, with default generated content. Since May 28, 2026, all three paths can be overridden with Liquid templates added to the theme.
• Wix automatically generates and maintains your site's llms.txt file, editable from the SEO dashboard, with a toggle to opt out. It requires a paid site, a connected custom domain and indexing enabled.
• Webflow offers an llms.txt upload in site settings, served on the custom domain, UTF-8 encoded and under 100 KB. It is not published to the Webflow staging domain.
• Framer hosts static files at the root or inside a .well-known folder from its hosting settings, on the Pro and Enterprise plans.

The practical lesson is short. If you gave up on llms.txt because a 2025 article told you your platform made it impossible, that advice has expired. Check before you conclude.

What is still out of reach today?

The opening applies to llms.txt, not to everything else. Per-page Markdown variants, which the llms.txt specification proposes as .md versions of each URL, are not supported by Webflow, which states plainly that its implementation is deliberately scoped to the llms.txt file itself. Shopify, for its part, serves three paths that Shopify defines, and nothing more: agent files remain out of reach for a store, and Shopify documents that robots.txt can only be edited through a Liquid template its support team does not cover.

The .well-known folder, which holds an MCP server capability card, an agent skills index and authentication metadata, follows a pricing-tier logic. Webflow reserves it for its Enterprise API, with a documented ceiling of 30 files under 100 KB each. Framer opens it from the Pro plan upward. The agent layer is therefore not locked by technology alone. It is locked by your subscription.

How should an honest scan treat a signal your platform forbids?

A signal your platform does not let you install should be marked impossible rather than counted against your site. It drops out of the action plan, and it never turns into advice you could not act on. Faulting a Shopify store for not exposing an agent capability card would be like faulting it for the weather, and worse, it would mean selling work that cannot be delivered. That principle is set out on our methodology page.

Multi-platform setups deserve a note, because they are common. Plenty of companies run a WordPress site alongside a Shopify store on the same domain or on two subdomains. In that case a signal counts as impossible only if it is impossible on every platform involved: a WordPress site can happily serve the llms.txt that the store cannot. The constraint is computed on the real estate you actually run, not on your most restrictive platform.

Why should an unmeasured signal never read as a failure?

An unreachable file, a timeout or a server error produce an absence of measurement, not a measured absence. Those signals drop out of the calculation instead of scoring zero, and every report states how many of the 19 signals were actually verified. An unmeasured signal also cannot trigger the harshest verdict: a site is only declared invisible on a failure that was genuinely observed, never on a doubt. Where that line falls is developed in our article on what an AI visibility scan actually measures.

This distinction has a direct commercial consequence. A tool that scores zero for everything it failed to read produces low, dramatic, wrong numbers. It sells fear instead of selling a measurement.

When is your platform not the problem?

Most of the time, the platform is not the main cause of invisibility. Two more ordinary problems come first. The first is having nothing worth citing: an AI will not recommend a company whose site is five pages of slogans, however clean its technical files are. The second is having no third-party presence: if nobody mentions you elsewhere on the web, no root file will manufacture that.

A Shopify store with real product pages, genuine answers to the questions buyers ask, and coverage in trade press will get cited ahead of a custom-built site that is technically flawless and empty. In that situation your budget belongs in content, not in a platform migration. Migrating makes sense once the content already exists and the platform becomes the limiting factor, not before.

What should you check on your own site this week?

Three questions place your site well enough. Does your platform serve an llms.txt file, and does it describe your business or recite a generic template supplied by the vendor? Is your content readable directly in the page source, or does it require JavaScript to appear? Does your robots.txt let in the crawlers used by answer engines, the ones that fetch a page to answer a question being asked right now?

None of those three checks guarantees a citation. They establish eligibility: without them a citation is very unlikely, with them it becomes possible. That is the only honest promise anyone can make in this market today.