We checked 368 small business sites. 1 in 4 are partly invisible to AI crawlers.

What we measured (and what we did not)

For a decade, "JavaScript SEO" meant one worry: can Google see content that only appears after scripts run? That worry is largely settled. Google renders. What changed is the arrival of a new class of crawler. As documented by Vercel and others, the major AI crawlers, OpenAI's GPTBot, Anthropic's ClaudeBot, and PerplexityBot, fetch your pages but do not execute JavaScript. Google's renderer does. That gap is the subject of this study.

We measure first-party crawl visibility to non-rendering AI crawlers: what an AI company's own crawler retrieves when it fetches your page directly without running JavaScript. This is not a claim that "ChatGPT cannot know your business exists." AI assistants also draw on search indexes, licensed feeds, and third-party citations beyond their own crawlers. Every number here is scoped to the direct-crawl path, a path we believe matters and will matter more.

The counterintuitive finding

The platform you build on is the strongest predictor in our data, and it cuts against conventional SEO advice. The usual warning is that hosted builders like Wix and Squarespace are bad for search while a "real" CMS or a custom build keeps you safe. For AI-crawler visibility, the opposite is true.

The issue is not builder versus custom. The issue is whether the platform produces readable HTML before JavaScript runs. Wix and Webflow server-render or static-export answer-critical content by default, so they were the safest by a wide margin. Bespoke custom builds and JavaScript single-page apps, the sites whose owners most often assume "full control" equals safety, were the worst.

Findings by platform

Share of sites on each platform with at least one answer-critical element invisible to a non-rendering crawler.

Read this as a proxy for how each platform renders by default, not a verdict on the technology itself. The safe platforms server-render content into the HTML; the risky ones lean on client-side JavaScript. The same tool stack can land at either end: a site built to server-render is the best case of all, regardless of platform (we prove this with our own below).

What goes missing

Across all sites, these are the elements most often present for Google but absent for a non-rendering crawler:

Methodology

We drew a purposive sample of real North American small-business websites across five verticals (professional services, health and wellness, home services, legal, and local services). For each site, on a single pass, we captured it two ways:

• Rendered DOM (Google-equivalent). Headless Chromium with full JavaScript execution, the way Googlebot's renderer sees a page.
• Raw HTML (AI-crawler-equivalent). A single HTTP GET with a GPTBot user-agent and no JavaScript execution, consistent with how GPTBot, ClaudeBot, and PerplexityBot retrieve a page on a direct crawl.

An element is counted AI-invisible when it is present in the rendered DOM but absent (or reduced below a set threshold) in the raw HTML. We attempted 405 sites and obtained analyzable dual captures for 368 (a 90.9% response rate). The 37 excluded (9.1%) failed on a connection error, TLS error, or timeout, the signature of dead domains, bot-protected sites, or servers too slow to respond. Exclusions are reported, not imputed.

This is a purposive sample, not a random one. It skews toward small B2B and local-service businesses and is not representative of the web as a whole. Findings describe this population. The raw fetch is a documented approximation of the AI-crawler view, not a capture of the crawlers themselves, and we measure crawl-time visibility, not whether an AI ultimately cites a business.

Why it matters

Search is no longer the only way a business gets discovered. Answer engines and AI assistants increasingly summarize, recommend, and route customers, and they lean on what their own crawlers can read directly. A site whose phone number, services, or location only appear after JavaScript runs is handing those systems a blank page.

The fix is one principle: if a piece of information matters for being found, it has to be in the HTML the server sends, not injected by JavaScript after the page loads. Google will run your JavaScript and see the injected version. A non-rendering AI crawler will not. The simplest self-check: open your page, view source (not the inspector, the actual source), and search it for your phone number, your H1, and your menu links. If they are not there, they are JavaScript-injected, and a direct AI-crawler fetch comes back without them.

Open dataset

The full per-site dataset, detected platform, vertical, raw and rendered word and link counts, and every per-element invisibility flag, is released under CC BY 4.0. It is anonymized at row level (domains removed). Journalists and researchers may reproduce or cut their own angles with attribution.