Fix Your Raw HTML or Vanish from AI Search

TL;DR: A study of 274 top fintech homepages found that 36% fail to deliver their content in a raw HTTP fetch — the only read most AI crawlers make. Forty-seven of those sites return zero readable content to GPTBot, ClaudeBot, and PerplexityBot. For operators running paid acquisition into brand-heavy, research-driven funnels, this structural gap is a direct threat to top-of-funnel AI visibility.

The Measurement That Changed a Design Argument

Until May 2026, “rendering independence” was a principle you could argue about. You could read the spec, look at crawler behavior patterns, and still walk out of the engineering review unconvinced that it was a real production problem. Now there is a number: 36%.

Researcher Slobodan Manic ran two sequential measurements against 274 fintech homepages drawn from the CNBC World’s Top Fintech Companies 2025 list. The first was a raw HTTP fetch with no JavaScript execution. The second was a full browser render using Playwright and Chromium, captured at five seconds post-TTFB and again at network idle. The gap between those two readings is exactly the gap an AI agent has to close on its own — and most of them don’t close it, because they can’t.

Of the 274 homepages, 99 returned less than 80% of their final content before any JavaScript ran. Inside that group, 55 sites returned less than 30%, and 47 returned zero. A blank shell. Major exchanges, neobanks, public companies — brands a person in finance would recognize without prompting — invisible to the crawlers feeding the AI systems your prospective customers are already using for research.

Why AI Crawlers Don’t Render JavaScript

GPTBot, ClaudeBot, PerplexityBot — these are not browsers. They make HTTP fetches. They walk away with whatever bytes come back in the raw response. Running a full Chromium instance per page at the scale these systems crawl the web is a compute cost that multiplies into the millions fast. So they don’t do it by default. They take the raw HTML and move on.

Google’s crawler runs a deferred rendering pipeline for some pages. Some AI systems will render selectively when the raw response looks empty. But those are exceptions. The production default for the crawlers feeding the largest AI systems today is: raw HTTP fetch, no JavaScript, done.

This creates a visibility gap that real users never see. A human visitor opens your site in Chrome. JavaScript runs. The page assembles. The hero loads, the trust badges appear, the value proposition renders. The visitor sees what you built. The AI agent fetched the raw response before any of that happened. Whatever wasn’t in the first HTML response is, for that agent, not there.

For operators spending $10K+ per month on paid media management, that means you’re buying traffic into a brand that AI systems may not be able to accurately describe or recommend — because your homepage returned empty bytes when the crawler came calling.

The Recovery Curve Proves This Is Fixable

Here is the part that matters operationally: 273 of the 274 sites in the study reached 80%-plus content visibility once a real browser rendered the page for five seconds. The content exists. The sites are not broken. They are gated behind a JavaScript runtime that production AI crawlers don’t pay to run.

Stripe, Adyen, Plaid, Marqeta, Remitly, Starling Bank — 101 sites in the sample returned 100% of their homepage content in the raw HTTP fetch. Fiserv returned a complete homepage in 58 milliseconds. Acorns in 76ms. Ledger in 100ms. These are not companies running legacy server-rendered PHP. They are using modern stacks, CDNs, and content management systems. They made architectural decisions that put content in the raw HTTP response, and they didn’t let framework preferences override that requirement.

The pushback that server-side rendering means going backwards five years in engineering is the wrong shape of objection. Next.js has SSR and static generation built in. Astro and SvelteKit ship server-rendered by default. React applications can use Prerender.io or Cloudflare Pages’ prerendering layer to serve a rendered snapshot to crawlers without changing the runtime experience for users. Most teams don’t need to rebuild. They need to add a server-rendering layer to a specific set of routes: homepage, pricing, product pages, anything carrying trust signals or regulatory disclosures.

What This Means for High-CAC Verticals

Fintech is where this study was run, but the structural failure it exposes is not unique to fintech. It applies to any vertical where buying decisions are research-heavy and the comparison loop has migrated into AI surfaces.

Consider forex broker acquisition. A prospective trader evaluating brokers is running exactly the kind of multi-round comparison that now happens inside AI chat before a single click-through occurs. Eric van Buskirk’s clickstream study of 846,000 Google sessions found that AI Mode users close their loops inside the AI 64% of the time, never clicking through. If the broker’s homepage returns zero content to GPTBot, the broker is absent before that comparison begins.

The same dynamic applies to iGaming operator marketing, where the trust layer — licensing badges, responsible gambling disclosures, payment method logos — is precisely the content most likely to be rendered client-side because someone wanted it to feel dynamic. That is the content an AI agent never sees when JavaScript is off. It is also the content a prospective depositor needs to see before they fund an account.

For crypto exchange acquisition, the problem is structurally identical. Exchange comparisons — fees, security certifications, supported assets, jurisdictional availability — are research-heavy decisions that users increasingly delegate to AI. If the exchange’s homepage carries those signals only in JavaScript-rendered components, the AI agent assembling the candidate set works from a shell.

Legal operators face the same exposure. A personal injury firm or mass tort practice spending aggressively on brand runs detailed regulatory and credentialing copy on its homepage — bar admissions, case results, AV ratings. If that content is client-side rendered, law firm visibility in AI search degrades exactly the way fintech visibility degrades: before the comparison even starts.

Run the 30-Second Audit Right Now

Open Chrome. Open DevTools. Hit Cmd+Shift+P on Mac or Ctrl+Shift+P on Windows. Type “Disable JavaScript” and hit enter. Reload your homepage.

What loads is what the agent saw. If your hero, value proposition, product description, trust signals, CTAs, and regulatory disclosures are all visible, your homepage passes the rendering-independence requirement. If the hero is there but the body is gone, you’re in the partial-visibility band. If the page is blank, you’re in the same tier as the 47 zero-content sites in the fintech study.

This is the cheapest visibility audit in the AI search category. It takes 30 seconds, requires no paid tools, and produces a result binary enough that there is no methodology argument. Either the agent can read your homepage, or it cannot.

Interior pages matter too. The fintech study measured homepages only, from one geographic origin, on a single day. A site with a clean homepage and JavaScript-heavy product or pricing pages would pass the test and still have the visibility gap on the routes that actually drive conversion. A complete audit covers the full set of high-intent pages: homepage, pricing, product pages, blog index, landing pages carrying trust signals. That is the scope that determines whether your brand enters the candidate set on the queries that move money.

The audience targeting infrastructure you’ve built for paid acquisition assumes your brand is legible to the systems your prospects already use for pre-click research. If the raw HTML version of your homepage is a shell, that assumption is broken at the foundation. Schema markup doesn’t help when the agent can’t read the page. Citation strategy doesn’t help when the model never saw the content. Everything stacks on the structural layer, and for one in three of the top fintech companies in the world, that layer is currently broken.

The fix paths are known. The audit is free. The 36% that got it wrong includes brands that passed through several rounds of senior engineering review without anyone naming AI visibility as a constraint. That review needs to happen now, before the AI surface your next depositor, trader, or claimant uses to compare providers assembles its candidate set without you in it.

If you’re running AI-assisted lead qualification on top of your current stack, the same rendering-independence logic applies to the pages those agents surface during qualification flows. The structural floor has to be solid before anything else you build on top of it works.

Originally reported by Search Engine Journal, June 2026.