A Third Of Fintech Is Invisible To AI Agents

Jun 26, 2026 08:30 PM - 2 hours ago 87

A 3rd of the apical fintech websites successful the world present little than 80% of their homepage contented successful earthy HTML. That is the type of the page an AI supplier gets erstwhile it visits, earlier it decides whether to walk the compute connected a afloat browser render. Most of them do not.

The Structure pillar of Machine-First Architecture says captious accusation must not dangle connected client-side JavaScript. Rendering independence. Until past month, this was a creation principle. Now it is simply a number, and the number is uncomfortable.

On May 25, I measured 274 fintech homepages from the CNBC World’s Top Fintech Companies 2025 list. I made 2 sequential measurements connected each one: a earthy HTTP fetch pinch nary JavaScript execution, and a afloat browser render pinch Playwright. The spread betwixt the 2 readings is the spread an AI supplier has to adjacent connected its own. 36% of these websites unit it to do that activity for the astir important page connected the property. The afloat study is published connected Web Performance Tools.

Most AI Visibility Coverage Skips The Rendering Step

Most AI visibility sum focuses connected schema markup, system content, brand authority signals, and optimization for AI Overviews, ChatGPT search, Perplexity citations, and Gemini’s grounding pipeline. The proposal stacks up fast.

All of that assumes the supplier saw your contented successful the first place.

Most AI crawlers do not render JavaScript. GPTBot, ClaudeBot, PerplexityBot, the AI user-agent landscape feeding the models you are trying to beryllium cited by, they make HTTP fetches and locomotion away. They are not browsers. Running a existent Chromium lawsuit per page costs compute that multiplies crossed the millions of pages these systems want to read. So they don’t, by default. They return what comes backmost successful the earthy HTTP consequence and move on.

There are exceptions. Google’s crawler runs a deferred rendering pipeline for immoderate pages. Some AI systems will render for high-value targets, aliases render selectively erstwhile the earthy consequence looks empty. The shape is not absolute. But the accumulation default for the crawlers that provender the largest AI systems connected the web coming is earthy HTTP fetch, nary JavaScript, return what is there.

This creates a spread that users do not see. A existent visitant opens your website successful a browser. JavaScript runs. The page assembles successful the viewport. Content loads, layout settles, the leader image arrives. The visitant sees what you built. The AI supplier fetches the consequence earlier immoderate of that happens. Whatever does not show up successful the first HTML consequence is, for that agent, not there.

This is what the Structure pillar of Machine-First Architecture is about. Critical accusation must not dangle connected client-side JavaScript. The page must beryllium parseable from the earthy HTTP response, not from the browser-rendered position 5 seconds later. This is not a capacity penchant dressed up arsenic architecture. It is simply a visibility request for the AI agents that now publication the web connected behalf of users.

Until recently, the rendering-independence request was an argument. You could publication the spec, look astatine the crawler behaviors, tie the conclusion, and still disagree astir really overmuch of a problem it is successful practice. There was nary number you could constituent to.

The fintech information gives you the number.

Two Reads On The Same Page: Raw HTML, Then Browser Render

The trial connected each of the 274 fintech homepages was simple: 2 sequential measurements, tally connected May 25, 2026, from Portugal. The first was a earthy HTTP fetch against the canonical homepage, nary JavaScript executed, immoderate bytes came backmost successful the response. The 2nd was a afloat browser render utilizing Playwright 1.60.0 pinch Chromium 148.0.7778.96 successful non-headless mode, capturing the page astatine 5 seconds post-TTFB and again astatine web idle. All measurements ran from Portugal connected May 25, 2026, connected residential broadband, viewport 1280 by 800, nary web throttling.

For each website, contented was extracted from the <main>, <article>, aliases <body> constituent and converted to Markdown to sphere structural elements. The raw-fetch matter was measured arsenic a percent of the network-idle text. If the earthy fetch returned 80% aliases much of what the browser yet rendered, the website passed astatine afloat visibility. Between 60% and 79% was partial. Between 30% and 59% was low. Below 30% was near-zero.

Three sounds connected the aforesaid page, successful the aforesaid session, separated by what the browser had to do to make the page complete: earthy fetch, five-second render, web idle.

The absorbing portion of the curve is not the network-idle reference astatine the end. Almost each website successful the sample resolves to afloat contented by web idle. The absorbing portion is the raw-fetch reference astatine the start, because the earthy fetch is the publication astir AI crawlers really take.

36% Deliver Less Than 80% Of Their Content Without JavaScript

Out of 274 fintech homepages measured, 99 returned little than 80% of their last contented from the earthy HTTP fetch. That is the header number. Thirty-six percent.

Inside that 99, the distribution is steep. Fifty-five websites (20% of the afloat sample) returned little than 30% of their contented without JavaScript. Forty-seven of those websites returned zero. The HTML consequence carried a shell, the layout scaffolding, immoderate inline scripts, nary readable content. Whatever the homepage was meant to pass required a JavaScript runtime to pass it.

The 47 zero-content websites see awesome exchanges, well-known neobanks, ample lending platforms, respective nationalist companies, and brands a personification successful finance would admit without prompting. I americium not going to telephone them retired individually. Naming the websites would distract from the architectural study underneath. Whether your homepage shows up to an AI supplier astatine each is presently a usability of decisions that cipher connected the squad was reasoning astir successful those position erstwhile they were made.

The 24 websites successful the 60-to-79 partial-visibility set person a different problem. They show up to the agent, but not each the way. The supplier gets a leader headline, possibly the superior navigation, possibly a worth proposition. It does not get the merchandise descriptions, the spot signals, the calls to action, the third-party logos. Whatever was decided to render connected the customer broadside is the portion the supplier does not see, and that portion tends to beryllium the portion that was made move because personification wanted it to consciousness interactive.

There is simply a betterment curve, and the betterment curve is wherever the communicative sharpens. Of the 274 websites, 273 scope 80%-plus visibility erstwhile a existent browser renders the page for 5 seconds. Ninety-nine percent. The contented exists. The websites are not broken. They are gated down a runtime that the accumulation AI crawlers do not salary for.

The median website successful the sample takes 21 times longer to scope web idle than to return its earthy HTTP fetch. Thirty-four websites (12%) do not scope web idle astatine each wrong the 30-second cap. That is simply a abstracted problem, but it points to the aforesaid guidelines cause. The costs spread betwixt fetching a website and reference it is widening, and the crawlers cannot support absorbing the difference.

Stripe, Adyen, And Plaid Prove The Stack Is Not The Problem

One 100 and 1 websites successful the sample returned 100% of their homepage contented successful the earthy HTTP fetch. Full visibility earlier immoderate JavaScript ran. The database includes Stripe, Plaid, Adyen, Marqeta, Remitly, Starling Bank, Neo Financial, Backbase, Thought Machine, and 92 others.

Fiserv returned a complete homepage successful 58 milliseconds. Acorns successful 76. Trustly successful 89. Ledger successful 100. Look astatine what those websites really are. Fiserv is simply a payments-and-banking infrastructure institution astatine a $60 cardinal scale. Acorns runs a user app. Ledger is simply a hardware-wallet vendor pinch a merchandise catalog. They are utilizing modern stacks, contented guidance systems, location CDNs, the works. They person decided that the contented the homepage exists to pass will beryllium location successful the earthy response, and they person not fto the model prime override that decision.

This is the reply to the evident pushback. The pushback is that a modern stack requires client-side rendering, that single-page applications are really the web is built now, that asking for server-rendered HTML is asking engineering to measurement backmost 5 years. The fintech sample disproves that connected its own. The websites moving the fastest earthy responses took the rendering-independence request earnestly erstwhile they made architectural choices, aliases rebuilt to return it earnestly aft the fact.

There are exceptions to this publication wrong the sample. Three websites underperformed astatine the five-second render window, moreover though they yet completed. All 3 are Asian companies measured from Portugal, and the latency punishment apt accounts for the curves alternatively than the architecture.

The Web Performance Tools study tested homepages only, from 1 geographic origin, connected 1 day, pinch 1 measurement per website. It did not measurement interior pages. It did not trial aggregate regions. It did not trial contented gated down scroll aliases click. The image this dataset gives you is the homepage of the biggest fintech companies successful the world, connected a azygous time successful May, fetched from Western Europe. That is simply a slice. A useful portion for the load-bearing mobility this article is about, but a slice.

The load-bearing mobility is whether the rendering-independence request holds astatine standard crossed a large, modern, well-resourced commercialized cohort. The fintech sample answers it. Most of the cohort gets it right. A 3rd gets it wrong, and the 3rd that gets it incorrect includes brands large capable that the architectural decisions almost surely passed done respective rounds of elder engineering reappraisal without anyone naming AI visibility arsenic a constraint.

Fintech Is Where The Homepage Is The Trust Signal

Most categories tin spend immoderate of their homepage to beryllium invisible. A user SaaS institution tin suffer a leader subhead, and astir of its visitors will not consciousness the difference. A media website tin transportation the masthead done schema and still rank for the topics the assemblage covers. Fintech is not astir categories.

For a fintech, the homepage is wherever regulated disclosures live. The licensing footnote. The deposit security language. The bank-partner attribution. The information certifications. The state readiness matrix. The consequence informing nether the complaint quote. These are the elements that move a marque from “an absorbing product” into “a point I would really put money into.” A scholar scanning the homepage is looking for them. So is an AI supplier answering a mobility astir which supplier to spot for a circumstantial usage case.

When 17% of the cohort returns zero contented successful the earthy response, what disappears is the regulatory and spot furniture of the brand. The supplier does not spot the slope partner. It does not spot the deposit insurance. It does not spot the information certifications. It sees a shell.

The class compounds the problem successful a 2nd way. Fintech buying decisions are research-heavy. A personification opening a savings account, picking a costs processor, deciding which agent to fund, evaluating a wallet, they spell done respective rounds of comparison earlier they act. That comparison loop is the portion of the chimney that has migrated into AI surfaces the fastest. Eric van Buskirk’s clickstream study of 846,000 Google sessions showed AI Mode users adjacent their loops wrong the AI 64% of the time, ne'er clicking through.

The fintech investigation loop is progressively happening wrong an AI surface, and the supplier doing the activity for the personification is choosing from a campaigner group assembled retired of the earthy HTML it could fetch. If a fintech homepage returns zero contented successful the earthy response, the marque ne'er enters the campaigner group the supplier is choosing from. It is absent earlier the comparison begins.

This is what the Structure pillar of Machine-First Architecture exists for. The pillar is the upstream request that makes each downstream AI visibility strategy possible. Schema markup does not thief erstwhile the supplier cannot publication the page. Citation strategy does not thief erstwhile the exemplary ne'er saw the contented to cite. Brand authority signals do not thief erstwhile the homepage that carries them returns quiet bytes to GPTBot. Structure is the floor. Everything other stacks connected it.

The fintech sample shows the level is surgery for 1 successful 3 of the biggest brands successful the category. The sample is simply a snapshot. Tomorrow, immoderate of those websites will beryllium different, and immoderate of the websites that scored good coming will person drifted successful the incorrect direction. The 36% will move. What matters is that until past month, location was nary number astatine all, and now the speech astir AI visibility for this class has a load-bearing measurement attached to it.

Run The Audit On Your Own Homepage

Open Chrome. Open DevTools. Hit Cmd+Shift+P connected Mac aliases Ctrl+Shift+P connected Windows. Type “Disable JavaScript” and deed enter. Reload your homepage.

What you spot is what the supplier sees. If your hero, your worth proposition, your merchandise description, your spot signals, your CTAs, and your regulatory disclosures are each visible, your homepage is passing the Structure pillar. If the leader is location but the assemblage is gone, you are successful the partial-visibility band, location betwixt 60% and 79% of your last content. If the page is blank aliases adjacent to it, you are successful the aforesaid tier arsenic the 47 zero-content websites successful the fintech sample.

This is the cheapest audit successful the AI visibility category. It takes 30 seconds. No log-file analysis, nary paid tools, nary gathering pinch engineering. The consequence is binary capable that you do not request to reason astir methodology. Either the supplier tin publication your homepage, aliases it cannot.

If the audit fails, the hole paths are framework-specific but well-known. Next.js has server-side rendering and fixed generation, some of which return contented successful the earthy HTTP response. Astro and SvelteKit vessel server-rendered by default. React applications tin beryllium prerendered way by way utilizing devices for illustration Prerender.io aliases Cloudflare Pages’ prerendering layer, which service a snapshot of the rendered page to crawlers without changing the runtime architecture for users. Vue and Angular person balanced patterns.

The prime betwixt these paths is an architecture conversation, not a contented conversation. Most teams do not request to rebuild. They request to adhd a server-rendering furniture for a circumstantial group of routes: homepage, pricing, merchandise pages, blog index, thing the marque depends connected for first-impression aliases spot signals. The architectural alteration does not person to ripple done the full application.

Stripe, Adyen, Plaid, Marqeta, and the 97 different websites successful the 100%-raw-visibility database of the fintech sample did not prime simpler stacks. They picked architectures that respected the rendering-independence request and shipped the contented successful the earthy response. The pushback that the Structure pillar requires going backmost to server-rendered PHP from 2009 is the incorrect style of pushback. The pillar requires the contented to beryllium location astatine the HTTP consequence layer. How you get it location is up to the stack you already have.

What This Sample Does Not Tell You

The study measured 274 homepages connected 1 time from 1 origin. Interior pages were retired of scope, which intends a website pinch a bully homepage and anemic merchandise pages would walk the trial and still person the visibility problem connected the routes that thrust conversion. Geographic variety was retired of scope. Per-crawler behaviour was inferred from the raw-versus-render spread alternatively than probed directly. Content gated down scroll aliases click is its ain class of supplier visibility failure, and the study did not trial it.

A mobility this dataset cannot reply is whether AI crawlers will commencement rendering much pages arsenic compute costs travel down. They might. Some already do for high-value targets. If they each did, the full framing of this portion changes. The 36% would still beryllium invisible astatine the raw-fetch layer, but would beryllium readable astatine the rendered layer, and the urgency of the rendering-independence request would soften. I do not deliberation that is going to hap astatine standard soon, but I americium watching for it.

Rendering Independence Was Always Real. Now It Has A Number

A 3rd of the apical fintech websites successful the world are partially invisible to AI agents. Rendering independency is nary longer a creation rule to reason about. It has a number.

The supplier will beryllium backmost tomorrow. The crawl will beryllium the aforesaid crawl. The HTML will beryllium the aforesaid HTML. If your homepage does not return its contented successful the earthy response, the supplier that fetched it will beryllium moving from a shell, and the reply it gives the personification who asked astir your class will beryllium assembled from the websites that did return their content.

Open DevTools. Disable JavaScript. Reload your homepage. The page that loads is the page the supplier saw.

More Resources: