By Mohamed Ali - Technology & Digital Economy Commentary
August 11, 2026 - Nobody tells you when the machine skips you.
A lost bid usually leaves a trail. A bad demo gets feedback. But when an AI recommends your competitor instead of you, there is no notification. No bounce rate. No abandoned form. The deal simply never shows up, and you go on thinking it was a slow quarter.
Here is the mechanism behind it.
When a business owner types into ChatGPT, "which provider has the lowest fees for sending money from the UK to Kenya?" , the model does not open a browser. It does not click tabs, expand accordions, or wait for a JavaScript widget to load pricing from an API. It sends a crawler. That crawler reads whatever your server delivers as raw HTML, and then it leaves. If your fee table lives inside a React component, a calculator widget, or a script that fires after page load, the crawler sees an empty shell. Not a partial view. Nothing.
Vercel and MERJ tracked over 500 million GPTBot visits. They found zero evidence it ever executed JavaScript. Not slow execution. Not partial rendering. Zero. GPTBot downloaded script files in roughly 11.5% of visits; ClaudeBot did so in about 23.8%. Neither executed a single line. Googlebot renders JavaScript. Almost nothing else does.
This is the infrastructure gap most teams miss. They assume that if Google can read their site, AI can too. That assumption is structurally wrong, and it is costing real money.
The contrast is worth looking at directly. Google search sends fewer clicks than ever, 83% of searches now end without a website visit when an AI answer appears. But at least Google saw your content before deciding not to send traffic. The AI crawlers building answers for ChatGPT, Claude, and Perplexity may never see your content at all. You are not being outranked. You are being unread.
And here is where the financial services angle gets specific.
Remittance and stablecoin businesses in Africa have built beautiful customer experiences. A user picks a corridor, enters an amount, and a dynamic widget fetches the rate from an API in real time. Perfect for humans. Completely invisible to crawlers. The exact questions your prospective customers are asking AI, "cheapest way to send $200 to Mogadishu," "which stablecoin corridor settles fastest into Kenya," "is this MTO licensed?", require numbers and references sitting in plain text. If yours live inside a widget, behind a click-to-expand tab, or buried in a PDF, the model cannot quote you. It quotes whoever has a boring static page with legible data.
Related articles
The same applies to compliance posture. MSB registration, licensing footprint, settlement times, if these are trapped in downloadable documents or JavaScript-rendered sections, corporate treasury teams asking AI to shortlist providers will never see your name. And nobody will ever tell you why.
Now consider where AI actually gets its answers. Muck Rack studied 25 million links cited by major AI engines. Earned media, reviews, news, forums, directories, other people writing about you, accounted for 84% of citations. Journalism alone made up 27%. Paid and advertorial content? 0.3%. You cannot buy your way into an AI recommendation. Ad budget has a hard ceiling here. Credibility, however, is earnable. The old playbook of spending more on ads until something works is not just inefficient in this environment. It is structurally mismatched.
So this is not a marketing problem. It is a plumbing problem.
Your website now has two audiences. One is human. One is machine. Only one of them can click, but both determine whether the click ever happens. If your server-side rendering is clean, your headings match real customer questions, your entity definitions are consistent across Wikidata and LinkedIn and Crunchbase, and your pricing is readable in raw HTML, you are legible to both. If half your value proposition lives in a widget, you have built a beautiful storefront with the lights off.
The fix is almost insultingly simple. Paste your URL into ChatGPT or Claude and ask why it cannot see you. View your own source code, not the inspect element. Search for a sentence from your pricing page. If it is not in there, the machines closing deals in your market have never read it.
Most businesses will wait until the traffic drop is visible in a dashboard. But AI-driven deal loss does not show up in dashboards. It shows up as silence. As quarters that feel slow for no clear reason. As competitors who seem to be winning on brand alone when in fact they just made themselves machine-readable.
Your website looks great. The question is whether anything that matters is actually in the HTML.