Adnan
Adnan
AboutWorkJournalSchema generatorSay hello
hello@adnankhan.techInstagramXGitHubLinkedIn

Build

Your website has a second audience now, and it does not have eyes

Automated traffic has overtaken human traffic on the web. Agents read markup, not layouts — which quietly changes what a well-built site means.

4 min readAdnan Khan, Delhi

Something crossed over this year without much noise: automated requests overtook human ones. Bots now account for roughly 57.5% of HTML traffic, against 42.5% from people. Not all of that is friendly — scrapers and vulnerability probes are in there too — but a growing share is agents acting on someone's behalf. Comparing prices. Checking your hours. Pulling your services into an answer someone asked for.

Meanwhile Chrome shipped an early preview of WebMCP in Canary in February 2026 — a protocol for letting agents interact with a site in a structured way, rather than by squinting at pixels.

So your site has a second audience. It never sees your hero animation, does not care about your font pairing, and will not scroll. It reads your markup.

What agents actually read

Agents parse structured data far more reliably than they parse visual layouts. JSON-LD, schema.org markup, semantic HTML, real headings, honest link text. The stuff that has been filed under “accessibility, we’ll get to it” for a decade.

That is the interesting part. This is not a new discipline you have to go and learn. It is the same list you were already supposed to be doing, and it just acquired a second, commercially obvious reason to exist:

  • Semantic HTML. One h1. Real heading order. <nav>, <main>, <article>, <time datetime>. A <div> soup is legible to a browser and ambiguous to everything else.
  • JSON-LD for the facts about you. Who you are, what you sell, where you operate, what it costs. Stated once, unambiguously, in a format that does not rely on anyone inferring it from your layout.
  • Content that exists without JavaScript. If your prices only appear after a client-side fetch, you have made them optional. Some agents will wait. Plenty will not.
  • URLs that do not change. An agent that cited you six months ago should still resolve. Redirect properly or do not move things.
  • Honest link text. “Read more” tells a model nothing. The link text is the label on the edge of a graph.

None of this is exotic. What has changed is that the cost of skipping it is no longer theoretical.

Now the correction

I want to be careful not to sell you a panic, because one is coming.

Gartner expects over 40% of agentic AI projects to be cancelled by the end of 2027, on cost, risk, and nobody being able to say what the value was. That sounds like a contradiction of everything above. It is not.

The agent hype will correct. The markup requirement will not.

The projects that die will be the ambitious ones — autonomous systems given budgets and standing permission. What survives is the boring, useful layer: assistants that read the web on someone's behalf and need to understand what they find. That behaviour is already ordinary, and it does not depend on any particular company's roadmap surviving.

Which is the reassuring part of this. You cannot bet wrong here, because every item on the list above independently pays for itself. Semantic HTML is accessibility. Structured data is rich results in Google. Content that renders without JavaScript is a faster first paint. Stable URLs are not losing your backlinks. If agentic browsing stalls completely, you have spent your effort on a genuinely better site. If it does not, you were early.

That asymmetry is rare enough to act on.

What this looks like in practice

The site you are reading is the example I can actually vouch for, since I built it.

It ships ProfessionalService and OfferCatalog JSON-LD describing who I am and what I offer, generated from the same file that renders the visible services list — so the markup cannot drift out of sync with the page, because there is only one source for both. Every post here carries BlogPosting structured data. The prose renders server-side and needs no JavaScript to be read; the animation on the home page is decoration layered on top of markup that stands up without it. Nothing is hidden behind a client-side fetch.

That last point is worth dwelling on, because it is where most modern sites quietly fail. It is very easy in a React application to build a page where the content only exists after hydration. It looks identical to a person on a fast laptop. To a crawler, an agent, or a customer on a weak connection in a lift, it is an empty page.

The one-hour version

If you do nothing else:

  1. View source on your homepage. If you cannot find your own business name, address and main service as text, fix that before anything else.
  2. Add LocalBusiness or ProfessionalService JSON-LD with your real details. Google's Rich Results Test will validate it in a minute.
  3. Check you have exactly one h1, and that it says what the page is about.
  4. Turn JavaScript off and reload. Whatever disappears is invisible to a meaningful and growing share of your traffic.

That is most of the value, and it is an afternoon.

Sources

Bot traffic share and agent parsing behaviour from WorkOS on AI agent web traffic and Reptile Haus on agentic browsers. The WebMCP Canary preview and browser landscape are covered by No Hacks. Gartner's cancellation forecast via Olostep's agentic AI trends.