Skip to content
The Other Web
Address
aadhar.shThe Other Web

The Other Web

Every page has two audiences: the person looking at it and the machine reading over their shoulder. The semantic web asked publishers to mark meaning; today’s models scrape the human page; the next web must decide how machines act. Paste one URL to compare the human page, the HTTP response, and the browser-rendered result. Fetched server-side, honestly, as AadharshBot.

The machine web, 2026-08:56.0% of HTML requests are bots10.0% of desktop pages have llms.txt>10k public MCP servers$898K in x402 volume over 30 days
Try:
Compare puts Human, HTTP Machine, and Browser Run side by side. Predict the machine pane; the miss is the lesson.
Human view · the live page
Machine view
Server-rendered machine summary. JavaScript can enhance this into the full selected lens; this fragment is the no-script evidence floor.
Observed document observed
The minimum contract a machine can recover from this response.
urlhttps://en.wikipedia.org/wiki/Semantic_Web
titleSemantic Web - Wikipedia
response200 OK
Lens field evidence75/100
local standards mirror13/100
content-typetext/html; charset=UTF-8
payload431420 B
headings23
fetched asAadharshBot/1.0 (+https://aadhar.sh/bot)
Available surfaces human-only
Evidence found during the server-side inspection.
robots.txt
Selected state anatomy
View: both. The browser enhancement can open the complete lens without changing the URL.
Browser Run · Rendered
Browser Run view. Ask Cloudflare to open this URL in a real headless browser and return the rendered page, screenshot, Markdown, accessibility tree, and a clear WebMCP lab boundary.
This is opt-in: browser execution is slower and can run page JavaScript. Runtime WebMCP discovery is reported separately and requires the Chrome-beta lab.
200 OKComparetext/html; charset=UTF-8431420 B660 msfetched as AadharshBot/1.0 (+https://aadhar.sh/bot)
← aadhar.sh · · · fetched by AadharshBot
The state of the machine web
56.0%
Automated clients sent 56.0% of HTML page requests in Cloudflare's latest published window: 47.2% from non-AI bots, 5.8% from AI bots, and 3.0% from mixed-purpose bots.
Cloudflare Radar · checked 2026-08
10.0% / 9.7%
HTTP Archive found an llms.txt file on 10.0% of desktop pages and 9.7% of mobile pages in June 2026. Both shares were about 4.7 times their July 2025 level.
HTTP Archive · checked 2026-08
>10k servers
More than 10,000 MCP servers had been published when MCP joined the Linux Foundation's Agentic AI Foundation in December 2025.
Linux Foundation · checked 2026-08
100+ agent skills
Google Workspace's CLI ships more than 100 agent skills, returns structured JSON, and builds its command surface from Google's Discovery Service at runtime.
Google Workspace CLI · checked 2026-08
2 pay-per-use partners
Cloudflare is testing pay per use with Ceramic.ai and You.com, tying payment to content appearing in results instead of the number of times a crawler fetches it.
Cloudflare · checked 2026-08
$898.13K / 9.04M txns
x402scan recorded $898.13K of volume across 9.04 million transactions over the past 30 days. That is about $0.10 per transaction on average.
x402scan · checked 2026-08
A page's second life as data is now the busier one. The lenses here measure whether a machine can actually read, understand, and act on a page once it has fetched it. Paste a URL to see one site's answer, or watch the movement over time in the weekly census of 16 representative sites.
About The Other Web
Every page now has two audiences: people, and the machines reading over their shoulders. This instrument watches the second one. Its story runs in three acts, and each act opens the evidence beneath it.
Past1995-2011
The semantic web asked first.
Publishers marked meaning up front: meta tags (1995), microformats (2005), RDFa (2008), microdata (2009), Open Graph (2010), and JSON-LD (2011). These layers let machines read facts instead of guessing from prose. The ones that delivered visible traffic, especially link previews, survived; the rest mostly fossilized in place.
See it live:
Present2022-now
Models pay the difference.
Models changed the bargain. They scrape pages built for people and pay the difference in tokens, so publishers no longer have to mark everything up first. Sites answer that appetite with crawler rules, AI-use signals, bot challenges, and pay-per-crawl.
See it live:
Future2024-on
From reading to acting.
The next question is action. Can a site publish tools, identity, payment, and instructions that let an agent do something without driving a human interface? Lens probes those surfaces and lets you switch the missing ones on as counterfactuals.
See it live:
How this instrument behaves
  • Every fetch is identified and cryptographically signed as AadharshBot, apart from the bot-view samples, which are the one disclosed exception.
  • A bot-view sample sends one crawler's published user-agent, read-only and unsigned, to record what that identity gets served. Chrome and curl ride along as controls, because a 403 on every crawler says nothing until something proves the door opens at all.
  • Every verdict stays pinned to evidence you can open. A probe that never answered reads as unknown.
  • Absent metadata renders as absent. Nothing is guessed or backfilled.
  • Delta's simulations stay amber. Green is reserved for signals actually observed.
  • One fetch per ask, server-side, with no logging.
Rubric after IsItAgentReady. Pedagogy after Seymour Papert's micro-worlds and Geoffrey Litt. Window chrome after Redmond, 2001.