Is "Agent Readiness" a Real Category - or Just Snake Oil?
TLDR Dev
"Agent readiness" scores are popping up everywhere, but they mix up two totally different problems. Most of the buzz is hype — the standards that matter are set by ChatGPT, Gemini, Claude and Perplexity, not scorecards.
Suddenly everyone wants to grade your website on how "agent ready" it is. Cloudflare rolled out a public Agent Readiness Score at isitagentready.com. Google slipped an Agentic Browsing category into Lighthouse. And behind them, a wave of scanners and audits appeared, each happy to hand you a number between 0 and 100 and a checklist of things to fix. The problem is that nobody has really proven any of this moves the needle, and the term itself is doing a lot of heavy lifting it shouldn't.
Dig in and you find "agent readiness" is actually smuggling two unrelated problems into one score. There's discovery — will ChatGPT, Gemini, Claude, or Perplexity actually cite or recommend your site, which is really just AI-era SEO. And there's usability — once something lands on your page, can it search, fill a cart, or check out. Getting good at one does nothing for the other. A site can be perfectly citable and still choke the moment an agent tries to click "buy." Most scorecards blend both into a single tidy number, which sounds authoritative but tells you almost nothing about where to actually spend engineering time.
The bigger issue is that "the agent" doesn't really exist as an abstract thing worth optimizing for. Nearly all measurable AI traffic funnels through four apps — ChatGPT, Gemini, Claude, and Perplexity — with ChatGPT eating the majority share by most counts. That means a protocol is only as useful as its adoption by those four. Ahrefs recently found llms.txt, the file everyone hyped as the new sitemap.xml for AI, gets ignored 97% of the time, and most of the remaining traffic is bots, not agents. So chasing every new spec that shows up on Twitter this week is mostly performance, not preparation.
Google's own take, tucked into Lighthouse's new Agentic Browsing audit, is refreshingly boring by comparison. There's no single grade — Google says outright the standards are still shifting — just signals across accessibility, layout stability, WebMCP, and llms.txt, the last of which is explicitly optional. Most of what's being measured is just solid engineering that happens to also help agents: accessible markup, stable layouts, nothing exotic. WebMCP, which lets a site expose structured actions instead of forcing an agent to click around blindly, is the one genuinely new idea in the mix. Google also just floated Agentic Resource Discovery, an experimental standard for publishing a discoverable list of a site's tools and APIs, with Microsoft, GitHub, Salesforce and Nvidia attached — early, but a name-brand coalition is at least a signal worth watching.
The practical takeaway is unglamorous: stop building for a hypothetical future agent and start following what ChatGPT, Gemini, Claude, and Perplexity actually reward today. Skip the weekly protocol-of-the-month club. The moment one of those platforms says it will boost rankings for agent-friendly sites, the calculus changes — until then, most of the "readiness" industry is just anxiety with a scorecard attached.
My take
I've watched enough SEO cycles to recognize this pattern instantly: a vague future threat, a flood of consultants selling scores, and a standard nobody's actually using getting called the "holy grail." llms.txt being ignored 97% of the time is the tell — build for the four apps that actually send traffic, not for an imaginary universal agent that doesn't exist yet.
Read more about this at: TLDR Dev
