OPEN BENCHMARK · AI RESEARCH

The Agentic Readiness Index

How ready is the web for AI agents? An open, reproducible measure of how well leading websites can be read, understood, acted on, and trusted by AI agents (ChatGPT, Perplexity, agentic browsers). Scored with Google's Lighthouse agentic-browsing audit.

0.50
Median score (of 1.0)
0 / 22
Support WebMCP (the "act" layer)
68%
Fail the agent-readability check
23%
Serve a valid llms.txt

The clearest finding: budget does not predict readiness. The two lowest scorers are martech giants (Salesforce 0.10, HubSpot 0.19), while a single-location small business scored above Apple, Nike, and Walmart. Readiness is not bought, it is chosen.

Leaderboard · v1 · June 2026 · click a column to sort

#SiteTier ScoreReadTrust UnderstandAct

Test your own site (free, one command)

The Index is fully reproducible. Run this and read the Agentic Browsing category:

npx lighthouse@13.3.0 <your-url> --only-categories=agentic-browsing --chrome-flags="--headless --no-sandbox"

The category appears only in the Lighthouse CLI/JSON report, not the DevTools panel or PageSpeed Insights UI.

What the checks mean

Read = accessibility tree well-formed (can an agent parse the page). Understand = valid llms.txt (does the site tell AI what it is). Act = WebMCP tools registered (can an agent operate it). Trust = layout stability/CLS (is it a stable target). Single-run, home page, experimental category — read the pattern, not any single number.

Data & methodology: dataset · Full study (DOI): 10.13140/RG.2.2.10508.48000
By Rahul Kumar (@4AMKarmYogi) — AI Researcher & Founder, Incrementors · LinkedIn · ResearchGate
Open data, CC BY 4.0. Cite the DOI. To suggest a site for the next Index, open a discussion on the dataset.