Deep dives on Agent Experience Optimization, AI agents, and making the web work for machines.
Web strategy in the AI era means optimizing for systems you are not allowed to look inside. The UK's June 17 order forcing Google to explain how it ranks, AI Overviews included, is the first real crack in that opacity. The durable move does not wait for the box to open: build a website a machine can read, parse, and verify.
AI search visibility is not one number, and the metric most tools sell you, citations and mentions, is decoupled from the outcome you actually want, being recommended. Here is what to measure instead, why a single visibility score is noise, and how six practitioners and the 2026 data say to read it.
Google published the Open Knowledge Format this week, a way to store knowledge as a folder of linked markdown files. It was built for internal company data, but it could give a website something better than a flat markdown copy of its pages, a graph of how its ideas connect. I tried it on the No Hacks website to show what that takes.
WebMCP lets your website hand AI agents a set of callable tools. Chrome's security guidance warns those same tools can be used to hijack the agents, through hidden instructions in a tool's manifest or injected into the content it returns. Making a website agent-ready with WebMCP now means securing the tools you expose.
A German court ruled that Google's AI Overview is the company's own speech, not a neutral summary of other pages, and held it liable for false statements about two businesses. The lawsuit is not the real consequence. What matters is what an answer engine does once it can be sued for what it says, and why that makes machine-readable brand identity table stakes.
Apple and Google both put an AI agent in the phone this month, but they act on different surfaces. Apple's Siri AI reads the web for answers and acts inside apps. Google's Chrome Auto-Browse visits your website to complete tasks, by default, on hundreds of millions of Android phones.
The agentic web's protocol layer is splitting into two bets: an identity file (llms.txt) that tells AI who you are, and a capability standard (WebMCP) that lets an agent actually do something on your website. Most websites have been defaulted into the first without choosing it. The second is the one that moves a task forward.
AI models cite Reddit more than almost any other source, so an industry has formed to manufacture those citations: aged accounts, paid upvotes, and ghostwritten threads sold as Answer Engine Optimization. A citation surface you can buy is a citation surface that gets filtered, the same way Google's Penguin update filtered bought links. The honest move is to not depend on a surface you neither control nor trust.
Google added AI-search visibility reporting to Search Console, and where it filed the feature is the message: AI visibility is search visibility, measured in the search tool. The report shows impressions only, on Google's surfaces only, which will pull operator attention toward the one engine that now has a free dashboard. Treat it as one input, not the whole picture.
Chrome's new Lighthouse Agentic Browsing audit shipped in version 13.3.0. I ran it on nohacks.co. Five audits passed or returned not-applicable. One failed, the llms.txt audit, because my file is a `.txt` file that the parser treats as markdown, and my links were not encoded in markdown link syntax. The fix is five characters per link. The file is still served as plain text. Only the audit result changed.
36% of the top fintech websites in the world deliver less than 80% of their homepage content to AI agents in the raw HTTP response. That number puts a measurement on the Structure pillar of Machine-First Architecture, the upstream requirement that every AI visibility strategy depends on. Here is what the data says and how to audit your own homepage in thirty seconds.
Sundar Pichai says Google Search is becoming an agent manager where users complete tasks, not browse links. Nick Fox says the optimization is the same for both. If you are treating search and agents as two disciplines, you are running two playbooks for one product.
YC's "make something people want" was always a distribution insight. Agents are now part of that distribution. Cloudflare, Shopify, Stripe, Supabase, Netlify, and Google independently shipped agent-facing infrastructure in the same quarter. Your website needs to work for this visitor class because agents are how people find you now.
Google AI Mode surpassed one billion monthly active users. Their queries are triple the length of traditional search. They arrive at your website post-research and pre-action, ready to complete a task. Your landing pages are still built to persuade someone who has not decided yet.
Google announced at I/O 2026 that Search will render custom interactive layouts per query instead of returning the ten-blue-links template. The unit of measurement has shifted from page views to whether your data was the one rendered. Inside the announcement and what changes for websites.
In one week, the CEOs of Condé Nast and Expedia Group publicly recalibrated the channel-mix conversation from opposite ends: Roger Lynch (Condé Nast) told TBPN to plan as if search traffic will be zero, and Ariane Gorin (Expedia Group) told a Q1 earnings audience that Answer Engine Optimization is now Expedia's fastest-growing channel. Two unrelated verticals on the face of it. The same paid OpenAI counterparty behind both quotes.
Google just published its first official guide for showing up in AI Overviews and AI Mode, naming AEO and GEO as "still SEO" and explicitly debunking five tactics by name. The debunking is correct for the scope it names. The scope it does not name is what happens when an AI agent acts on your website instead of citing it.
If you searched "agentic commerce protocol specification for small merchants," you are looking for the wrong document. The right answer is in your platform admin, not the spec. Here's the decision tree by platform (Shopify, BigCommerce, Wix, Squarespace, WooCommerce, direct Stripe, direct PayPal), what to skip (AP2, UCP Cart, Stripe Projects), and the 90-day playbook to get fully agent-ready.
Stripe launched Projects on April 30, 2026 with Cloudflare, Vercel, and Netlify as launch partners. AI agents can now create accounts, buy domains, upgrade plans, and deploy infrastructure on behalf of human owners. Infrastructure buying is the second commerce category of the agentic web after retail, and the audit questions are different.
Amazon sued Perplexity over its Comet browser shopping on Amazon under user authorization, won a preliminary injunction, then lost it to a Ninth Circuit stay. At the June 11 oral argument the judges wrestled openly with a 1986 hacking statute that has no concept of an AI agent's intent or access, and did not rule. The case is the first real test of who counts as an authorized visitor when the visitor is an agent, and the law has no answer yet.
Google announced today that Chrome auto-browse lands on Android phones in late June 2026, baked into the operating system itself. Not an app you download. Not a browser extension you install. The agent ships as part of Android. Every Pixel 10 and Galaxy S26 user gets it by default, with a rollout to 200 million Android devices by end of year. Read together with nine other Google announcements from the last six months, the agentic-web stack is shipped and the question evolves from "does your website get cited?" to "does your website let the agent complete the booking?"
Google's web.dev published seven rules for building agent-friendly websites. I tested nohacks.co against all seven; six passed and one failed on every native button on the website. The cause is a Tailwind v4 default that ships with no warning, and the fix is three lines of CSS.
Google donated the Agent Payments Protocol to the FIDO Alliance today, and the headline writes itself. The story underneath is the 60-organization list sitting next to Google: Mastercard, American Express, PayPal, Revolut, Coinbase, Adyen, Etsy, JCB, UnionPay International, Ant International, Intuit, Salesforce, ServiceNow, Mysten Labs, Worldpay. Cross-industry payment infrastructure does not arrive on the same standards-body announcement by accident.
Google's Gemini Deep Research Max, launched on April 21 as a public preview on the paid Gemini API tier, is the first production agent that blends public-web retrieval with private MCP servers and file stores in a single reasoning loop. This is extremely early. The pattern it ships is a leading indicator for how the agentic web evolves, not something most websites will encounter this quarter.
The agentic web is the layer of the internet where AI agents, acting on behalf of humans, discover, read, and transact with websites. It exists alongside the human web and is measured separately.
Cloudflare Browser Run added experimental WebMCP support on April 15, becoming the first Tier 0.5 vendor to run agent infrastructure at scale against the browser API that has sat in Chromium 146 behind a flag since February. Every model vendor will have to follow because the economics of screenshot-analyze-click agents are unsustainable.
Cloudflare shipped isitagentready.com on April 17, a scanner that scores any website on emerging agent-legibility standards. I ran my own website through it and scored 33/100 Level 2 "Bot-Aware" after one small fix. The score is useful as a trigger for audits, not as a target. Here is what the audit actually checks, why a low score can be structurally misleading, and what to do with the result this week.
Adobe Analytics published the 2026 Q2 AI Traffic Report on April 16. AI-referred traffic to US retailers grew 393% year over year, and now converts 42% better than non-AI traffic, a full reversal from roughly half the non-AI rate twelve months earlier. The conversion sign flip is the real story, and most websites still don't know it changed which number matters.
On April 15, Cloudflare and OpenAI shipped competing AI agent runtimes within hours of each other. The next day, Cloudflare added five more pieces of agent infrastructure. Web professionals have been asking which AI model to optimize for; the more useful question, after this week, is which agent runtime your website is legible to.
A definitive reference to the AI user agents visiting your website in 2026. Five functional categories, every verified user agent string, every published IP range, and an honest take on robots.txt, Web Bot Auth, and llms.txt.
Google's Universal Commerce Protocol is the first production implementation of the Interaction pillar of machine-first architecture. The architecture generalizes into five principles any website can apply, long before a standard protocol arrives for the rest of the web.
On March 20, 2026, Google added Google-Agent to its official fetcher list. It ignores robots.txt, signs requests with cryptographic keys, and browses websites on behalf of users. The hybrid web is no longer theoretical.
Google's Universal Commerce Protocol now lets AI agents build shopping carts, query live catalogs, and carry over loyalty benefits. With simplified Merchant Center onboarding and Salesforce, Stripe, and Commerce Inc implementing the protocol, UCP is moving from announcement to infrastructure.
Your website used to be the destination. Now it's the source. AI agents summarize, recontextualize, and relay your content without your layout, fonts, or scroll animations. If your message only works inside your design, you don't have a message. You have a brochure.
Google patented a system to replace your landing page with AI. Chrome browses on your behalf. Agents handle checkout. For the first time, we have infrastructure for a web where no human creates the page and no human visits it. Here's what that means.
Checkout is becoming a protocol, not a page. Here's how the Agentic Commerce Protocol, the Universal Commerce Protocol, Stripe Projects, AP2, and the payment-network standards are turning AI agents into buyers across retail and infrastructure, and what it means for your business.
Cloudflare's new Markdown for Agents feature converts HTML to markdown on the fly when AI agents request it. An 80% token reduction, built into the CDN layer. Here's what it means for your website.
AI agents don't see your website the way humans do. They read the accessibility tree. Here's how the major platforms actually perceive web pages, what the research says, and how to make your website agent-ready.
WebMCP (Web Model Context Protocol) is a W3C browser standard that lets a website expose its own features to AI agents as structured, callable tools through a new navigator.modelContext API. As of June 2026 it's in a public origin trial in Chrome 149. Here's what it is, how it works, and what it means for your website.
Four protocols are forming the infrastructure of the agentic web. Here's what MCP, A2A, NLWeb, and AGENTS.md do, how they fit together, and what the Agentic AI Foundation means for your business.
A practical guide to Answer Engine Optimization (AEO). How AI search engines parse content, what gets cited, and what Google, Microsoft, and OpenAI actually recommend.
The evolution from SEO to AEO to AAIO, and why December 2025 marks the turning point for optimizing websites for AI agents.
The complete guide to AI-powered agentic browsers in 2026. Every browser, automation framework, and enterprise API from Chrome's auto browse to Claude for Chrome, and what they mean for your website.
Google's browser can now scroll, click, and fill forms autonomously. With 3 billion Chrome users, the agent traffic wave just went mainstream. Is your website ready for visitors that aren't human?
On January 12, 2026, Shopify auto-opted merchants into agentic shopping. Customers can now discover, evaluate, and purchase products without ever visiting a website. What does brand control look like when AI is the interface?
Agent Experience Optimization (AXO) makes your website work for AI agents. Learn the four pillars: structured data, content accessibility, information architecture, and server-rendered content.
