Latest insights on Agent Experience Optimization, AI search trends, and strategies for optimizing content for LLM agents.
Ten patterns from our agentic interaction patterns reference, from the entry point to progressive automation, each with a shipping product that does it today and a scenario where you would need it.
Agentic experience design has no pattern for the moment an agent gives up. The research on what a handoff is worth, why knowing when to stop is its own failure class, and why the oversight literature disagrees with itself.
Reliability is quoted per action and experienced per task, and the gap is where agent integrations fail. What compounding does to a five-step flow, and what it changes about agentic experience design.
Hold the model and the task constant, change only what the agent is allowed to observe, and success moves by thirty points. Why the interface, not the model, is the agentic experience design problem.
Meta's Muse and OpenAI's Dots both open a browser and act on a person's behalf rather than reading and summarising. What that changes for your site, and what to do first.
Wikipedia's article on this discipline has been under an open rename dispute for a month. Where each term came from, why GEO and AEO are used interchangeably, how Google and Microsoft split on the vocabulary, and why the naming is unstable in the first place.
OpenAI killed Atlas, Microsoft retired Copilot Mode, Google folded Project Mariner into Gemini. Standalone AI browsers failed as products while agentic browsing succeeded as a feature — and that changes what arrives at your server.
Search Console now separates AI Overviews and AI Mode impressions from ordinary Search. What the report shows, what it cannot tell you, and how to read the trend — worked through with this site's own data.
OpenAI now runs four distinct user-agents — GPTBot, OAI-SearchBot, ChatGPT-User and the new OAI-AdsBot. A blanket block on all of them removes you from ChatGPT citations. A worked robots.txt that treats each as a separate decision.
Amazon won an injunction blocking Perplexity’s Comet browser from shopping on its site. The Ninth Circuit vacated it, reasoning that a user-directed agent is the user acting. What that changes about how you write terms, rate limits, and bot policy.
Cloudflare replaced Pay Per Crawl with Pay Per Use and split crawler permissions into Search, Agent and Training. New ad-monetized domains get a restrictive default from September 15 — including a trap for mixed-purpose crawlers.
The share of AI Overview citations coming from top-ten organic results roughly halved in a year, and ChatGPT never tracked Google closely at all. Why the link broke, and what to measure instead.
A 137,000-site study found almost nobody fetches llms.txt, and Google has documented that it does not use the file. We recommended it too. Here is the correction — and what the file is genuinely good for.
A W3C draft co-edited by Google and Microsoft lets a site expose callable tools to agents running in the browser, instead of leaving them to guess at your DOM. What it is, when it beats an MCP server, and why your tool surface is now a design problem.
Which websites consistently appear in AI responses, and the AXO strategies behind their success. Updated for 2026 with current citation-share data.
Cloudflare accused Perplexity of stealth crawling. A year on, the dispute has reshaped crawler economics, agent identity, and the law around who an agent acts for.
Traditional SEO optimized for search engines and human clicks. AXO optimizes for LLM agents and direct citations. Why both matter, and how they now diverge.
AI-powered search changed how people find information. What that means for content strategy, and how to prepare for what comes next.
How structured data helps AI agents understand and cite your content — and where the evidence for schema as a citation lever actually stands.
Analysis of AI search, crawler policy and agent standards — sent when there is something worth reading, roughly twice a month. Unsubscribe anytime.
