Last updated: 2026-10-09
A measured, non-commercial guide to being found in the age of AI agents. Sixteen lessons across five modules: how search has shifted, how AI agents read a website (screenshots, HTML, accessibility tree), structured data and clean HTML, web.dev agent-UX patterns, content for machines and humans, off-site signals and freshness, agent protocols (MCP, AP2, NLWeb, UCP, WebMCP), measurement, the five-step agent-ready audit, the first-mover window, and a final calibration against Google's May 2026 official guide. No pitches, no upsells, no proprietary frameworks invented for this course.
Self-paced. Honor-based. Not accredited.
Five short modules covering the new vocabulary (GEO, AEO, AIO), how AI agents actually read a website, structured data and clean HTML, writing for both machines and humans, the off-site signals that build trust, the agent protocols already shipping (MCP, A2A, AP2, ACP, NLWeb), measurement that catches what classical analytics misses, and a one-week audit that takes most sites from invisible to recommended.
Each module stands on its own. Together they give a working frame for thinking about AI search optimisation — one you can apply to your own site, push back on, and extend.
Classical SEO, GEO, AEO, and AIO defined. The delegated customer journey. The five signals AI agents use to read a website and the three interaction modes that shape the work.
Schema.org and JSON-LD. Semantic HTML, headings, ARIA, and the text-to-HTML ratio. Concrete agent-UX patterns from Google web.dev. robots.txt rules for GPTBot, ClaudeBot, and friends. sitemap.xml with lastmod. The llms.txt and AGENTS.md debate.
The four-question page (what, who, cost, how). Tables, lists, FAQ schema, and HowTo blocks. Topical authority and the category-association flywheel.
Citations, mentions, and off-site signals. Freshness and the last-updated signal. APIs, feeds, and the agent-protocol stack — MCP, A2A, AP2, ACP, and NLWeb.
Prompt testing, server-log analysis, citation tracking, and agent-readability scoring. The five-step agent-ready audit. The first-mover window. A final calibration against Google's May 2026 official guide.
Start with the question that won't go away. From Keynes's 1930 forecast to today's AI wave, there is a single thread — and it is worth pulling on.
This course is written for the average website owner — no SEO certification or developer background assumed. Terms are defined where used; claims are sourced; and every lesson points you to the primary readings, including Google's May 2026 generative-AI search guide, the web.dev agent-UX article, and the Vercel, Microsoft, and Schema.org references. If you have read a few breathless blog posts on AI SEO and want something more measured, this is for you.
This course deliberately draws on a wide set of sources — Google's May 2026 generative-AI search guide and the accompanying web.dev agent-UX article, Vercel's agent-readability specification, Microsoft's NLWeb project, Schema.org, the llms.txt proposal, the Model Context Protocol specification, Google's Agent Payments Protocol, OpenAI's Agentic Commerce Protocol, and analyst reports from McKinsey, Gartner, Bain, BCG, Salesforce, and Adobe. Each lesson links to the primary references so you can read them yourself and form your own view.
Course progress, exam results, and certificate details are stored in this browser on this device. They may be removed if you clear browser data. Nothing is sent to a server.
Search optimisation used to mean ranking a page on Google. Today an increasing share of web traffic begins inside an AI assistant or browser agent that never shows the user a list of blue links. This lesson defines the new vocabulary so the rest of the course makes sense.
For twenty-five years, the customer journey assumed a human at every step: a problem, a search, a few clicks, a comparison, a decision. AI agents collapse most of those steps into a single conversation. A site that wants to be chosen has to convince something that does not skim, scroll, or feel.
An AI agent does not see your hero image, your colour palette, or your clever tagline. It sees code, structure, and signals. To optimise a site for agents, you first need a clear picture of what they are actually scanning.
Structured data is the language machines speak. A page that publishes its facts as schema.org JSON-LD is legible to every search engine, AI assistant, and agent on the web; one that does not is asking each visitor to guess. Of all the changes covered in this course, schema markup gives one of the highest returns for the lowest effort — for traditional search, for rich results, and for the AI systems that prefer structured input.
Below the JSON-LD layer, the HTML itself either helps or hinders agents. Pages with clean semantic markup, a clear heading hierarchy, and accessibility annotations are scanned, summarised, and quoted by AI systems. Pages built mainly out of styled divs are not.
Before an agent can read a page, it has to be allowed to fetch it, find it, and parse it efficiently. Two tiny files at the root of a site — robots.txt and sitemap.xml — control most of that. A handful of newer files (llms.txt, AGENTS.md, markdown mirrors) have been proposed; this lesson covers what each one actually does, which assistants honour it, and where Google's May 2026 guidance comes down.
An AI agent reading a page is fundamentally trying to answer a small set of questions on behalf of its user. Pages that answer those questions plainly and quickly are quoted; pages that bury them inside brand storytelling are skipped.
Once a page is written in plain language, the next lever is format. Some structures — tables, lists, FAQ blocks, comparison grids — are radically easier for AI systems to extract and cite than free-flowing prose. Choosing the right format is part of the craft.
An AI assistant asked for the best small-business CRM in the United Kingdom does not pick at random. It picks from a small mental list of brands it associates with that exact category. Owning that association — being the default in the AI's head — is the long game.
AI agents do not trust a website's claims about itself in isolation. They cross-reference. A site whose facts match what other sites say is rewarded; a site that exists in an echo chamber of self-praise is discounted. This is the off-site half of the work.
Stale information is a trust killer. AI assistants weight recent content heavily because users punish them for citing prices, openings, or product details that turn out to be wrong. A maintenance routine — small, regular, visible — is one of the cheapest ways to stay quoted.
The most agent-friendly sites do not just publish HTML for agents to scrape. They provide structured endpoints — APIs, product feeds, conversational interfaces — that agents can query directly. This lesson surveys the protocols that are emerging in 2026 and what to do about them as a non-developer.
If AI traffic does not show up in classical analytics, the temptation is to assume it does not exist. It does. Measuring it requires a small set of new techniques: prompt testing, server-log analysis, citation tracking, and scoring frameworks. None are difficult; almost no one does them.
The last fourteen lessons map onto a single five-step audit that a small team can complete in a week. This lesson is the synthesis — a one-page checklist for taking a site from invisible to recommended.
Every major shift in the web has had a first-mover window — a period in which early adopters locked in advantages that compounded for years. The agentic shift is no exception; the difference is that the window appears to be unusually short. This final lesson sets out the realistic timeline and how to think about it.
The previous fifteen lessons draw on analyst forecasts, vendor frameworks, and emerging protocols. In May 2026, Google published its first formal generative-AI optimisation guide. This final lesson calibrates the course against that official document — where it agrees, where vendor frameworks have softened the picture, and what to read once a year to stay current.