
For most of the history of the web, when a company thought about its audience, it meant people. People searching. People visiting. People shopping. That is changing.
Your customers are beginning to use machines to discover, evaluate, and shop from businesses on their behalf. The customer asks an AI assistant. The assistant opens five websites. The customer gets one combined answer. Maybe they click a source. Maybe they do not. Your analytics were built for the customer visiting you directly, so you may see one strange machine request in the middle of that journey, or nothing at all. That machine is not another customer. It is another audience for the information your company put on the web.
This course is a measurement discipline, not a list of AI SEO hacks. It is built around four questions, answered in order:
Are machines finding us?
What are they looking at?
What do they understand when they get there?
Does any of this lead to customers and revenue?
You will work through a live machine-traffic dashboard and real ecommerce product pages. Along the way you will learn to:
Separate machine traffic into four roles: traditional search, model development, answer indexing, and live readers
Explain why GPTBot, OAI-SearchBot, and ChatGPT-User are different events sitting at different distances from present human intent
Verify a claimed AI user-agent against published network ranges, and state precisely what a failed check does and does not prove
Read machine traffic as fetches rather than people, and understand why browser analytics never saw those requests
Inspect what a machine reader actually receives from a product page, side by side with what a shopper sees
Score six commercial facts (price, availability, shipping, returns, size or options, and current offer) as confident, ambiguous, wrong, or absent
Diagnose four distinct failure modes: blocked at the door, a successful response that is commercially empty, the wrong price, and a single missing fact
Work the two levers a company actually controls: access at the CDN or bot-management layer, and the product information in the response
Design a machine-readable statement of product facts, and keep a proposed treatment separate from proof that it changed an answer
Design a one-variable experiment on a property you control, using fixed responses, fresh sessions, and repeated runs
Place instruments along the path from request to response to referral to transaction, instead of waiting for one magical AI dashboard
Underneath all of it is one rule: name the event you measured. A fetch is not influence. A citation is not a recommendation. A referral is not a transaction. Every claim gets labeled Observed, Supported, Unknown, or Speculation, and those four are never mixed together.
This is not a prompt-engineering course, and it is not a promise to rank in AI answers in 7 days. By the end you can take any product URL, classify the machine traffic around it, inspect what live AI readers actually receive, score which commercial facts survive, design a controlled experiment, and distinguish what you measured from what you inferred.
Includes screen-by-screen walkthroughs of a live machine-traffic dashboard and machine-eye inspections of real direct-to-consumer product pages from Bombas, Fellow, BLK & Bold, and Unbound Merino.