Calibrating the course against Google's official line
Why this lesson exists
The fifteen lessons before this one draw on analyst forecasts, vendor frameworks (Vercel, Microsoft, Tealium, Cloudflare), and a set of emerging protocols. None of that work is wrong. None of it is Google. In May 2026 Google published its first formal generative-AI optimisation guide. The document tightens the claims that can be made about generative-AI search optimisation and changes some of the marketing vocabulary around it. This lesson summarises Google's stated position, identifies where it agrees with the course, and flags the places where the course's vendor-driven framing has been softened or contradicted by the official document. Read it alongside lesson 1, where the same source is introduced.
Four things Google explicitly says you do not need to do
Google's mythbusting section lists four practices the company says are not load-bearing for its AI features. First: do not create special machine-readable files such as llms.txt or other novel AI markup formats; Google's systems do not look for them. Second: do not artificially fragment content into tiny chunks for AI consumption; the systems read paragraphs and whole pages perfectly well. Third: do not rewrite content in some special AI register; the underlying language understanding handles synonyms and meaning. Fourth: do not chase inauthentic mentions across the web; the same spam systems that block link-buying also discount mention-chasing. None of these are mortal sins. Some of the listed tactics may still help with non-Google assistants. But they are not Google-search levers, and they should not be the first place a small team puts its time.
What Google does emphasise: helpful content, technical structure, RAG and query fan-out
The same document is clear about what does help. Foundational SEO is still the baseline: indexable pages, clean technical structure, a fast page experience, valid HTML, working internal links. The single most repeated piece of advice in the guide is to write helpful, original, non-commodity content based on direct experience. Structured data is described as not required for generative AI features but valuable for rich results, and recommended as part of the overall strategy. Two technical mechanics get named directly. Retrieval-augmented generation, or RAG, is how AI Overviews ground their answers in Google's index: pages are retrieved and read at answer time. Query fan-out is the practice of generating concurrent related queries before answering. The example in the guide expands 'how to fix a weedy lawn' into separate queries for herbicides, chemical-free removal, and prevention, then synthesises the result.
Where the course needs to be read with Google's caveat in mind
Lessons 1 through 15 have been edited to acknowledge Google's stance, but the original frame is still useful. Treat AEO, GEO, and AIO as industry shorthand for a real shift in user behaviour, not as separate disciplines Google itself recognises. Treat llms.txt as a proposed convention that some assistants honour and Google does not. Treat schema markup as high-impact for rich results and traditional SEO, and as a clarifying tool for AI systems, but not as a hard requirement for AI visibility. Treat the three agent-viewing channels covered in lesson 3 (screenshots, HTML, accessibility tree) and the design patterns in lesson 5 as Google-endorsed work that helps human and machine readers at the same time.
Preferred sources, browser agents, and what to watch next
Google's guide ends by pointing forward in two directions. The first is preferred sources, a Search Console feature that lets users mark which publishers should be favoured in their AI Overviews. For sites with a loyal audience this is a quietly significant ranking input. The second is the agentic web: the rise of browser agents covered in lessons 3 and 5, and protocols such as the Universal Commerce Protocol, WebMCP, and the merchant-side Business Agent that aim to let agents act on the user's behalf using structured commerce signals. Google's framing is conservative. These are emerging, worth tracking, not yet required. The direction is the one this course has spent fifteen lessons describing. Re-read Google's document at least once a year; the field will keep moving, and the official position is the one that should anchor every other source.
Read the previous fifteen lessons and Google's May 2026 guide on optimising for generative AI search. Where do the vendor frameworks and Google's official document agree, and where would acting on one make a competent person look foolish to the other?