A new official reference
After this lesson you can describe what Google has published about optimising for its generative AI features, in your own words, and you can use it as a filter for the advice you meet.
Dated: read in 2026-11. Source: Google Search Central blog post of 2026-05-15 announcing a new resource for optimising for generative AI in Google Search, and the guide it points to, which sits in a section Google calls generative AI fundamentals. All points below marked documented come from those pages; I have not quoted them, and the exact wording is worth reading yourself.
What the guide covers (documented)
According to the announcement, it covers: the value of unique, non-commodity content; local, shopping, image and video content; myths about AEO and GEO; early guidance on AI agents; and why SEO best practice still underpins success.
Its main recommendations, in my summary:
Foundations. Keep core SEO practice. Publish original, first-hand, non-commodity content. Use clear headings, sections and readable paragraphs. Add relevant, good images and video. Apply Search's usual rules to AI-assisted content too.
Technical. Meet the Search technical requirements; the page must be indexed and snippet-eligible. Keep content crawlable. Use semantic HTML where practical. Follow JavaScript SEO basics. Improve page experience (mobile display, speed, clear main content). Reduce duplicate content. Verify the site in Search Console.
Business and commerce. Use Merchant Center feeds for products and keep Google Business Profile details current.
Measurement and emerging areas. Use the generative AI performance report in Search Console. Consider whether browser-based agents can use your site. Watch emerging protocols (the guide mentions UCP, which I have not studied; check it before commenting).
What Google says you can skip (documented)
The guide includes a list of myths, tactics it says are unnecessary for Google Search and its generative features:
- llms.txt and other special AI or machine-readable files.
- Special markup or Markdown meant for AI systems.
- "Chunking" content into small pieces to suit models.
- Rewriting content specifically for LLMs.
- Hunting for long-tail keyword variants.
- Mass-creating pages for query variations or fan-out queries, which Google treats as scaled content abuse.
- Seeking inauthentic brand mentions across the web.
- Over-focusing on structured data: it is not required for generative features, though it can still be useful for rich results that exist.
- Trusting third-party tools that claim ranking success or knowledge of internal Google metrics.
- Treating AEO and GEO as separate from SEO.
Read that list slowly. Several items are advice sold widely elsewhere. For Google specifically, the vendor says it is not needed. That does not prove the tactic does nothing for other assistants (unknown), but it removes the justification "Google needs this".
Applying the guide
Where the guide is strongest: it describes its own product. A statement like "the page must be indexed and snippet-eligible" is as solid as it gets.
Where care is needed. The guide is one company's guidance about its own product, and it has an interest in how people behave. Treat it as the best available source for Google and not as proof about ChatGPT, Copilot, Perplexity or others. Also note it is a recommendation document, not a guarantee; Google does not promise inclusion.
What stays hypothesis. Anything about how a specific paragraph, heading or phrase is chosen. The guide says clear structure is good; it does not say that a 40-word answer box triggers inclusion.
A filter you can reuse
Take any tip and ask three questions:
- Does it help a human reader? If yes, low risk, do it regardless.
- Does the vendor document it, or document the opposite? Look it up.
- Would it still be acceptable if the AI feature disappeared tomorrow? If not, be careful.
Example: "Split every article into 100-word chunks with a heading each." Question 1: it might make skimming worse. Question 2: Google lists chunking among tactics to ignore. Question 3: the page would read oddly. Verdict: skip.
Example: "Add a short definition near the top of an explanatory page." Question 1: yes, readers like it. Question 2: not forbidden. Question 3: still fine. Verdict: do it, without expecting a particular AI effect.
Tip: "Add llms.txt to rank in AI Overviews"
Google (2026-05 guide): llms.txt can be ignored for Search.
Label: not needed for Google; usefulness elsewhere is hypothesis.Common mistakes
- Quoting the guide as if it covered every AI product.
- Ignoring the foundation items because they sound dull; Google puts them first.
- Treating the "ignore" list as proof that structured data is worthless; it says "not required for generative features", and rich results are a separate matter.
- Relying on a summary article instead of opening the guide; summaries drift.
- Building a pile of near-duplicate question pages. Google lists this as scaled content abuse.
Exercise
- Open the guide and make two lists from it: five things to do, five things to skip. Check them against this lesson and note where your reading differs.
- Take three tips from an agency blog and run the three-question filter on each.
- Write one sentence explaining why the guide cannot tell you what Perplexity does.
Show a sample answer
For step 3: the guide documents Google's own Search features, so its statements apply to Google's crawlers, index and products. Perplexity is a different company with its own documentation.
Previous: What Google documents about AI Overviews and AI Mode. Next: Bing and Copilot.
Key points
- Google published a guide to generative AI features in Search on 2026-05-15 through the Search Central blog.
- It says SEO fundamentals still underpin success and that original, non-commodity content matters most.
- It lists tactics it says can be ignored, including llms.txt, special markup, chunking and rewriting for LLMs.
- It warns against mass-creating pages for fan-out queries and against inauthentic brand mentions.
- The guide is one vendor's view of its own products; other assistants are not covered.