What did Google actually say in May 2026?
The guide's core message: there is no secret AI markup — appearing in AI Overviews and AI Mode requires being indexed, rankable and useful, full stop. No special files (llms.txt explicitly unnecessary), no new meta tags, no separate AI sitemap. The signals that earn AI selection are the ones that earn rankings, weighted toward extractability and trust.
The strategic read: your AI Overview strategy is your SEO strategy with an answer-engineering layer on top. Teams selling 'AI Overview optimization' divorced from rankability are selling astrology.
The extractability layer
AI Overviews assemble answers from passages, so passage quality decides selection. Every money page should open with a 40–60 word direct answer to its title question; every H2 should be phrased as a question users actually ask; every answer should stand alone without surrounding context. Tables and tight bullet lists extract disproportionately well for comparison and pricing queries.
Schema seals the identity layer: Organization (who you are), FAQPage (labeled Q&A pairs), Article with named authors. Roughly 65% of AI-cited pages carry structured data — it functions as a machine-readable trust signal.
- Answer block above the fold, 40–60 words
- Question-form H2s matching natural phrasing
- Tables for pricing/comparison content
- Organization + FAQPage + Article schema sitewide
Freshness, authorship and the March 2026 update
The March 2026 core update re-weighted information gain: original data and first-hand experience now outrank synthesis. For AI Overviews this compounds — models prefer fresh, dated, expert-attributed sources when synthesizing answers. Practical cadence: money pages refreshed every 7–14 days to quarterly with visible dateModified; every page bylined to a named human with a real bio and sameAs links.
Anonymous content is being priced out of AI answers. If your pages can't say who wrote them and why that person is credible, they're donating their information to better-attributed competitors.
How do you measure AI Overview visibility?
Build a fixed query set (your money keywords plus question variants), check monthly whether AI Overviews appear and whether you're cited, and track the citation position. Pair with Search Console: AI Overview citations show as impressions with characteristic position patterns.
Expect volatility — AI Overview presence per query shifts constantly. The trend across your set matters; individual query wins and losses are noise. Chalk Labs runs this as a standing monthly scoreboard for GEO clients alongside ChatGPT/Perplexity tracking.
Questions we hear about this
They reduce clicks on informational queries — zero-click share keeps climbing — but cited brands gain visibility and authority in the answer itself. The strategy shift: optimize for being the citation, and design pages so the click that does come lands on conversion-ready content.
Per Google's own guidance, no — AI features source from indexed, rankable pages. Fix crawlability, rankings and content quality first; answer-engineering multiplies a working SEO base, it cannot replace one.
Blocking Google-Extended affects Gemini/AI training but AI Overviews run on regular Google indexing. Blocking GPTBot, PerplexityBot or ClaudeBot removes you from those engines' retrieval. If AI visibility is the goal, allow them all — our robots.txt does.
Yes. The rich-result display shrank, but structured Q&A pairs are still parsed during answer synthesis by Google and every other AI engine. FAQ blocks with schema remain among the highest-leverage AEO assets on commercial pages.