Ranking in answers, not just results

Google ended the argument on May 15, 2026

Two outcomes by the end of this hour: one of your real money pages GEO-audited and fixed, and a citation log running weekly. The third is immunity to an entire category of snake oil. For two years, an industry of GEO/AEO consultants sold special tactics for ranking in AI answers: llms.txt files, content chunked for retrieval, AI-specific rewrites, secret schema. Then on May 15, 2026, Google published its official guide to optimizing for generative AI features, and the headline was blunt: AEO and GEO 'are still SEO'.

The guide explicitly says you do not need llms.txt, do not need content chunking, do not need AI-specific rewrites, and do not need special schema markup to appear in AI Overviews or AI Mode. What gets you cited is what got you ranked: genuinely useful, crawlable, well-structured content.

The llms.txt verdict

llms.txt deserves its own autopsy because it was 2025's most-hyped tactic. The idea: a file telling AI crawlers what matters on your site, like robots.txt for LLMs. The reality: no major AI company committed to reading it, and one AI-visibility vendor (Limy) monitored over 500 million AI-bot visits across 90 days and counted only 408 requests for llms.txt. The bots are not asking for the file.

The meta-lesson outlasts the tactic: in a gold rush, verify against crawler-log data and official guidance before adopting anything. That habit, applied quarterly, is your entire defense against the next llms.txt.

What actually moves AI citations

Still-SEO does not mean nothing changed. Answer engines select and quote differently than ten blue links did, and the cross-platform consensus on what earns citations is consistent.

• Answer the query completely in the first ~200 words. Answer engines extract; they don't reward suspense. The full answer up top, the depth below.
• Extractable structure: bullets, FAQs, and tables get lifted into answers, often near-verbatim. Prose walls don't.
• Factual density and original data. Your own numbers, benchmarks, and research are the most citable thing you can publish - nobody else has them.
• Entity consistency: your company, product, and category described the same way everywhere, so models resolve you as one coherent entity.
• Third-party presence - the big one. 65%+ of content cited by AI answers comes from publishers, review sites, UGC, and community threads, not the brand's own site.

Notice every item on the list also helps a human skimmer and classic Google. That's the point Google was making: there is no fork in the road between SEO and GEO. There's just better-structured, denser, more-cited content.

Schema: still worth automating

Special schema for AI is a myth, but ordinary structured data still matters for Google surfaces - FAQ, HowTo, and LocalBusiness JSON-LD measurably helps inclusion in AI Overviews. And it's perfectly mechanical work, which means it belongs in your publish-cms skill, not in anyone's afternoon. Schema and SEO-brief skills are exactly the generic plumbing the open catalogs carry - install the library's version if one exists, then adapt it to your site's conventions and gates.

{
  "@context": "https://schema.org",
  "@type": "FAQPage",
  "mainEntity": [{
    "@type": "Question",
    "name": "Do AI answer engines read llms.txt?",
    "acceptedAnswer": {
      "@type": "Answer",
      "text": "Effectively no. In one vendor's sample of 500M+ AI-bot visits,
        only 408 requested llms.txt, and no major AI company
        commits to reading the file."
    }
  }]
}

1. Extend publish-cms: when a post contains Q&A material, generate FAQ JSON-LD from the actual questions and answers in the content - never invent Q&A for schema's sake.
2. Validate every generated block with Google's Rich Results Test before the publish package is finalized - broken schema is worse than none.
3. Backfill: run the generator across your existing top 10 pages.

The citation-monitoring habit

You can't manage what you never look at, and almost nobody looks at AI answers systematically. The habit is simple: a fixed set of category queries, run weekly across ChatGPT, Perplexity, and Google's AI Overviews, with results logged. Who gets cited? Are you mentioned? Which sources show up week after week?

1. Write 10-15 queries a real buyer would ask in your category - 'best <category> for <segment>', 'how to <job your product does>', '<you> vs <competitor>'.
2. Each week, run them across the three surfaces and log to data/analytics/citations.md: query, surface, brands mentioned, sources cited.
3. After four weeks, read the pattern: the repeatedly-cited third-party sources are your outreach shortlist - those are the places worth being reviewed, listed, or quoted in.
4. Feed gaps into the calendar: queries where competitors get cited and you don't are briefs waiting to be written.

Do this now

Sources and further reading

• Google: optimizing your website for generative AI features
• SEJ: Google's guide calls AEO and GEO 'still SEO'