Google quietly did something the AI-visibility industry should be nervous about: it published official documentation on optimizing for generative AI features — with a mythbusting section. For a category full of confident advice and few primary sources, this is the closest thing to ground truth we're going to get. It's worth reading in full. Here's the plain-English version, with the parts most vendors won't quote.
What Google says doesn't work
- llms.txt. "You don't need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search." The file gets no special treatment — it neither helps nor harms in Google Search. (Some AI agents and developer tools do read it; Google's crawlers don't care.)
- Special schema for AI. Structured data "isn't required for generative AI search, and there's no special schema.org markup you need to add." It remains useful for rich results and unambiguous facts — but it is not an AI-visibility lever on its own.
- Chunking and rewriting content "for AI." No requirement to fragment pages or write in a special AI-friendly register. Google's systems read normal, well-organized pages fine.
- Manufactured mentions. Seeking artificial "mentions" across the web isn't effective compared to genuinely useful content. The link-exchange networks sold as "AI backlink building" are exactly this.
What Google says does work
The unglamorous list: unique, genuinely useful content; clear structure and headings; crawlability; good page experience; real images and video; consistent business data (Business Profile, Merchant Center). And the sentence that should reframe the whole category: generative AI features are "rooted in core Search ranking and quality systems." AI search is still search. When ChatGPT answers a buying question, it runs a web search and composes from what it trusts — the same fundamentals decide both.
The tactics Google just debunked are the easiest ones to sell — a file, a plugin, a bulk content run. The things it endorses are the ones you can't fake.
What we changed the same week
Incisory's readiness check used to score llms.txt as one of six equal signals — a site without one lost points. Per Google's own guidance, that was too strong: llms.txt is now an unscored bonus check in every audit, labeled for what it is — read by some AI agents, ignored by Google. Schema stays checked, reframed around what it's actually for: rich results and unambiguous facts, not a magic AI switch. When the primary source speaks, the honest move is to update — publicly.
What this doesn't change
The measurement problem is untouched. Google's guide tells you how to be eligible; it doesn't tell you whether ChatGPT, Gemini, or Perplexity actually name you when your buyers ask — or who they name instead, or which third-party sources tipped the answer. That's still only knowable by asking the engines, blind, and reading the answers word for word. That part — measure, fix what's real, prove it moved — is the whole job. Run the free audit and see where you actually stand today.