Continuing on from Google's new guide for ranking in AI features, there's more worth pulling out, both a useful clarification on the content variations point, and a genuinely valuable section that clears up some of the most persistent myths currently circulating in the GEO space.
Clarifying the content variations point
To build on the earlier point about content variations: targeting 12 different demographics with 12 genuine, distinct variations of a piece of content is actually very effective, and doesn't put you at any real risk of the scaled content abuse penalty Google warns about. The distinction that matters is between genuine, unique value created for each audience versus low-effort, near-identical duplication scaled purely for volume. The former is a legitimate content strategy. The latter is exactly what the penalty exists to catch.
The warning against inauthentic mentions
Google also explicitly warns against pursuing inauthentic mentions of your brand, in other words, spamming mentions and recommendations of your brand across the internet in a low-quality, scattershot way.
Their stated reasoning is straightforward: Google is already very good at detecting spam, and as a result, they simply ignore the pages where these kinds of inauthentic mentions get built. That's a reasonable claim, and it's largely true. Google does ignore the majority of genuinely spammy pages.
This is exactly why our approach has always been to only build recommendations on pages that are already being cited by Google in the first place. There's no guesswork involved in figuring out whether a page is spammy or being ignored. Google tells you directly, through its own citation behaviour. If a page is showing up as a cited source, it's clearly not being filtered out as spam. If it isn't being cited, building a mention there was never going to move the needle regardless of Google's spam detection.
The genuinely useful part: myth-busting
Here's where the guide actually earns its place. Google used this as an opportunity to directly clear up a number of myths that have been circulating widely across the GEO space, several of which I've addressed before, and it's genuinely valuable to have Google confirm them directly. According to Google's own guidance:
You don't need an llms.txt file. This lines up with data from Ahrefs showing the overwhelming majority of llms.txt files aren't being referenced by AI search at all, and Google confirming it directly here removes any remaining ambiguity for ranking purposes specifically.
You don't need a markdown version of your pages. A myth that's circulated widely as businesses try to make their content more "machine-readable." Google is telling us directly that this isn't a ranking factor.
You don't need to worry about chunking your content in any special technical way. Structuring content clearly for readability and clear answers still matters, but that's a content and structure decision, not some separate technical chunking requirement to implement.
Adding structured data doesn't help you rank in AI features. This one might be the most surprising to a lot of people, given how heavily structured data has been pushed as a GEO best practice. Google is stating plainly that it isn't a ranking lever here.
Where to look for genuinely practical guidance
If you're looking for a guide that actually offers useful, actionable information on how to get cited in AI, rather than one that's more useful for its warnings than its recommendations, I'd point you toward Microsoft's guide instead. It's considerably more practical and specific than what Google has published here.
The overall takeaway
Between the two pieces of this guide, the warnings and the myth-busting, Google has quietly given the GEO space something genuinely useful: a clear signal on what doesn't matter, even if the guide is light on what actually does. Cutting several persistent myths off at the source is worth something, even from a guide that otherwise leans heavily on the generic "just write good content" advice that's always been technically true and rarely actionable.