AI Search Myths: GEO Tactics That Are Probably Overhyped

Which claims lack evidence?

Several tactics sold as GEO wins don't trace to a source that actually tested them. Below are six, checked against Google's own guide and the largest brand-correlation study we could open, with a dated verdict on each — opinion, clearly labeled, not fact dressed up as one.

What does Google's guide say to ignore?

Directly and by name: content rewritten specifically for AI systems, chasing inauthentic mentions across the web, keyword-variant stuffing, content chunking, and llms.txt or other machine-readable files. Google's own framing is that "optimizing for generative AI search is optimizing for the search experience, and thus still SEO" — its guide doesn't introduce a parallel set of AI-specific hacks, it rules several out explicitly.

That's one company's statement about its own features. It says nothing about how ChatGPT, Claude or Perplexity weigh these same tactics, since none of the three has published an equivalent statement either way.

Which claims have partial evidence?

Backlinks, domain authority and page volume aren't zero-evidence — they're weak-evidence. Ahrefs' correlations put them well above zero but well below branded mentions and YouTube presence. That's a real signal, just not the signal some GEO advice treats it as. The honest read: these factors probably help a little, as part of a bundle with everything else that makes a brand well-known, not as a standalone lever worth optimizing in isolation.

What would count as proof?

A study that manipulates one variable — adds schema to one set of matched pages and not another, publishes llms.txt on half a prompt set's target sites and not the other half — and measures a real difference in citation or recommendation rate, with enough runs to call the result more than noise. Correlational studies like Ahrefs' 75,000-brand analysis are useful and large, but the authors themselves are explicit: "correlation isn't causation." Nothing we opened for this post is a controlled test of any of these six tactics. If one surfaces, or if OpenAI, Anthropic or Perplexity publish documentation confirming active use of any of them, we'll update this table and the change log below.

What's the strongest counterargument?

That absence of a controlled test isn't absence of effect — these tactics could still matter in ways a correlation study or a single company's guide doesn't capture, especially for engines other than Google's own features. That's fair, and it's exactly why this post is labeled opinion rather than fact: we're reporting what's been tested and what hasn't, not claiming these six tactics definitely don't work anywhere.

What should a team do instead?

Spend the time these six tactics would have taken on the items with Strong-rated evidence instead: crawler access, original information rather than rewritten summaries, and brand mentions on sites AI engines already pull from. See the full SEO, AEO and GEO checklist for the complete, evidence-rated list this table draws its six items from.

FAQ

Does this mean llms.txt and schema are useless? No. It means neither has demonstrated evidence of improving AI citations specifically. Schema still helps classic rich results; llms.txt has no identified downside. Neither is the lever some advice claims it is.

Why pick these six tactics and not a bigger, rounder number? Because a fixed count like 15 or 25 would mean padding the list with items we can't source as well as these six. We'd rather show six well-evidenced verdicts than twenty guesses.

Is this the same as saying SEO doesn't matter for AI search? No — the opposite. Google explicitly ties its AI features to its core ranking systems. This post is about specific AI-only tactics layered on top of SEO, several of which don't hold up, not about SEO itself.

What would change your mind on any of these? A controlled test — matched pages, one variable changed, enough runs — showing a real citation or recommendation difference. See the experiment framework for what that kind of test actually requires.

What to do next

1. Check your own team's current AI-visibility task list against the six tactics above — deprioritize any that rank above crawler access or original content.
2. Read the full checklist for the complete evidence-rated version of this exercise.
3. Revisit this post's verdicts if a primary source or controlled study changes the picture — we will, and we'll log it below when we do.

ContextOwl sells one-time AI visibility reports. The sample report shows what an evidence-backed check of your own brand looks like.