Why "AI SEO" advice on LinkedIn is mostly wrong.

This one is half-right in a way that makes it mostly misleading. FAQ sections that contain specific, accurate answers to questions buyers actually ask, structured with proper Question/Answer schema, do help AI Overviews and some retrieval-augmented engines identify citable passages. That part is true.

The version that gets shared on LinkedIn is the part that isn't: that adding generic FAQ blocks to pages you haven't otherwise improved will move your citation rate meaningfully. It won't. A FAQ section on a page that doesn't have genuine topical authority, a named author, or a source-worthy footprint is a formatting intervention on a trust problem. You've made the page slightly easier to parse; you haven't changed whether the model has any reason to cite it.

FAQ sections are a finishing touch on pages that are already citable. They are not a route to citability for pages that aren't.

This advice is aimed at the right target (making content legible to language models) but it points in the wrong direction. The evidence from prompt testing is that models cite content that is precise and specific, not content that is colloquial. "Our platform uses a microservices architecture to enable sub-100ms query response times" is more citable than "we built our platform to be really fast." Both are "natural language." One gives the model something to quote; the other doesn't.

What "write conversationally" is trying to address is the real problem of keyword-stuffed, robotic content that answers no actual question. But the correction isn't to write casually. It's to write precisely and specifically in response to real questions. Precision and natural language are not opposites. Vagueness is not the same as natural.

Schema is useful. It is not the key. The version of this claim that gets shared most often frames structured data as if it's a signal that overrides everything else, as if a properly marked-up page will get cited regardless of its content quality or its off-site footprint. It won't.

The actual mechanism: schema helps a model correctly classify a page's content type, author, and subject matter. A well-structured Article schema with a Person author block that links to the author's external presence is a real positive signal. What it does not do is create source-worthiness where none exists. A thinly-written page with perfect schema is still a thinly-written page. The models have read enough of the web to know the difference.

Schema is infrastructure, not a lever. Structured authorship markup specifically (Article and Person schema with proper sameAs links) is the highest-leverage schema investment for content pages. Everything else is marginal.

There is a correlation between long-form content and AI citations. The causal story is more complicated. Long-form content tends to be cited more because it tends to be more specific, more structured, more thoroughly sourced, and produced by organisations with more developed editorial practices: these are the actual citation drivers. Length is an outcome of those qualities, not a driver of them.

A 4,000-word post that repeats the same claim forty different ways is not more citable than a 1,200-word post that makes four specific, original observations with clear methodology. If anything, the bloated long-form post is less citable because the density of useful, quotable information per passage is lower. Models retrieving passages to synthesise an answer don't benefit from repetition.

The right question is not "how long should this be?" It's "what specific, original, well-supported claim does this page make that a model would want to quote?"

This one is wrong in both directions. Traditional SEO is not dead. Google's traditional index is still the largest and most widely used search system in the world, and it feeds directly into Google AI Overviews. A brand with strong traditional SEO fundamentals (topical authority, clean technical architecture, good backlink profile, consistent entity signals) has a structural advantage in AI search as well as in traditional organic. The work compounds, not competes.

What has changed is the marginal priority. The interventions that matter most at the frontier (entity coherence, source-worthiness, structured authorship, off-site footprint across five source categories) are different from the interventions that moved rankings five years ago. But that doesn't mean ignoring crawlability, canonicalisation, internal linking, or page quality. Those are the foundation that everything else sits on.

The practitioners telling you to abandon traditional SEO for GEO are usually selling something. The practitioners telling you GEO doesn't matter are usually defending a practice that hasn't kept pace with how search actually works now. Both are wrong. The work is to understand which fundamentals still apply, which have diminished in importance, and which new signals matter, then build a programme that accounts for all three.

That programme is what I lay out in the GEO roadmap I'd build if starting today.

The reason bad AI SEO advice travels so well on LinkedIn is that it sounds actionable and specific ("add FAQs," "use schema," "write long-form") without requiring any investment in the slower, harder work that actually moves the number. That harder work is building genuine source-worthiness: editorial relationships, reference source coverage, community presence, and the kind of original thinking that earns citations not because you asked for them but because you produced something worth citing.

If I had to summarise what separates the AI search programmes that work from the ones that don't: the ones that work start with the question "why would a model cite us?" rather than "how do we hack the algorithm?" The answer to the first question is a strategy. The answer to the second is a tactic with a short shelf life.