GEO myths vs what the research actually says

What the research actually measures

The most-cited GEO study tested content edits across thousands of queries and measured a metric called Position-Adjusted Word Count: roughly, how much of your page's text the engine pulls into its generated answer, weighted by where it appears.

This matters because of what it is NOT. It is not 'odds of being cited', it is not clicks, and it is not whether a human saw you or bought anything. A page can score well on this metric and still be left out of the actual buying journey. When people say a tactic improves 'visibility', this is usually the number they mean, and it is narrower than it sounds.

One more caveat the loud posts skip: the study ran against a system built to resemble Bing Chat, not live ChatGPT. Treat the results as directional, not as a guarantee about any specific assistant today.

What genuinely works (verified)

Three content changes measurably increased how much of a page's text the model reproduced in its answer:

• Quoting named experts was the single strongest lever (around +40% on the visibility metric).
• Adding concrete statistics helped (around +30%).
• Citing credible sources helped (around +30%).
• Adjacent, well-supported practices: answer the question in the first sentence, use question-shaped headings, and add a short FAQ. These map to how people actually prompt.

The myths to stop repeating

These circulate constantly and are not supported by the primary study:

• '3 or more stats doubles your citation rate.' Not in the paper. The real result is a ~30% lift on a text-share metric, not a doubling of citations.
• 'Comparison tables and listicles get cited 3 to 5x more.' The study did not test list vs prose as a variable. This number has no primary source.
• 'SEO is dead, GEO replaces it.' False, and the most harmful one. The paper explicitly did not touch metadata like backlinks, so it cannot claim they don't matter.
• 'You can't influence AI recommendations, it's a black box.' Half true: you can't control it, but the inputs (what your pages say you're for, and whether credible third parties corroborate it) are workable.

How AI seems to pick who it recommends

Beyond the lab metric, here is a working model based on how the engines retrieve, stated as a hypothesis rather than a proven law. Your own pages mostly define what you ARE (your attributes, who it's for), which gets you into the consideration set. Between comparable options, the deciding signal tends to be earned, third-party corroboration: reviews, roundups, community threads.

There's a further layer. When a buyer adds constraints (budget, team size, use case), the option that survives tends to be the one whose third-party evidence consistently frames it for that exact role. If three sources position you three different ways, the model gets a muddy signal. Consistency of positioning across sources may matter as much as raw volume of mentions. This is observation, not settled science, but it lines up with how retrieval works.

Why SEO still feeds AI citations

The 'different game' framing makes people stop doing things that still work. In practice the channels overlap: ChatGPT's search leans on Bing's index, and Perplexity leans heavily on Reddit. So your existing SEO and off-site presence still feed whether AI names you.

The honest takeaway: GEO is additive to SEO, not a replacement. Backlinks, rankings and third-party coverage remain part of the picture because they shape what the retrieval layer can find about you.

Frequently asked questions

Is the GEO study about live ChatGPT? No. It ran against a Bing-Chat-like system on a benchmark, so treat it as directional, not a measurement of any current assistant.

Does 'visibility' mean traffic? No. The study's metric measures how much of your text appears in the answer, not whether a human saw it or clicked.

So is GEO advice useless? No. The directional advice (cite sources, add real stats, quote named experts, answer directly) is supported. The fabricated multipliers are what to ignore.

What's the fastest honest check? Ask Perplexity or ChatGPT the exact questions your buyers would, without naming your brand, and see who gets listed. Run it a few times, the answers vary.