SEO Is Not Dead. It Just Stopped Being Enough.
I was in Taxbiex.
It was evening. I was sitting at the marina — the kind of evening where the water is flat and the light is doing something specific that only happens in Malta at that hour — and I was drinking a Cisk. Not because I particularly love Cisk. Because it was what was available and my mood didn't require anything better.
I go to that marina sometimes when I need to think, or when I need to stop thinking. That evening it was the latter. The kind of tired that comes from building something every day and occasionally needing to sit somewhere that doesn't require anything from you.
I opened LinkedIn — which I do despite hating it, as previously documented — and I read something that made me put the Cisk down.
Noa Vernick, an AI Visibility Specialist at AEO Agency Arobis AI, had published a piece on GEO versus SEO versus AEO. I've read dozens of articles on this topic. Most of them are recycled definitions dressed in new acronyms, ending with the same conclusion: do all three, they work together, here's my LinkedIn. Safe. Useless.
Noa's piece was different. She had read the actual studies. Not the summaries. The primary research — Princeton, Ahrefs on 75,000 brands, Semrush on 150,000 citations. And what she found contradicts most of what gets said on this topic, including things I had been repeating myself.
I messaged her that night from the marina. I told her I was going to write my own version of this, with FreeMalta's numbers, and that I'd reference her work. She said yes. This is that piece.
The Three Terms, Quickly
SEO is getting a page to rank in a list of links. The user still has to click. Success is a position.
AEO — Answer Engine Optimization — is getting your content used as the answer instead of as a link. It predates the AI wave by about a decade. Featured snippets, voice assistants, the answer box. Success is owning the extraction. The user usually doesn't click.
GEO — Generative Engine Optimization — is getting your brand surfaced inside AI-generated responses. ChatGPT, Perplexity, Gemini, Claude, Google AI Overviews. The term comes from a 2023 Princeton and IIT Delhi paper that built a 10,000-query benchmark and tested which content changes actually improved inclusion in generated answers. Success is a citation or a mention.
AEO and GEO overlap significantly. Both are about being extracted rather than clicked. The honest difference is that AEO optimizes for a machine choosing one passage, and GEO optimizes for a machine synthesizing many sources and deciding which brands to name. If you're spending real energy arguing which acronym belongs in your internal deck, you're arguing about the wrong thing.
The Finding That Changes Everything
Noa's piece contains a finding that I've been thinking about since the marina.
Semrush ran a study across 5,000 keywords and 150,000 citations, comparing what AI platforms cite against Google's top 10 organic results. The domain-level overlap was high — AI Overviews pulled about 86% of citations from domains already ranking in Google's top 10. Perplexity was higher, over 91%.
Then the URL-level numbers. AI Overviews: 67%. AI Mode: about 35%. ChatGPT: mostly citing pages ranking beyond position 21.
Read those two sets of numbers together and the strategic implication is clear: AI engines mostly trust the same domains Google trusts. They just cite different pages on those domains.
This matters because it collapses two bad arguments simultaneously. The first is "SEO is dead" — it isn't, domain credibility still feeds AI citation directly. The second is "just rank page 1 and you're fine" — you can own position 3 for your main keyword and watch an AI system cite a different page entirely, from position 4, because that page happens to be more extractable.
Same domain. Different page. Your SEO work got you credit with the domain. Your content work determines which page gets cited.
What Actually Predicts AI Mentions
Ahrefs published a correlation analysis across 75,000 brands, measuring what predicts brand visibility in ChatGPT, AI Mode, and AI Overviews.
Branded web mentions: 0.664 to 0.709 correlation across platforms.
Domain Rating: 0.266 to 0.326.
The metric that most agencies build their entire retainer around is roughly a third as predictive as simply being talked about.
YouTube mentions were the strongest signal of all — approximately 0.737. Higher than any web metric measured. This is the finding that surprised Noa most, and it surprised me too. The places where AI models form their opinion of a category are not your blog posts. They are forum threads, video reviews, comparison pages, Wikipedia, community answers. GEO is largely something that happens to you on pages you don't control. SEO and AEO are things you do to pages you own. The structural difference between those two statements explains most failed GEO programs.
FreeMalta's Numbers
FreeMalta is eight months old. Here is what the data looks like.
In the last three months: 1.5 million+ Google Search impressions. 178,000+ Google AI feature impressions across AI Overviews, AI Mode, and Discover. 47,000 Bing impressions. 40,000+ Bing AI Performance impressions across 49 cited pages.
Across AI platforms, FreeMalta content is referenced in ChatGPT, Microsoft Copilot, Gemini, and Google AI Overviews. Zero ad spend. Zero agency. One person.
I'm not citing these numbers to impress anyone. I'm citing them because they are the result of doing specific things consistently, and those things map directly to what the research says actually works.
Every piece of content on FreeMalta is written for extractability — specific claims, clear structure, sourced assertions, concrete data points. Every piece is submitted through IndexNow the moment it publishes, which means it enters Bing's index within minutes and becomes available to Copilot and ChatGPT's web search. The content pipeline runs continuously: News Beast publishes 40+ digests per day, each one a specific, indexed, citable piece of content on a specific topic.
This is not a magic formula. It is the research applied at scale: be specific, be indexed, be mentioned, be consistent. The Ahrefs data says branded mentions predict AI citation better than domain authority does. The way to generate branded mentions is to publish things worth mentioning, consistently, for long enough that the ecosystem starts to notice.
Eight months. That's how long it took to go from zero to being cited across the major AI platforms. Not because we did anything clever. Because we did the right things without stopping.
The On-Page Work That Actually Matters
The original Princeton and IIT Delhi GEO paper — the one that named the field — tested nine optimization methods on 10,000 queries across 25 domains.
The methods that improved AI citation: adding direct quotations, adding statistics, citing authoritative sources, improving fluency. Quotation Addition and Statistics Addition delivered 30 to 40% relative improvement on their visibility metric.
The method that hurt: keyword stuffing. It dropped the metric below baseline. The oldest SEO trick makes you less visible to generative engines. There is something satisfying about that.
The lesson is not complicated. Write content that is easy to quote and hard to doubt. Specific data points. Clear attribution. Authoritative citations. Clean structure that allows a model to extract a passage without ambiguity about what it means.
What the paper also makes clear, and what Noa emphasizes, is that on-page work multiplies presence you already have. It does not create presence from nothing. If the six sources your category gets cited from don't mention you, perfect on-page GEO optimization produces a perfectly invisible page faster. Fix the third-party record first. Then optimize the page.
The Decision Rule
Noa's framework for deciding where to focus: three questions.
Does your category get recommended or searched? If buyers describe a problem and ask for options, AI is now your top of funnel and GEO is the priority. If buyers arrive typing your brand name, keep doing what's working.
What is your baseline? Take twenty to thirty prompts a buyer would type who has never heard of you. Category-level, not brand-level. Run them across ChatGPT, Perplexity, Gemini, and Claude. Count the share of responses that name you. This number is your starting point. Everything else is noise until you have it.
Is your gap coverage or consensus? If you're never mentioned, the model hasn't meaningfully encountered you. The fix is presence in the sources it reads. If you're mentioned but described wrongly, the fix is correcting the third-party record. These are different problems with different solutions. Almost everyone assumes coverage. In practice, most have consensus — which is harder to fix and can't be solved by publishing more content.
What I'd Do This Quarter
Measure first. Get a baseline across a fixed prompt set before changing anything. Every vendor — including the good ones — will claim credit for noise if you don't have a pre-change number.
Fix the record before making more content. Find the sources your category actually gets cited from and get an accurate description of your product into them. This is worth more than thirty new blog posts and considerably less fun, which is why almost nobody does it first.
Keep doing SEO. The domain overlap data is clear: domain credibility still feeds AI citation. Stop expecting page-level rank to deliver AI visibility — the URL-level data says it doesn't. But stop doing SEO means stopping the domain-level work that still matters.
Treat third-party surfaces as owned media. Reddit, YouTube, review platforms, comparison pages. If your plan doesn't name specific threads, videos, and listicles, you don't have a GEO plan. You have a content calendar with new vocabulary.
Remeasure at ninety days. The systems are too noisy for anything shorter to mean much.
Noa's full analysis is worth reading directly — she goes deeper into the methodology and the five studies than this piece does, and her framework for measuring recommendation rate is the right way to track this. You can find it on her LinkedIn, and if your company needs proper AI visibility work, Arobis AI is where she operates.
The Cisk was warm by the time I finished reading. I ordered another one.
FreeMalta is Malta's first AI-native intelligence platform. For more on AI search visibility and web intelligence, visit freemalta.com/hub/web-intelligence.
