Is GEO Just SEO? The Honest 80/20 Answer

The 80% that really is just SEO

If you strip the acronyms away, most of what it takes to get recommended by an LLM is by-the-book SEO: authority, clean structure, schema markup, answer-first content, and narrow topical depth. That’s it. Those are the same fundamentals that have moved rankings for a decade. So let me say the uncomfortable part out loud: if someone is selling you GEO services and the deliverables are blog posts, FAQ sections, and JSON-LD, you are getting ripped off — you’re buying SEO at a markup. There’s nothing wrong with those deliverables. They work. But call them what they are, and don’t pay a “generative” premium for search fundamentals you could get anywhere. If you want the baseline, our guide to what generative engine optimization actually is and how to rank in Google’s AI Overviews cover it without the markup.

Why the other 20% is genuinely new

Here’s the shift that the “it’s all just SEO” crowd misses. Before large language models, search engines ranked pages, factoring in the domain they lived on. Now, models rank identities — a picture of who you are, assembled from your entire web presence. That’s not a marketing distinction. It changes what you optimize, and it means a chunk of AI search visibility can’t be fixed on your own website at all. Two stories from real client work make it concrete.

Proof #1: the client whose identity got rewritten

One of my clients, a New York actress, ranks well for the terms I target for her. Recently, Google’s AI Overview started associating her with a casting agency on the other side of the country. She’s never lived there. She’s never been cast there. I checked her website thoroughly — every page, every bit of schema — and there was absolutely no signal pointing anywhere near that organization. Nothing on the page was wrong. After some digging, I found the culprit: a handful of old Facebook comments loosely linking her to it. No amount of on-page optimization fixes that, because the page isn’t the problem. The AI Overview had built her identity away from her website, using third-party surfaces she didn’t control. That’s the 20%. That’s where the real work is. This is what practitioners mean by “entity confusion,” and it’s why controlling what shows up for your name now reaches well past your own site.

Proof #2: an llms.txt file that was read and acted on

People love to argue that llms.txt — the file that formats a site for AI crawlers — does nothing, that there’s no evidence it moves anything. I can’t fully settle that debate, but I can tell you what I saw. While researching a large fintech company, I clicked a link from Google’s AI Overview and hit the dreaded 404. I traced it back: the broken link was coming straight out of that company’s llms.txt file. I can’t prove the file improved their visibility. What I can prove is that it was being read and acted on — a stale entry in their llms.txt was actively sending real users to a dead page. So when someone tells you these files are inert, know that at minimum, the machines are reading them. Treat them like production infrastructure, not a curiosity.

The real playbook: manage your identity, not just your pages

My advice to everyone reading this: don’t just optimize your pages for AI — manage your identity. In practice that means keeping consistent information across the third-party surfaces the models learn from. These systems train on and cross-reference sites like Wikidata, Crunchbase, LinkedIn, industry directories, and news sources. The models weigh what others say about you just as much as — if not more than — what you say about yourself. If your name, your role, your location, and your work are contradictory across those surfaces, the model resolves the contradiction on its own, and you may not like the answer. This is the part that isn’t SEO, and it’s the part almost nobody is selling well yet.

A metric traditional tools don’t report

Through SEMPITE I track how often models name a brand versus how often they cite its domain — and you’d be surprised, both at the results and at the fact that traditional SEO tools don’t report these metrics at all. A brand can be named constantly and cited rarely, or the reverse; those are two different visibility problems with two different fixes. If your reporting only counts links and rankings, you are blind to half of how AI search actually describes you.

One honest caveat before you spend a dollar

Being surfaced doesn’t equate to more traffic. I had an article jump to 528 impressions in a single week and pull exactly one click. So be skeptical of anyone quoting you a tidy ROI on a GEO service — visibility and revenue are related, but they are not the same number, and honest practitioners will tell you so. The 20% that’s real is worth doing. Just make sure you’re paying for the identity work, not for search fundamentals dressed up in a new acronym. I’m happy to show the AI Overview screenshots and the citation-probe data behind everything above.