After Google’s AI guide: is GEO still a discipline—or just good SEO?

On May 15, Google published its AI Optimization Guide and told the industry that GEO and AEO aren’t new disciplines—they’re SEO with fresh labels. My inbox that week looked like a crisis hotline. Here’s why I think Google is half right, and why the other half is the whole product.

The headline that rattled German agencies

The guide’s thesis is one clean sentence: “From Google Search’s perspective, optimizing for generative AI search is optimizing for the search experience, and thus still SEO.” Around that, Google added a “mythbusting” section that read, to a lot of practitioners, like a demolition notice. No need for llms.txt. No special AI files or Markdown—Google says it ignores them, and that they “will neither harm nor help” your visibility. Structured data “isn’t required for generative AI search, and there’s no special schema.org markup you need to add.”

In the DACH market, that landed hard. The German-speaking agency scene had spent a year building a vocabulary around this—AEO-Optimierung, GEO-Audits, KI-Sichtbarkeit—and pricing services against it. Google had just told their clients the whole category was jargon. I got the emails. “Do we refund the retainer?” “Was the audit a scam?” “Did we just build a business on a myth?”

So let me answer the question a marketing lead actually types into an assistant: is GEO worth investing in after this guide? Short version—yes, but not for the reasons the loudest people on either side are shouting. Let me show my work.

What Google gets right

I’m not going to strawman this. The core of Google’s argument is correct, and dismissing it makes you look like someone protecting a product line.

First, the rank-citation link is real. A July 2025 Ahrefs study of 1.9 million AI Overview citations found that 76% of cited pages ranked in the top 10 organic results, and that pages ranking nowhere in the top 100 were the exception. If your site can’t earn an organic top-10 spot, you are starting a foot race with your shoelaces tied. Foundational SEO is table stakes, full stop.

Second, the guide is right that you can’t game your way in with junk. Its advice to build non-commodity content—the kind that “goes beyond common knowledge” rather than recycling what a model could already generate—is sound. Thin content dressed in schema does not get cited. I’ve watched it fail.

So when Google says quality content and a clean technical structure are prerequisites, I agree completely. My disagreement isn’t that Google is wrong. It’s that Google answered a narrower question than the one my clients are asking—and then let the answer sound universal.

Gap 1—URL citation is not entity recommendation

The 76% figure measures one thing: when Google’s AI Overviews cite a URL, how often does that URL also rank in the top 10? It’s a question about which pages get linked. It says nothing about which brand a model names when a user asks an open-ended commercial question with no URL in sight.

Those are different queries with different physics. “What is generative engine optimization” is informational—the model summarizes a well-ranked explainer and cites it. “Who’s the best GEO audit provider in Berlin” is commercial and navigational-adjacent—the model assembles a recommendation from what it knows about entities: who they are, what category they occupy, whether third parties vouch for them. Rank is one input among several, and often not the deciding one.

There’s a second problem with leaning on 76%: it’s already gone. A March 2026 re-analysis of 863,000 keyword SERPs and 4 million AI Overview URLs found the overlap had fallen from 76% to 37.9%—roughly halved—coinciding with Google moving AI Overviews to Gemini 3. Even inside Google’s own product, rank explains less of the citation picture every quarter. Building a 2026 strategy on a 2025 correlation is building on sand that’s actively washing out.

Gap 2—Google’s guide is about Google

This is the gap the guide never names. As the Frase.io team put it, the document “says nothing about how ChatGPT, Perplexity, Claude, or Gemini’s standalone app decide what to cite.” That’s not a criticism of Google—it’s Google Search Central documentation, of course it’s about Google. But readers keep generalizing “you don’t need structured data” into “you don’t need structured data anywhere,” and that inference is false.

The numbers make the scope problem concrete. An August 2025 Ahrefs study of 15,000 long-tail queries found that only 12% of URLs cited by ChatGPT, Gemini, and Copilot appear in Google’s top 10. Perplexity—the platform most correlated with Google rank—still pulls nearly 70% of its citations from pages outside the top 10, with only 28.6% landing in it. Optimize purely for Google’s ranking model and you’ve covered a minority of the surface where AI now recommends brands.

The platforms select differently on purpose. As one analysis frames it: “Google ranks pages for click satisfaction. Perplexity selects sources for extraction quality—whether the system can quote the page accurately without distortion.” A page “can rank well in Google and never be cited in Perplexity if its claims are buried, ambiguous, or structurally difficult to extract.” Same page, two verdicts, because the tests are different. Winning one guarantees nothing about the other.

Gap 3—“do good SEO” is not a checklist

Here’s the gap that actually costs money. Strip the guide down to its actionable core and you get: continue prioritizing foundational SEO, build a clear technical structure, create unique, valuable content. Every site owner has heard that for 15 years. None of them can do it on Monday morning, because it isn’t a task—it’s a mood.

GEO’s entire value is that it converts that fog into named, auditable, fixable signals. A real GEO audit doesn’t sprinkle AI words on an SEO report. In the words of one public methodology, “it checks whether a model can find, parse, and quote you, which is a different question than whether you rank.” It scores concrete things: whether content is chunked into liftable passages, whether schema exists and is connected with stable identifiers, whether the brand’s entity story is legible, and whether the site is already being cited.

Make it tangible. A client of ours ranked #2 for their core service term and got cited by no AI assistant we tested. The audit found the mechanism in minutes: their H2s were phrased as questions—“How much does a migration cost?”—but the paragraph underneath opened with a marketing wind-up, not the answer. A model scanning for a liftable passage found nothing to lift. We rewrote the first sentence to answer the question directly, added a self-contained price range, and left the ranking untouched. Citations followed. “Do good SEO” would never have surfaced that. A passage-structure check did, in one line of an audit.

What the audit scores actually show

Let me put real numbers down. These are anonymized scores from GEO audits we’ve run for German SMBs, on a 100-point scale that follows the public four-dimension method—chunk readiness, schema coverage, entity clarity, citation presence—each scored 0, 1, or 2 and weighted to 100. A site sitting at a flat “1” across the board—mediocre but present on every axis—lands at 50. Comparable organizations that have done deliberate work cluster in the 45–70 range.

The sites I keep opening score 32, 29, 28. And here’s the part that matters for this debate: several of them ranked fine organically. Rank did not predict the score, and it did not protect it. Three failure categories showed up again and again, none of which a rank check would catch:

• Disconnected schema. Either no JSON-LD at all, or two or three schema blocks that never reference each other with stable @id values—so a machine can’t tell that the Organization, the service, and the author are the same story.
• Questions without answers. H2s phrased as questions where the first sentence underneath doesn’t answer them. The passage is unliftable, so it’s uncitable.
• No entity self-declaration. The brand never states its category, its proof, or its boundaries in language a model can repeat. There’s nothing for the model to recommend, because the site never said what it is.

A #1 ranking does nothing about any of these. That’s not a gap in someone’s SEO effort—it’s a gap in what SEO measures. To be clear about provenance: these three scores are from our audit practice, not a third-party study. I’m reporting what I see, not publishing a survey.

The DACH market has already voted

While Google was calling the category jargon, the German-speaking market was doing the opposite. Using DataForSEO figures collected May 25, 2026, comparing the May–July 2025 quarter against February–April 2026, German search volume for “aeo optimierung” rose 1,300%, “geo audit” rose 630%, and “ki sichtbarkeit” rose 364%.

Let me be honest about scale, because the percentages flatter the story: these are specialist terms with small absolute volumes. “Aeo optimierung” grew from roughly 3 to 47 monthly searches; “geo audit” from 33 to 243. This isn’t a mass-market wave. It’s a professional vocabulary being built in real time—the people whose job is to get brands cited are the ones searching.

When practitioners surge toward a topic the market leader just called unnecessary, it’s one of two things: mass misunderstanding, or a real need Google’s guide didn’t serve. The audit data tells me it’s the second. People aren’t searching “geo audit” because they’re confused about SEO. They’re searching it because they ranked, checked their AI visibility, and found nothing—and “do good SEO” didn’t tell them why.

GEO isn’t a competing discipline—it’s the layer SEO leaves unfinished

So drop the binary. “GEO vs. SEO” is a false choice, and both the hype merchants and Google are selling a version of it. Here’s the framing that survives contact with real audits.

SEO gets a page ranked: it earns the authority, the links, the technical health, the quality that makes the page worth citing at all. GEO is the structured layer on top—the specific practices that make an already-good page machine-readable: connected entity schema, passage formatting that answers the question in the first sentence, review and third-party presence, and freshness metadata. A page can rank #1 and fail all four. That failure is exactly what the audit measures, and closing it is the product.

Whether you rank and whether a model can find, parse, and quote you are two different questions. SEO answers the first. GEO is the discipline of answering the second on purpose.

Is that “still SEO”? Call it what you like. The label is a semantic fight. The work—the four checks a ranked page still fails—is real, nameable, and not covered by “create valuable content.” Google is right that it’s not magic. It’s just carpentry SEO never framed as its own job.

The verdict, and five things to do this week

Is GEO worth investing in after Google’s guide? Yes—as the structured layer, not as a replacement for ranking. If you’re not in the organic top 10, fix that first; the guide is right about foundations. If you already rank and you’re still absent from AI answers, GEO is precisely the diagnosis you need. Here are five moves that each map to an auditable signal, cheapest first.

1. Add connected Organization JSON-LD. Emit Organization schema with sameAs links to your real LinkedIn, GitHub, and Crunchbase profiles, and give it a stable @id that your other schema blocks reference. This is the entity story a model reads to know who you are. A 30-minute fix on most sites.
2. Rewrite one service-page section as a self-contained answer passage. Take one H2 phrased as a question and make the very first sentence answer it in about 150 words—no wind-up, no “in today’s landscape.” Include the concrete number a reader would want. That’s the unit a model lifts.
3. Claim and complete a Trustpilot profile. Review presence is a third-party signal SEO doesn’t touch. Assistants weigh what others say about you, not only what you say about yourself.
4. Add dateModified to every service page and keep it honest. Perplexity’s median cited URL is about 32.5 days old versus 108 for Google—a 3.3× freshness premium. Stale-looking pages get skipped on the platforms that move fastest.
5. Run a baseline AI visibility audit. You can’t fix a score you haven’t measured, and rank won’t tell you. Establish where you actually stand across ChatGPT, Perplexity, Gemini, and Google AI Overviews, then track it—a tracker like Cited watches how these models describe you and your competitors over time and turns the gaps into to-dos.

The debate over whether GEO is “really” SEO is a distraction from the work. A ranked page that no model can quote is leaving citations on the table, and “just do good SEO” won’t find the leak. If you want the score for your own site—the same four-dimension methodology I’ve described here—that’s exactly what our AI Visibility Audit delivers: a baseline, the specific failures, and the fixes ranked by effort. Starting at 99 €. Measure first, then argue about what to call it.