- Only 12% of URLs cited by ChatGPT, Gemini and Copilot also appear in Google’s organic top 10 for the same prompt, showing how far AI citations can diverge from traditional rankings.
- AI-referred sessions rose 527% year over year across 19 GA4 properties in the first five months of 2025, but marketers still lack a clean way to measure AI search visibility.
Ranking well in Google used to be the clearest sign that a brand was visible in search. It still matters. But it is no longer the whole visibility problem.
AI search is creating a second discovery layer that does not map cleanly to traditional organic rankings. A page can rank well in Google and still be absent from ChatGPT, Gemini or Copilot answers. Another page can be cited by AI systems even if it has little or no Google visibility.
That split is now showing up in the data. For marketers, the uncomfortable part is not that SEO has become irrelevant. It is that SEO alone no longer explains where a brand appears when users ask AI systems for recommendations, comparisons and answers.
AI Citations Do Not Follow Google Rankings Cleanly
Ahrefs analyzed 15,000 prompts and found that only 12% of URLs cited by ChatGPT, Gemini and Copilot also ranked in Google’s organic top 10 for the same prompt.
That is a large gap. It suggests that traditional rankings and AI citations are connected, but not interchangeable.
ChatGPT appears even less tied to Google’s organic results. In another Ahrefs analysis of 118,931 fan-out queries, only 6.82% of ChatGPT search results appeared in Google’s top 10 for the same query. A separate Ahrefs study found that 28.3% of ChatGPT’s most-cited pages had zero Google organic visibility.
That does not mean ranking is pointless. It means ranking is no longer the whole signal.
Google’s AI Overviews also appear to select sources differently from the classic organic list. A 2026 paper on Google AI Overviews reported that nearly 30% of cited pages did not appear in the co-displayed first-page results, pointing to a source-selection layer that is not identical to Google’s ranking algorithm.
The pattern is becoming clearer: AI systems may use search results, but they do not simply copy the search results page.
Why AI Search Behaves Differently
A traditional Google result page gives users a ranked list of links. AI search systems do something different. They retrieve information, compare sources, compress facts and generate one answer from several inputs.
That changes the role of the source page.
In classic SEO, a page can win by being the best match for one query. In AI search, the system may break a user question into several smaller searches before forming an answer. Ahrefs found that ChatGPT pulls an average of 1.78 search queries per prompt, with 75% of prompts triggering exactly two searches.
That fan-out behavior matters. A page optimized around one keyword may not be enough if the AI system is looking for definitions, comparisons, examples, prices, risks, alternatives and supporting context before deciding what to cite.
This is why the SEO versus GEO debate is often framed too narrowly. The issue is not whether marketers should abandon SEO for generative engine optimization. The issue is that AI answers reward a different kind of source readiness.
What GEO Actually Adds
SEO is still the foundation. It helps search systems crawl a site, understand its topics, assess authority and connect pages through internal structure.
GEO, or generative engine optimization, adds another layer. It focuses on making content easier for AI systems to retrieve, understand and reuse inside generated answers.
That usually means clearer entity relationships, concise definitions, structured sections, original data, named sources, comparison points and updated facts. It also means covering the related questions around a topic, not only the primary keyword.
For example, a traditional SEO page about a software category might target one high-volume phrase. An AI-ready page has to help answer the follow-up questions too: who the tool is for, how it compares to alternatives, what the limitations are, what evidence supports the claim and what a buyer should check before deciding.
GEO is not magic formatting. It is not adding a few FAQs and hoping ChatGPT notices. It is closer to building source material that an AI system can confidently use without guessing.
Technical Access Is Now Part of Visibility
There is also a technical layer many teams still overlook. AI crawlers, AI search systems and browsing agents need some form of access to the content they are expected to understand.
That access is becoming more complicated. Cloudflare moved toward a permission-based model for AI crawlers, asking new domains whether they want to allow or deny AI crawler access. Its AI Crawl Control tools also give site owners more control over which AI crawlers can access content.
For publishers and brands, the trade-off is real. Blocking AI crawlers may reduce unwanted scraping. But it may also reduce visibility in some AI search environments. Leaving everything open may improve access, but it does not guarantee citation, attribution or traffic.
That means technical SEO now needs an AI access check. Teams should review robots.txt, Cloudflare settings, bot rules, server-delivered HTML, schema markup and whether important content is available without relying on client-side rendering.
We covered the broader access question in our piece on Google’s llms.txt guidance and Chrome’s agentic browsing audits.
AI Referral Traffic Is Growing, But It Still Hides the Bigger Shift
AI search traffic is still small for many websites. But the direction is hard to ignore.
Previsible reported that AI-sourced sessions across 19 GA4 properties rose from 17,076 to 107,100 between January and May 2025. That is a 527% year-over-year increase.
EMARKETER projects that 31.3% of the US population will use generative AI search in 2026. That puts ChatGPT, Google AI Overviews, Gemini and Perplexity inside the discovery layer rather than outside it.
The important detail is not only the traffic number. It is where the influence happens.
AI systems often show up in decision-stage journeys. Users ask for best tools, agency recommendations, product comparisons, vendor shortlists, local services and explanations before they click anything. A brand can influence that moment even if the final visit comes later through direct traffic, branded search or another channel.
That makes AI visibility harder to evaluate than normal referral traffic. The click is only one part of the value.
The Measurement Problem Is Still Unsolved
This is where the current GEO market gets messy.
SEO has imperfect but familiar metrics: rankings, impressions, clicks, backlinks, crawl data and conversion paths. AI search does not offer the same clean reporting layer.
Referral traffic tells part of the story. Brand mentions tell another part. Citation tracking can show whether a site appears inside AI answers. But none of those metrics fully captures how often a brand is recommended, ignored, summarized or used without a click.
There is also the problem of instability. AI answers can vary by prompt wording, model, session, location, timing and source availability. One answer may cite a brand. The next may ignore it. A small change in wording can change the source set entirely.
The paper “Don’t Measure Once: Measuring Visibility in AI Search” argues that one-off observations are unreliable because AI answers vary across runs, prompts and time. The authors recommend repeated measurement and treating AI visibility as a distribution rather than a single ranking position.
That is the right mental model. GEO is not a rank tracker with a new label. It is a visibility layer with more variance and less native data.
What Marketers Should Do Next
The practical takeaway is simple: do not stop doing SEO. But do not assume SEO alone will explain AI visibility.
Marketers should start by identifying the prompts and questions that matter most to buyers. Not just keywords, but the actual questions people ask when comparing products, choosing agencies, looking for tools or trying to understand a market.
Then they should check whether their brand appears across ChatGPT, Gemini, Perplexity, Copilot and Google AI Overviews. One check is not enough. The same prompts should be tested repeatedly over time.
On the content side, teams should make important pages easier to extract from. That means clear definitions, direct answers, original data, source links, comparison tables where useful, author credibility and topic clusters that cover more than one search phrase.
On the authority side, brands need to think beyond their own websites. AI systems often cite third-party sources, directories, review sites, media articles, documentation pages, forums and trusted industry resources. If those sources shape the answer, visibility cannot depend only on owned content.
The most exposed businesses are not necessarily the ones with no SEO. They are the ones that built strong Google visibility and stopped there.
Their rankings may still look fine. Their traffic reports may still look familiar. But inside AI-generated answers, they may already be missing from the conversation.
In 2026, ranking is no longer the whole visibility problem.
