Search “geo vs seo” and every page on the first result page has the same shape: a definitions section, a similarities list, a differences table, and “do both”. So does the AI Overview above them, and so did ChatGPT, Claude and Perplexity when we asked them on 21 September 2026. None of them has a number for the thing the title compares. This page has one. We took the 75 search results pages we captured for SEO and GEO queries in September 2026 and measured, page by page, how much of what the AI Overview cites is what ranks.
Is GEO replacing SEO?
No. Google answered this in its own documentation, under a heading that is almost the query. The guide Optimizing your website for generative AI features, last updated 10 July 2026, has a section titled What about “AEO” and “GEO”? and says:
From Google Search’s perspective, optimizing for generative AI search is optimizing for the search experience, and thus still SEO.
The companion page, AI features and your website, gives the eligibility rule in one sentence: to be shown as a supporting link in AI Overviews or AI Mode, “a page must be indexed and eligible to be shown in Google Search with a snippet”. There are, it says, “no additional technical requirements”.
That is Google’s side. The measured side, below, agrees with it and adds the part Google does not say: how often the ranking page is the cited page.
What the two words mean, from the documents that defined them
Generative engine optimization is a 2023 term. The paper that coined it, GEO: Generative Engine Optimization by Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan and Deshpande, submitted in November 2023 and accepted to KDD 2024, defined it against a category the authors called generative engines, which they expected to be “rapidly replacing traditional search engines like Google and Bing”. The paper’s headline result, that GEO “can boost visibility by up to 40% in generative engine responses”, is a benchmark score on the authors’ own dataset, GEO-bench. It is not a measurement of Google, and any page quoting “40%” as a fact about search is quoting a benchmark.
Search engine optimization needs no definition here. What matters is the relation Google draws between the two, quoted above, and what Wikipedia records about the vocabulary: as of early 2026, “no consensus definition distinguishing these terms had been established in the academic literature”, and GEO, AEO, LLMO and AIO “are frequently used interchangeably”. Wikipedia’s own article on the term has carried a notice since August 2026 that it “may document a neologism”. Answer engine optimization is the same idea under another name; this page uses GEO because that is the word in the query.
The overlap, measured
The method is the one from our AI Overviews study, re-read per results page. We captured 76 US SERPs for SEO, GEO and AI-search queries between 4 and 17 September 2026, in a browser with personalisation off. 65 of them carried an AI Overview with at least one citation. For each of those 65 we compared the domains the overview cited with the domains in the organic top ten on the same page.
| On the 65 AI Overviews we captured | Result |
|---|---|
| Citations, counted as unique domains per page | 356 |
| Citations whose domain ranks in the top ten on the same page | 163 — 46% |
| Citations whose domain does not | 193 — 54% |
| Pages where the #1 organic result is cited | 25 — 38% |
| Pages where any of the top three is cited | 47 — 72% |
| Pages where the overview cites nothing from the top ten | 10 — 15% |
| Overlap between the cited set and the top ten, as a share of their union | 0.25 mean, 0.25 median |
Read as a rule of thumb: ranking first on Google gets a page cited in the AI Overview a little more than one time in three. Ranking anywhere in the top three gets it cited nearly three times in four. One overview in seven cites nobody from page one at all. And the two lists, what ranks and what is cited, overlap by about a quarter.
That is neither “GEO is SEO” nor “GEO is a separate discipline”. It is the size of the gap between the two scoreboards, on the queries we captured.
One competitor number exists on the same question, and it agrees. Writesonic, whose page on this query is cited by the AI Overview and does not rank, reports that in its “analysis of over 1 million AI Overviews”, 40.58% of citations “come from Google’s top 10 search results”. Theirs is measured per URL over a million overviews; ours is per domain over 65, which should make ours higher, and it is. Roughly half of what an overview cites is on page one, on two samples four orders of magnitude apart.
What the citations that do not rank actually are
The 54% that does not rank is not a second ranking system rewarding a second checklist. In September 2026 we fetched the homepage of every domain in the corpus, 276 in all, and grouped them by whether they were cited, ranked, or both. The structured-data study published the census; here is the reading of it that belongs on this page.
| Domains in the corpus | Fetched | JSON-LD present | Organization or Person node | sameAs |
|---|---|---|---|---|
| Cited by an overview, never in a top ten | 46 of 56 | 76% | 63% | 48% |
| In a top ten, never cited | 128 of 139 | 88% | 76% | 55% |
| Both | 74 of 81 | 91% | 80% | 55% |
The domains the overview picks and Google does not rank carry less of the GEO checklist than the domains Google ranks and the overview ignores: twelve points less structured data, thirteen points less entity markup. Whatever separates the two groups, it is not the technical signals the checklists sell.
What does separate them is visible in the list of names. Nine of the 56 cited-only domains are platforms where a person, not a site, wrote the thing: LinkedIn, Medium, YouTube, Reddit, Quora, the two app stores. On the “geo vs seo” page itself, the overview’s eight citations split exactly in half: four rank (Informa TechTarget, Contentful, Semrush, seo.com) and four do not — a tool vendor’s blog, a LinkedIn post by a named practitioner, a Medium post and a YouTube video. The overview will cite an individual’s post on a platform whose homepage the ranking system would never surface for the query.
The mechanism Google gives, and why it fits
Google describes how an overview finds pages that do not rank, and the description reconciles “still SEO” with “54% do not rank” without a second ranking system. From AI features and your website:
Both AI Overviews and AI Mode may use a “query fan-out” technique — issuing multiple related searches across subtopics and data sources — to develop a response. While responses are being generated, our advanced models identify more supporting web pages, allowing us to display a wider and more diverse set of helpful links associated with the response than with a classic web search.
An overview runs related searches. A domain that ranks for a neighbouring query can be cited on this one without ranking on it. Same ranking system, wider query set. That is the account Google gives, and it fits the numbers: a quarter of overlap on the query itself, and the rest drawn from the queries next door. It is the mechanism Google describes, not a demonstration that it is the whole mechanism.
What the assistants said
On 21 September 2026 we asked ChatGPT, Claude and Perplexity the query verbatim: “geo vs seo”. All three answered with a differences table, all three said GEO does not replace SEO, and none gave a number. Claude and ChatGPT cited no sources. Perplexity cited two: Semrush’s page, which ranks third on Google for the query and is cited by the AI Overview, and a page from impactplus.com, which is on neither. Perplexity’s table had Semrush’s rows.
For our previous question, llms.txt, Perplexity had cited thirteen sources and none of them was on Google’s first page. Zero of thirteen, then one of two. Two questions are two observations, not a rate, and the honest statement is the narrow one: an assistant’s source set can overlap Google’s page fully, partly or not at all on consecutive questions. “GEO is a different game on assistants” is neither confirmed nor refuted by what we have.
One disagreement is worth noticing. All three assistants filed Google’s AI Overviews under GEO. Google files them under SEO. The argument about what the word covers is between the assistants and Google.
What this means for the work
Elivate SEO’s reading of the numbers above, for a site that wants to be cited:
- Rank first, then worry about citation. 72% of overviews cite something from the top three. A page that is nowhere on Google is rarely cited by Google’s overview, and Google’s eligibility rule says why.
- Do not buy a second checklist. The cited-only domains carry less structured data than the ranked-only ones, not more. Google’s guide names the things to ignore: “chunking” content, “creating unnecessary AI text files (like llms.txt)”, “pursuing inauthentic mentions”.
- Expect a quarter, not a match. Ranking first is cited one time in three. The rest of the overview comes from neighbouring queries, so a site’s coverage of a topic matters more to citation than its position on one query.
- Be present where a person can write. Nine of the 56 cited-only domains are platforms. That is a person’s LinkedIn post or a founder’s Medium piece being cited, not a company’s homepage. On ecommerce searches, a quarter of AI Overview citations go to platforms like these.
- Measure the second scoreboard separately. Citation is not position. Our free audit reports what a crawler and a parser get from a page; whether an overview cites it is read from Search Console’s AI features report, a page at a time.
The GEO service is the work in that list, done for a site rather than described.
What this does not show
- Why a cited domain does not rank. Fan-out is Google’s stated mechanism and fits the data; nothing here shows it is the only one.
- A rate for the web. 76 queries about SEO, GEO and AI search, US market, two days in September 2026. Cooking, medicine and retail may differ.
- The same measurement as Writesonic’s. Theirs is per URL over a million overviews, ours per domain over 65. They agree in direction and size, which is all that is claimed.
- What assistants do in general. Two questions, two observations.
- That structured data hurts. It does not separate the groups. That is a different sentence, and it is the one the census supports.
- Every page on the SERP. Contentful’s page, second on Google, rate-limited our fetcher twice and was not read; nor were the LinkedIn post, the Medium post or the video the overview cites.
Why we wrote it
Because every page ranking for “geo vs seo” is published by a company with a GEO product, and every one of them compares the two without measuring anything. Google’s answer to the question exists, under a heading that is nearly the query, and does not rank for it. We ran the numbers, quoted the documents, and wrote the comparison with the one thing the others lack. The measurement repeats in six months, when the September 2026 capture becomes a before.