GEO adds to SEO; it does not take its place. When an AI engine answers from the web, it starts by retrieving documents, and a document's relevance and its place in the context given to the model remain the most reproducible levers known so far. The change comes after that step: the result is no longer a ranked page but a brand named in a text that shifts from one request to the next, often on the word of third parties.
For a CMO, the rule is simple: do not fund GEO by cutting SEO. Add the measurement SEO cannot give you, then allocate with it.
The market tells a different story, built on acronyms and a 40% figure taken out of context. As of 6 November 2026, Olivier Martinez's review of 45 studies, a preprint published in July, finds no technique whose causal effect on organic discovery or downstream behavior holds over time and across engines.
What is GEO?
GEO (Generative Engine Optimization) is the set of practices meant to increase a brand's or a piece of content's presence, citation or influence in the answers of AI engines: ChatGPT, Gemini, Perplexity, Claude, Copilot, Google AI Overviews and Google AI Mode.
The term comes from a paper by Pranjal Aggarwal and co-authors, submitted in November 2023 and presented at KDD in August 2024.
It covers two results that should not be added together. A brand is named when the answer describes or recommends it. A source is cited when the engine points to its page. Either can happen alone: your brand can be recommended on the strength of a comparison you did not write, and your page can be cited in an answer that recommends a competitor.
Do GEO, AEO and AIO mean the same thing?
In everyday use, nearly. At Google and Microsoft, no, and those two disagree with each other. Nobody has settled these terms: Wikipedia's GEO entry, read on 6 November 2026, treats them as names for one concept and notes that no consensus definition exists.
| Acronym | Stands for | What it covers | Watch out for |
|---|---|---|---|
| GEO | Generative Engine Optimization | Visibility in generated answers | The only one of the three with a documented origin (2023) |
| AEO | Answer Engine Optimization, or "Agentic" at Microsoft | Often the same as GEO. At Microsoft, content meant for assistants and agents | Older, first tied to featured snippets and voice. No reliable first use could be established for this guide |
| AIO | Artificial Intelligence Optimization, or AI Overviews | Either a practice or a Google product | "The impact of AIO" does not say which. Ask for the full term |
Sources: Aggarwal et al.; Microsoft Advertising; Darkroom, an agency, on AEO's origin (November 2026); Wikipedia on AIO as a practice; Martinez on AIO as AI Overviews.
Three readings coexist:
- Google: it is still SEO. Its guide to generative AI features says "optimizing for generative AI search is optimizing for the search experience, and thus still SEO". That holds for its own products, and Google has every reason to discourage manipulation of its results.
- Microsoft: three layers. In its January 2026 guide, centered on commerce and summarized by Search Engine Journal, SEO gets the product found, AEO helps the AI explain it, and GEO helps the AI trust and recommend it.
- The market: synonyms. The agency Onely asked five AI engines on 16 October 2026, once each. They diverged, and AI Overview filed AI Overviews under GEO while AI Mode filed them under AEO. It remains an anecdote.
The acronym matters less than three questions for whoever uses it: which engines, which metric, what outcome?
What actually changes compared with SEO?
The object you measure, how stable it is, who carries the message and what data you get.
| SEO | GEO | |
|---|---|---|
| Result | A page ranked for a query | A brand named or a source cited in a text the engine writes |
| Metric | Position, impressions, clicks | Frequency of appearance across a set of prompts |
| Stability | A position can be recorded and compared | The answer changes from one request to the next |
| Observed demand | Real queries in Search Console | Prompts you choose |
| Who carries the message | Your pages first | Often third parties |
| Google's data | Impressions and clicks | Impressions only in the generative AI report |
Stability. A team from Maastricht, Utrecht and Zurich universities submitted 117 shopping queries three times each, mostly in English and from the Netherlands (preprint, September 2026). Between repetitions, the pages in common average just 18.1% on ChatGPT, 28.4% on Gemini and 41.4% on AI Overviews. Yet whether the answer names at least one brand or product flips in only 11 to 13% of queries, while the recommended products barely overlap (0.178 on ChatGPT, on a scale of 0 to 1). An engine is fairly consistent about whether it recommends, far less about whom it recommends and which pages it uses.
Rank holds up even worse. SparkToro and Gumshoe (2,961 responses in late 2025, not peer reviewed, with a possible conflict of interest flagged by its author) put the odds of getting the same brand list in the same order twice at roughly 1 in 1,000. Carrying SEO's average position into a GEO dashboard means tracking noise. Frequency of appearance across many prompts looks more usable, says the same author, who leaves the question open.
Who carries the message. On US queries, University of Toronto researchers (September 2025) found 81.9% earned, third-party sources from a web-enabled GPT in automotive, against 45.1% in Google's results. One model, queried by API, so trust the direction of the gap rather than the decimal. Part of GEO falls outside the SEO team's remit.
The data. Search Console's generative AI report, open to all sites since 31 August 2026, announces impressions and no clicks. Clicks say little anyway: in Pew's panel (900 US adults, March 2025), 1% of visits to a page with an AI summary led to a click on a link in that summary. A mention may therefore shape a decision without leaving a trace, or shape nothing at all. Neither has been demonstrated.
Where do GEO and SEO overlap?
At the step where the engine picks what to read, which is the step that weighs most. Google describes this for its own features: they rest on its ranking systems, ground answers in the index and run several related searches, known as fan-out. Its documentation adds that "all existing SEO fundamentals continue to be worthwhile", with no additional requirements.
The most direct independent test points the same way. C-SEO Bench (NeurIPS 2025) simulates engines on more than 1,900 English queries. On one of the models tested, moving a document to the top of the model's context gains 0.70 to 2.77 citation ranks depending on the domain, against 0.36 at best for any content optimization method, and 0.18 for the best rewrite. The size varies by model; the direction does not. The authors call these methods "a complement and not a replacement for traditional SEO".
One limit: position in the context is not position in Google. The study tests no production engine, and the step from one to the other still has to be established engine by engine. Across 61 sites tracked by SEOWorks from September 2025 to August 2026, growth in AI referrals moves with organic growth (r = +0.534 after controlling for growth in the rest of the traffic): a weaker signal, only a correlation, but a consistent one.
So the shared ground is concrete: AI bot settings decided engine by engine (at OpenAI, OAI-SearchBot for search and GPTBot for training are controlled separately), proof readable in the HTML text, and pages that answer a real question. Your SEO team already does this work.
Does Google position still matter?
Yes, though not necessarily the one your reports show.
Ahrefs placed 76% of AI Overview citations in the query's top 10 in July 2025, then 37.9% in March 2026 (863,000 results pages). Ahrefs warns the two are not strictly comparable, since its method and Google's model both changed. Its hypothesis: cited pages increasingly come from fan-out sub-queries, which Google does not publish.
A low overlap therefore does not prove position has stopped mattering. It may mean position matters on queries you do not track. If so, the work changes: covering the sub-questions of a topic rather than defending a few keywords. Without building one page per variation to manipulate answers, which Google treats as scaled content abuse.
Dependence on Google also varies by engine. Across 3,311 head terms (Ahrefs, September 2025, so due for a recheck), 65.07% of URLs cited by Perplexity sat in Google's top 10, against 10% for ChatGPT.
What falls is the value of a click on part of the query set. Agarwal and Sen's experiment, known here through PPC Land's report (1,065 US desktop users, January to February 2026, working paper), randomly hides AI Overviews: where the block appears, outbound organic clicks fall 39.8%, driven by informational queries. On navigational and transactional queries the effect cannot be told apart from zero, with noisier estimates. In France, where these features opened on 22 July 2026, the first figures remain fragile: Ahrefs reports a median 8.8% click decline across 963 domains, over nine days only.
Search still carries most of the flow. Ahrefs counts 27.55% of visits from search against 0.46% from AI, across 118,671 sites in October 2026. Among French generative AI users concerned by this use, 59% prefer a search engine to look something up, against 28% for an AI (Arcep and CREDOC, June 2025 survey, before AI Overviews arrived there). Traffic does not capture influence, and these balances can move. Today they do not justify emptying the SEO budget.
Has GEO proven itself?
Not yet, and the most quoted figure does not say what people claim.
The "40%" comes from the founding paper. Martinez's review points out that it is a relative maximum, on one metric, in a simulator where five documents already sit in the context. The same paper's "real-world" test on Perplexity supplied source text as file uploads, on 200 samples, so it says nothing about organic retrieval. And the gain went to poorly placed sources: adding citations meant +115.1% for the fifth-ranked site and -30.3% for the first. A challenger's recipe at best.
Later work is less encouraging:
- C-SEO Bench finds significant improvement in only 3 of 54 cases. Adding statistics lowers the document's rank in 19 of 24 settings, and gains erode as competitors adopt the same method.
- LIVE-GEO (preprint, 31 October 2026, 80 fictional brands) sums up: "part of the aggregate GEO gain survives on the live Web, but most brand-level gains do not".
- Watanabe and Nakayashiki (June 2026, one domain) see ChatGPT referrals multiply 5.7 times across the domain, but already 3.5 times on untreated pages. Once the trend is isolated, 1.82 remains, but a conservative placebo test does not validate the effect (p = 0.16). The authors call it suggestive, not conclusive.
On the link from citations to clicks or revenue, Martinez rates confidence as "very low". Missing proof does not demonstrate uselessness. It means no recipe is established, and that any promised figure invites three questions: measured where, on what, against which control group? The evidence level of each tactic has its own guide.
What does nobody know yet?
Three points the studies read for this guide do not settle:
- What people really ask. No source consulted gives the real distribution of prompts. When SparkToro had volunteers write 142 prompts for one need, their mean similarity was 0.081. All tracking therefore rests on a chosen sample whose representativeness nobody knows.
- What a mention earns. No published incrementality test on revenue was found. A separate guide covers how to build one.
- Personalization and mobile. No peer-reviewed study was found on how memory, history or location affect brand recommendations, and the strongest experiment cited here is desktop only.
What should you decide now?
Five decisions, in this order.
- Take SEO out of the trade-off. Cutting it to fund GEO weakens the better documented of the two levers. Set the envelope with the GEO, SEO and paid budget in front of you.
- Measure before you optimize. Start from decision questions that do not name your brand. The volumes you need have their own guide.
- Read a frequency over several weeks. A University of St. Gallen study (April 2026, German-speaking Swiss markets) sees about 65% of sources change from one day to the next and recommends rolling windows of two to four weeks. Watch for false signals too.
- Give the topic to two functions. SEO keeps access and retrieval. PR and partnerships handle third parties, starting from the sources each engine cites.
- Separate reach from results. Visibility says where you appear; AI traffic in GA4 gives a last-click floor. Keep them apart in reporting.
Steps 2 and 3 are hard to keep up by hand, because the same reading has to come back across several engines and over time. That is the job of Nanga's GEO module: prompts grouped into topics, followed run after run (weekly by default, daily as an option), with Recommendation Visibility, "how often your brand naturally appears when a user expresses intent to an AI". In the same system, the SEO module reads your organic search and Brand Radar the demand on your name in Google, so you read both faces of demand before moving any budget.
To see what AI engines say about your brand, on your own prompts: request a demo.
Frequently asked questions
What is GEO?
GEO, short for Generative Engine Optimization, covers the practices meant to increase a brand's or a piece of content's presence, citation or influence in the answers of AI engines. The term comes from a research paper by Aggarwal and co-authors, submitted in November 2023.
What is the difference between GEO and AEO?
In everyday use, the two are often interchangeable. Microsoft Advertising separates them: AEO targets assistants and agents, GEO targets generative search environments. Google considers both to be still SEO. No definition has reached consensus, so ask which engines and which metric are meant.
Does GEO replace SEO?
No. The C-SEO Bench benchmark, presented at NeurIPS in 2025, concludes these methods complement SEO rather than replace it, because a document's place in the model's context outweighs rewriting it. In October 2026, search still sends about 60 times more visits than AI, according to Ahrefs.
What does AIO mean?
It depends who is talking. As a practice, AIO means Artificial Intelligence Optimization, close to a synonym for GEO. In measurement, it means Google's AI Overviews, which is how Olivier Martinez's review uses it. Spell the term out to avoid the misunderstanding.
Does GEO really raise visibility by 40%?
Not as a general rule. The figure is a relative maximum, on one metric, in a simulator where the documents already sit in the context, as Olivier Martinez's July 2026 review points out. The later C-SEO Bench benchmark found significant gains in only 3 of 54 cases.
Sources:
- Aggarwal et al., GEO: Generative Engine Optimization, arXiv November 2023, KDD August 2024.
- Puerto et al., C-SEO Bench: Does Conversational SEO Work?, NeurIPS Datasets and Benchmarks 2025, revised October 2025.
- Olivier Martinez, Optimizing Visibility in Generative Engines: A Critical Survey of Generative Engine Optimization (2023-2026), preprint, July 2026.
- Kong et al., Do GEO Gains Survive the Live Web?, preprint, October 2026.
- Watanabe and Nakayashiki, Disentangling Answer Engine Optimization from Platform Growth, preprint, June 2026.
- Uberti-Bona Marin et al., If I Had to Buy Just ONE: Auditing AI-Generated Product Recommendations, preprint, September 2026.
- Chen et al., University of Toronto, Generative Engine Optimization: How to Dominate AI Search, preprint, September 2025.
- Schulte et al., University of St. Gallen, Don't Measure Once: Measuring Visibility in AI Search (GEO), preprint, April 2026.
- SparkToro and Gumshoe, New research: AIs are highly inconsistent when recommending brands or products, January 2026.
- Google Search Central, Optimizing your website for generative AI features on Google Search, read 6 November 2026.
- Google Search Central, AI Features and Your Website, read 6 November 2026.
- Google Search Central, announcement of the generative AI performance reports, June 2026, update note of 31 August 2026.
- Google France, launch of AI Overviews and AI Mode in France, July 2026.
- Microsoft Advertising, From discovery to influence: A guide to AEO and GEO, January 2026, summarized by Search Engine Journal, breakdown of Microsoft's guide, and Search Engine Roundtable, Microsoft Advertising Releases Guide To AEO & GEO.
- OpenAI, Overview of OpenAI Crawlers, read 6 November 2026.
- Darkroom, SEO vs GEO vs AIO: Modern Search Optimization Guide, November 2026.
- Onely, GEO vs AEO vs AI SEO vs LLMO: What Our Research Shows They Mean, October 2026.
- Wikipedia, Generative engine optimization, read 6 November 2026.
- Agarwal and Sen, SSRN working paper 6513059, June 2026, summarized by PPC Land, Researchers find Google AI Overviews cut publisher clicks 39.8%.
- Pew Research Center, Google users are less likely to click on links when an AI summary appears in the results, July 2025.
- Ahrefs, AI vs Search Traffic Analysis, October 2026 data, read 6 November 2026.
- Ahrefs, Update: 38% of AI Overview Citations Pull From The Top 10, March 2026.
- Ahrefs, ChatGPT May Scrape Google, but the Results Don't Match, September 2025.
- Ahrefs, 1 mois d'AI Overviews en France et déjà -23,1% de taux de clics : qui est concerné ?, August 2026.
- SEOWorks, AI referral traffic benchmark, September 2025 to August 2026 data.
- Arcep and CREDOC, Baromètre du numérique, édition 2026, February 2026, survey of 5 to 21 June 2025.