Updated October 23, 2026: rebuilt the post around current evidence (Ahrefs, Seer and Pew click data, the C-SEO Bench and 2026 survey results on GEO tactics, Google’s July 2026 AI optimization guide, the new Search Console and Bing AI reports) and removed claims that did not hold up.
There is no stable “AEO ranking.” When SparkToro ran the same brand-recommendation prompts through ChatGPT, Claude and Google’s AI 2,961 times, there was less than a 1 in 100 chance of getting the same list twice, and about a 1 in 1,000 chance of getting it in the same order (SparkToro, January 2026). What you can measure is share: how often your brand or page shows up across many prompts, each run many times. And the evidence on how to raise that share is thinner than GEO vendors admit. A July 2026 survey of 45 GEO studies found that topical relevance and position in the model’s context are the most reproducible levers, and that no technique has shown a stable causal effect across platforms. In practice, that means solid SEO plus answer-first writing, measured honestly.
The stakes are real anyway. When an AI Overview appears, the top organic result’s click-through rate is 58% lower than Ahrefs’ model predicts it would be without one. Yet AI assistants sent just 0.43% of visits across 115,254 sites in September 2026. Clicks are leaving the results page, and they are not coming back as AI referrals. That gap is why GEO (Generative Engine Optimization) and AEO (Answer Engine Optimization) matter, and why most of what is sold under those labels deserves skepticism.
How AI Search Picks Its Sources
ChatGPT search, Perplexity, Copilot, Google AI Overviews and AI Mode all use some form of Retrieval-Augmented Generation (RAG). The model does not answer from memory alone. It turns the question into one or more searches, pulls documents from an index, and writes an answer that cites a few of them. Google describes its own version the same way: RAG on top of its core ranking systems, plus “query fan-out,” a set of related queries the model generates to fetch more results. Its example: “how to fix a lawn that’s full of weeds” fans out into “best herbicides for lawns” and “remove weeds without chemicals.”
That pipeline has a consequence most GEO advice skips: a page that never gets retrieved cannot be cited, however well it is written. Retrieval runs on ordinary search infrastructure. Google requires a page to be indexed and eligible to show with a snippet before it can appear in AI features, and since August 31, 2026 Search Console has a Search generative AI setting that includes or excludes a site from AI Overviews and AI Mode (included by default). OpenAI states that sites blocking its OAI-SearchBot crawler “will not be shown in ChatGPT search answers” (OpenAI crawler docs). Blocking GPTBot, the training crawler, is a separate switch and does not affect search.
Fan-out is also the likely reason ranking #1 no longer guarantees a citation. In July 2025, Ahrefs found that 76% of AI Overview citations came from pages in the organic top 10. Its March 2026 update, covering 863,000 keyword SERPs and 4 million cited URLs, put that share at 38%, with 31% from positions 11 to 100 and 31% from outside the top 100. Ahrefs changed its parsing method between the two studies, so part of the drop is measurement. But the direction matches an independent audit summarized in the 2026 survey: 53% of domains cited in AI Overviews did not appear in the organic top 10. Ranking for the sub-questions matters, not just the head term.
What Gets Cited, According to the Data
- Text near the top of the page. Kevin Indig analyzed 18,012 verified ChatGPT citations: 44.2% came from the first 30% of a page, 31.1% from the middle and 24.7% from the final third (Search Engine Journal, February 2026). Cited passages were also almost twice as likely to use definitive language (36.2% vs. 20.2%) and to contain a question, mostly in headings. These are correlations, not proof of cause, but they fit how retrieval and summarization work.
- Not schema markup. Ahrefs tracked 1,885 pages that added JSON-LD between August 2025 and March 2026 against about 4,000 matched control pages. Relative to the controls, citations fell 4.6% in AI Overviews (small, but statistically significant) and rose 2.4% in AI Mode and 2.2% in ChatGPT, both indistinguishable from zero. Their verdict: “Adding schema produced no major uplift in citations on any platform” (Ahrefs, May 2026). Google agrees: there is “no special schema.org markup you need to add.”
What gets ignored is predictable: pages blocked from crawling, content behind logins, and pages that never made it into the index.
GEO vs. AEO vs. SEO: Three Labels, One Pipeline
These terms get used interchangeably online, but they started out describing different things.
SEO (Search Engine Optimization) optimizes for ranking algorithms. You compete for position in a list of links. Success is measured by ranking position, click-through rate, and organic traffic.
AEO (Answer Engine Optimization) originally meant being the direct answer to a specific question: Google’s featured snippets, voice assistant responses, and answer boxes above the organic results. AEO predates AI search. The format is tight: concise, factual, usually one paragraph or a short list. In 2026 the term mostly gets used for AI answers in general, and tool vendors such as Conductor treat AEO and GEO as the same thing.
GEO (Generative Engine Optimization) optimizes for being cited inside an AI-generated response from ChatGPT, Perplexity, Copilot or Google’s AI features. The term comes from a 2023 research paper. Unlike classic AEO, GEO is not about being “the” answer. It is about being one of several sources the model weaves together: Pew Research found that 88% of Google’s AI summaries cited three or more sources.
What Google Says About the Labels
Google’s AI optimization guide, updated July 10, 2026, is blunt: “From Google Search’s perspective, optimizing for generative AI search is optimizing for the search experience, and thus still SEO.” Google has an interest in saying that, and it only speaks for Google. But it matches the independent evidence: the C-SEO Bench paper (NeurIPS 2025) found that “traditional SEO strategies, those aiming to improve the ranking of the source in the LLM context, are significantly more effective” than GEO-style rewriting.
So the useful distinction is not in the page work, which overlaps almost entirely. It is in the reporting. SEO measures rank, clicks and traffic. AEO and GEO measure whether you appear in the answer, how often, against whom, and whether the answer describes you correctly.
What “Ranking” Means in AEO
People who search for “AEO ranking” usually want one of two things: how to rank in AI answers, or a ranked list of AEO tools and agencies. Both run into the same problem. AI answers are probabilistic, so there is no fixed position to win or to report.
The variability is large and well documented:
- Lists rarely repeat. In the SparkToro study, ChatGPT and Google’s AI returned the same brand list twice less than 1% of the time; Claude was only slightly more consistent. Rand Fishkin’s conclusion: “visibility % across dozens to hundreds of prompts run multiple times is a reasonable metric,” while “any tool that gives a ‘ranking position in AI’ is full of baloney.” (SparkToro partnered with Gumshoe.ai, an AI-tracking startup, and disclosed it.)
- Sources churn daily. Across four engines and 45 days, Schulte et al. measured daily source overlap (Jaccard) of only about 0.34 to 0.42 for the same prompts, as summarized in the 2026 survey.
- AI Overviews churn faster than blue links. Kirsten et al. measured page overlap across two months of 18% for AI Overviews, against 45% for organic Google results (same survey).
- Sometimes there is no search at all. In one configuration Schulte et al. tested, 57.8% of ChatGPT runs did not trigger a web search, so no page could be cited.
In practice, an honest “AEO ranking” is three numbers for a defined prompt set:
- Visibility rate: the share of runs in which your brand is mentioned or your page is cited.
- Citation share: your citations as a share of all citations for those prompts, compared with competitors.
- Accuracy: whether the answer describes your product, pricing and claims correctly.
As for the ranked lists that dominate this search: read them as marketing. Conductor’s “10 best AEO tools” puts Conductor first. A widely shared list scores platforms with “Profound’s proprietary AEO ranking model” and puts Profound first, with a 10% weight on schema markup that Ahrefs’ controlled test could not connect to more citations. Google’s guide adds a fair warning: “No third-party tool has access to our internal ranking or AI systems.”
Does GEO Work? What the Research Shows
The famous “up to 40%” comes from the original GEO paper by Aggarwal et al. (KDD 2024). It is real, but narrow. The authors built a simulated engine: GPT-3.5-turbo writing answers from the top five Google results for each query. Adding quotations, statistics or source citations to one of those five documents raised its visibility score by 30 to 40% relative to the unedited baseline. Keyword stuffing made things worse. A 200-query test on Perplexity.ai also showed gains for quotations and statistics, though less consistently across methods.
Two things have happened since:
- C-SEO Bench tested published GEO methods across question answering and product recommendation, three domains each, with several competing sites optimizing at once. Its finding: “most current C-SEO methods are not only largely ineffective but also frequently have a negative impact on document ranking.” The more sites adopted them, the smaller the gains, which the authors describe as “congested and zero-sum.”
- The July 2026 survey of 45 studies (a single-author preprint by Olivier Martinez) concludes that the GEO paper’s gains are “conditional on a source already being present in a fixed context” and “establish neither organic discoverability nor durable traffic effects.” Topical relevance and context position replicate; generic heuristics transfer poorly; citation-oriented rewrites can hurt retrieval.
Here is how the common tactics stack up against that evidence:
| Tactic | What the evidence says | Verdict |
|---|---|---|
| Be indexed, crawlable and relevant | Required for Google AI features; ranking-focused work beat GEO rewrites in C-SEO Bench | Do it first |
| Answer in the first third of the page | 44.2% of ChatGPT citations come from the first 30% (correlational) | Worth doing |
| Add statistics, quotes, sources | Helped in the GEO simulator once the page was already a source; survey rates verifiable evidence “moderate to strong” and warns fabricated stats raise reuse while lowering quality | Real evidence only |
| Show fresh dates | Survey rates the effect moderate, mainly for time-sensitive and commercial queries | Conditional |
| Schema markup for AI | Ahrefs difference-in-differences test: no uplift; Google: no special schema needed | No measurable effect |
| llms.txt | Google Search ignores it | Not for Google |
| “Chunking” content for AI | Google: not required, no ideal page length | Skip |
| Mass pages for fan-out queries | Google: violates its scaled content abuse policy | Risky |
What to Do Instead
The survey’s own bottom line is the best summary: “produce a relevant, comprehensive, verifiable, clearly structured, and technically retrievable page; then measure retrieval, citation, and fidelity separately.” In practice:
1. Make Sure You Can Be Retrieved
Check that your key pages are indexed in Google and Bing, that they are not set to nosnippet, and that robots.txt allows OAI-SearchBot if you want to appear in ChatGPT search. Look at Settings > Search generative AI in Search Console to confirm your site is included. Then do the unglamorous part: rank for the specific questions your buyers ask, including the sub-questions an AI system would fan out to. If you want a single takeaway from C-SEO Bench, it is that this beats any rewriting trick.
2. Lead with the Answer, Then Expand
Every section should follow the inverted pyramid: claim first, evidence second, context third. If someone asks “how much does HubSpot cost?”, the first sentence of your pricing section should contain the number, not a paragraph about HubSpot’s founding story. Indig’s data shows ChatGPT cites the top of a page far more than the bottom, and question-style headings show up disproportionately in cited text. It is also just better writing.
3. Bring Evidence Nobody Else Has
AI systems need something concrete to cite. “Our product is fast” gives the model nothing. “Our CDN delivers 95th-percentile latency under 50ms across 280 edge locations” gives it a specific, verifiable claim it can attribute to you. Google’s guide pushes the same way: it favors “non-commodity” content with a first-hand point of view over summaries of what everyone else has said. The survey’s rule is the right one: not “add numbers,” but “provide relevant, verifiable, dated, and properly attributed evidence.” Original data, real test results and named expert sources are what make your page the one worth citing.
4. Update Facts That Change
Freshness helps when the question is time-sensitive, such as prices, versions, rankings or “best X in 2026.” It helps much less for stable definitions. Update the pages where the facts move, and change the date only when the content actually changed.
5. Skip the Hacks
The llms.txt, chunking and fan-out rows of the table come straight from Google’s guide, which also says to ignore rewriting content “just for AI” and chasing inauthentic mentions. Building a page for every fan-out variation “primarily to manipulate rankings or generative AI responses” violates its scaled content abuse policy. Other engines may treat llms.txt differently, but no study cited here shows it moving citations.
How to Measure AEO and GEO
Free First-Party Reports
- Google Search Console, generative AI performance report. Announced June 3, 2026 and rolled out to all sites worldwide by August 31, 2026. It shows impressions in AI Overviews, AI Mode and Discover’s generative AI features, broken down by page, country, device and date (down to hourly). The announcement lists no click or query data, so it tells you where you appear, not what it earns you.
- Bing Webmaster Tools, AI Performance. Public preview since February 10, 2026. It shows total citations, cited pages, page-level citation counts and “grounding queries” (the phrases the AI used to retrieve your content, shown as a sample) across Microsoft Copilot, AI summaries in Bing and select partner integrations. A June 2026 update added intent, topic and citation share per grounding query (Practical Ecommerce). No clicks here either.
- Your own analytics. Track referral sessions and conversions from ChatGPT, Perplexity, Gemini, Claude and Copilot. Expect small numbers: ChatGPT accounted for 0.34% of visits in Ahrefs’ September 2026 benchmark.
Third-Party Trackers
Profound, Peec AI, Otterly.AI and Ahrefs Brand Radar cover the engines Google and Bing cannot: ChatGPT, Perplexity, Gemini and others. Most work by running prompt sets against the engines and counting mentions and citations. They are useful if you use them like a statistician, not like a rank tracker:
- Repeat every prompt. Schulte et al. propose seven to eight runs per prompt as a starting point, from a small Swiss query set, so treat it as a floor.
- Vary the inputs. The survey’s rule: a GEO measurement has to vary “run, paraphrase, date, and engine.” Real users phrase the same need very differently.
- Count the misses. If a tool only computes citation share among answers that contained citations, it overstates your visibility, because many runs never search at all.
- Ignore “average position.” See SparkToro above.
Conversion Claims
You will see “AI visitors convert 4.4x better than organic” everywhere. It comes from a 2025 Semrush study, which says the figure is “based on conversion rate” but publishes neither of the two rates, the number of sites, nor the period. Measure your own: AI referral conversion rate against organic, on your site, over a full quarter.
GEO Statistics That Hold Up (October 2026)
| Statistic | Number | Source and caveat |
|---|---|---|
| ChatGPT weekly active users | 900 million | OpenAI, February 2026 (TechCrunch) |
| AI Overviews / AI Mode monthly users | 2.5 billion+ / 1 billion+ | Google, I/O May 2026, self-reported (Google) |
| Position-1 CTR when an AI Overview shows | 58% below forecast | Ahrefs, 300,000 keywords, desktop, Dec 2023 vs. Dec 2025, modeled counterfactual (Ahrefs) |
| Visits with a click on a result | 8% with AI summary vs. 15% without | Pew, 900 U.S. adults, March 2025; only 1% clicked a link inside the summary (Pew) |
| Organic CTR, informational queries | 2.07% cited in AIO, 0.94% not cited, 3.35% no AIO | Seer Interactive, 53 brands, Jan 2025 to Feb 2026, correlational (Seer) |
| Share of web visits from AI assistants | 0.43% (search: 26.82%) | Ahrefs, 115,254 sites, September 2026 (Ahrefs) |
| AI Overview citations from top-10 pages | 38% (was 76% in July 2025) | Ahrefs, 4M cited URLs; parsing method changed between studies (Ahrefs) |
| ChatGPT citations from first 30% of a page | 44.2% | Kevin Indig, 18,012 citations, correlational (SEJ) |
| Citation change after adding schema | No significant uplift | Ahrefs, 1,885 pages vs. ~4,000 controls (Ahrefs) |
| Same AI brand list twice | Under 1 in 100 | SparkToro, 2,961 runs, not peer reviewed (SparkToro) |
| Visibility gain from quotes and statistics | Up to 40% | GEO paper, simulated engine with 5 sources already in context (arXiv) |
Treat any “AI Overviews appear on X% of searches” figure with care: it depends entirely on the keyword sample. In Pew’s panel of real users, 18% of searches produced an AI summary, but 53% of searches with ten or more words did, and 60% of searches phrased as questions. So a tracker that samples long, question-style keywords will report a far higher rate than one that samples everything.
The Real Risk Is Not the Click
Being cited beats not being cited, but it does not replace the click. In Seer’s data, cited brands earned about twice the organic CTR of uncited ones on informational queries, and still about 38% less than on results pages with no AI Overview. Seer calls that correlational, since stronger brands are more likely to be cited in the first place. Pew found that people ended their browsing session after 26% of visits to a results page with an AI summary, against 16% for pages without one.
The 0.43% referral share understates the shift, because it only counts clicks. When ChatGPT recommends your competitor, the user may never visit your site to compare. The decision was made inside the chat window. That is the real threat: not losing clicks, but never being considered. The defense is less exotic than the GEO industry suggests: be retrievable, be the clearest and best-evidenced answer, and measure your share of answers instead of chasing a position that does not exist.
Frequently Asked Questions
What is the difference between GEO and AEO?
AEO (Answer Engine Optimization) started as the practice of winning featured snippets and voice answers: one short, direct answer to one question. GEO (Generative Engine Optimization) is about being one of the sources an AI system such as ChatGPT, Perplexity or Google AI Overviews cites when it writes an answer. In 2026 most vendors use the two terms interchangeably, and Google’s own guidance says optimizing for its generative AI features is still SEO.
Does GEO replace SEO?
No. AI search engines retrieve pages through ordinary search indexes before they cite them, and a page that is not retrieved cannot be cited. The C-SEO Bench study (NeurIPS 2025) found traditional ranking strategies significantly more effective than GEO-specific rewrites, and Google says its AI features rely on its core ranking systems. Search also still sends far more traffic: about 27% of visits across 115,000 sites in Ahrefs’ September 2026 data, versus 0.43% from AI assistants.
How do I measure my visibility in AI search?
Start with the free first-party reports: the generative AI performance report in Google Search Console (AI Overviews, AI Mode and Discover impressions, available to all sites since August 31, 2026) and the AI Performance report in Bing Webmaster Tools (Copilot citations and grounding queries). Third-party trackers such as Profound, Peec AI, Otterly.AI and Ahrefs Brand Radar add ChatGPT, Perplexity and Gemini. Because AI answers vary between runs, measure visibility rate and citation share across many prompts, each repeated several times, rather than a single position.
What kind of content gets cited in AI search?
Content that is relevant to the question, gets retrieved by the engine’s search step, and states its answer early. In Kevin Indig’s analysis of ChatGPT citations, 44% came from the first 30% of a page. Verifiable statistics and quotes helped in the original GEO experiments once a page was already among the sources, but independent tests found most rewriting tricks ineffective, and an Ahrefs controlled study found no meaningful citation uplift from adding schema markup.
Is there such a thing as an AEO ranking?
Not in the sense of a fixed position. SparkToro ran identical prompts through ChatGPT, Claude and Google’s AI 2,961 times and found under a 1 in 100 chance of getting the same brand list twice, and about 1 in 1,000 of getting it in the same order. The meaningful ranking is your share of answers: how often you are mentioned or cited across a representative set of prompts, compared with competitors. Published lists of top AEO tools are a different thing, and many are written by vendors that rank themselves first.
