Generative Engine Optimization Tested as Google Adds AI Links

A month of convergence and contradiction

October 2026 produced no single landmark launch in generative engine optimization (GEO). It did produce three developments that, read together, show where the discipline is heading. Google changed how links appear inside its AI answers. AI Overviews spread to queries they had largely left alone, including brand-name searches. And a fight over whether GEO exists as a separate practice reached the point where both sides started citing peer-reviewed papers.

My reading is that GEO is being folded back into search engine optimization, but not in the way skeptics claim. The page-level tricks that defined early GEO are losing credibility. Meanwhile, the evidence keeps showing that AI answer engines retrieve and cite sources by different rules than Google's ranked list does. What remains is closer to reputation management and measurement than to a new branch of technical SEO.

Google pushes links back into AI answers

On October 15, Google VP of Product Management Hema Budaraju announced changes that apply to both AI Overviews and AI Mode. The headline change is that inline links will appear more often and higher in responses.21 The package also includes a test of a section near the top of some answers that points to one or two official destinations for task-based queries, wider rollout of timely-article carousels that highlight Preferred Sources, thumbnail-style related-article links, and a test of clickable creator names on AI Mode video results.2229 Google said Preferred Sources usage has grown more than sevenfold this year, though it gave no baseline figure.21

Coverage of the update agrees on one important caveat. Google offered no data showing that these changes send more traffic to websites.2429 Search Engine Roundtable's headline noted that Google won't show click data.21 Practitioners also point out that Search Console's generative AI performance report, added in June, shows impressions rather than clicks.244 Some SEOs doubt users will click much when the answer is already on the screen.30

Other changes point the same direction. Google reportedly added UTM parameters to outgoing Gemini links so site owners can attribute those referrals.22 It also substantially expanded its "Creating helpful content" documentation, adding sections on effort, originality, accuracy and how content was created.22 Taken together, Google appears to be answering publisher complaints about lost traffic with more visible links and better attribution, while still not saying how many clicks it actually sends.

AI Overviews reach branded and commercial queries

The more consequential change may be where AI Overviews now appear. DemandSphere reported that AI Overviews on branded queries tripled in September. Search Engine Land described a jump from 26% to 80% of branded queries, and Ahrefs put the figure at 83%.21 Ahrefs' September crawl found AI Overviews on 73.3% of branded and 78.3% of non-branded US desktop keyword results.28 Semrush, tracking 400 commercial-modifier keywords, saw AI Overview coverage rise from about 46% to 88% in a year, and to 97% when weighted by search volume.21

This is where the reporting diverges most sharply, and the gaps matter. Recent prevalence estimates range from 4.5–12.5% depending on the study27, to roughly 18%23, to more than 20% in SparkToro's clickstream panel10, to 48%26, and up to the high 70s in Ahrefs' keyword crawl.28 The numbers measure different things. Some count share of all searches, and others count share of tracked keyword sets. They also come from different months during a period of rapid expansion. Anyone quoting a single figure without its method and date is misleading their audience.

OmniSEO calls this problem "stat drift." It found a Google AI Mode answer still repeating the 34.5% click-through-rate decline that Ahrefs had replaced eight months earlier.28 That same 34.5% figure still appears in current industry guides.127 The irony is clear: an industry built on getting AI systems to repeat accurate facts is spreading outdated statistics through those same systems.

How much traffic is actually lost

The direction of the traffic data is consistent even though the size of the effect is not. SparkToro found that 68.01% of US Google searches between January and April 2026 ended without a click, up from 60.45% in 2024.10 Its estimate of the click-through decline when an AI Overview appears is close to 60%.10 Other analyses cite about 18% for the top-ranked result26, and an NJIT and NTU difference-in-differences study found roughly a 15% causal reduction in daily traffic to English Wikipedia articles exposed to AI Overviews.23

AI Mode itself is a smaller factor than its user count suggests. Google reports more than 1 billion monthly AI Mode users10, but SparkToro found that only 0.34% of searches in its panel moved into AI Mode, so the tab played a limited role in the click decline.21 AI Overviews, not the chatbot-style tab, are doing most of the damage to referral traffic.

Google's business is not suffering. Alphabet reported 24% year-over-year revenue growth in Q2 2026, with Search and other revenue up 17%.10 Rising search revenue alongside falling publisher clicks is the economic tension behind every "AI SEO" sales pitch.

The "GEO is dead" fight

On October 20, the Hobo Web blog argued that GEO had been disproven, that work for AI answers is "still SEO," and that every distinct GEO tactic amounts to spam.2 Google's own position supports part of that argument. Its 2026 documentation, the first official statement of a view it had shared at conferences, mentions AEO and GEO as terms people may encounter but treats optimizing for generative AI search as SEO.3 Bing, by contrast, uses "generative engine optimization" in its documentation.1 Google has also said it does not use llms.txt files, and Ahrefs found that 97% of valid llms.txt files received zero requests in May 2026.4

A rebuttal from Novel Cognition checked the critique against its sources and found that much of it holds up. The 2023 paper that coined GEO allowed its rewriting prompts to invent statistics, quotations and sources, and later benchmarks found most of its tactics ineffective.2 The rebuttal also identified a quotation attributed to the C-SEO Bench paper that does not appear in it, and noted that the SAGEO Arena results came from a frozen 171,003-page offline snapshot, not the live web.2

The most important evidence is a University of Toronto study by Chen and Koudas, released October 10. It replayed ChatGPT's own searches and found that 97% of the pages ChatGPT retrieved were absent from Google's top 50 for the same queries. Even Gemini, which is grounded on Google Search, cited pages outside the top 50 93% of the time.2 The same study found no significant within-site benefit from FAQ blocks, schema markup or summary boxes.2

That finding undercuts both camps. If classic ranking is "not a proxy" for AI retrieval, as the authors put it, then saying "it's just SEO" is too simple.2 If FAQ blocks and schema don't move citations, then much of the standard AEO checklist is unsupported. Several agency guides still recommend answer-first FAQ sections and structured data as the easiest wins.11 Some even claim that 99% of AI Overview citations come from pages already ranking in the organic top 10.11 Other analyses put that overlap at 38%25 or 12% across LLMs.13 The disagreement is too large to be explained by methodology alone, and on balance the more recent and more rigorous work favors low overlap.

Where the evidence actually points

The signal that holds up across sources is off-site reputation. Muck Rack found that earned media accounts for 84% of links cited by AI engines, while paid content accounts for 0.3%.10 AirOps reported that brands are 6.5 times more likely to be cited through third-party sources than through their own domains.19 A Stacker and Scrunch study found a 239% median visibility lift from distributing content through third-party publications.10 BrightLocal found that use of AI tools for local business recommendations rose from 6% to 45% in a year.4 For small businesses, reviews and mentions are the inputs that matter.

That is also where vendors are building. Bazaarvoice partnered with Profound to track brand sentiment and citation frequency in generative answers using reviews.9 HubSpot sells an AEO product starting at $50 a month12, and tool roundups now list a dozen or more platforms tracking citations across ChatGPT, Gemini, Claude, Perplexity and Google's AI surfaces.141820 Agencies are relabeling existing SEO services as GEO. Digital Web Solutions reports average citation gains of 25% across accounts5, and Ignite Digital has added GEO to its SEO programs.6 None of these vendor figures is independently verified, and they should be read as marketing claims.

Measurement itself is becoming less reliable. Since September 20, rank-tracking firm Nozzle has seen Google return "poisoned" results pages to automated requests, peaking at 25% of pages and running at 5–11% a day in October. These pages showed AI Overviews less often than normal results did.22 If Google is degrading scraped data to deter AI crawlers, as Nozzle suspects, the dashboards that GEO tools sell become less trustworthy at the moment they are most in demand.22

The bottom line

Foundational SEO still determines whether a page is eligible to be cited. Google says being indexed and eligible for a snippet is the only requirement.24 Beyond that, the evidence does not support a separate technical playbook. It supports earning mentions on the sites AI systems actually read, and measuring the results carefully against a strong SEO baseline, which is the standard Novel Cognition proposes for both vendors and critics.2 GEO is not dead. Its most marketable shortcuts are failing under testing, and what survives looks a lot like public relations combined with analytics.