How AI Learned Who To Recommend
Google is no longer the only place people go to find out about you. It may not even be the first. Here is what that actually looks like, by the numbers.
For about twenty years, the way an organization got found online followed a fairly predictable logic. You built a website. You filled it with the right words. Google’s algorithm crawled those words, ranked them, and served them up as a list of blue links to anyone who typed a relevant question into a search bar. The entire discipline of search engine optimization, or SEO, grew out of this exchange. Billions of dollars were spent on it. Entire industries formed around it. It worked.
It still works, in the sense that Google still serves search results and people still click on some of them. But something else has started happening alongside it, and the data from the past two years suggests it is not a temporary disruption. It is a rearrangement of the basic mechanics of how people find information online.
The shift has two names, depending on who you ask. Some call it Answer Engine Optimization, or AEO: the practice of structuring content so that AI-powered tools like ChatGPT, Perplexity, and Google’s own AI Overviews can understand it, trust it, and surface it as a direct answer. Others call it Generative Engine Optimization, or GEO: the discipline of making your content visible inside the AI-generated responses that are increasingly replacing the traditional list of links. The terminology is still settling. The trend is not.
What Changed, Exactly
The simplest way to understand the shift is to look at what happens after someone types a question into Google in 2026. Increasingly, they never leave Google at all. The answer appears at the top of the page, synthesized by AI from multiple sources, formatted as a readable paragraph with citations. Google calls these AI Overviews. By mid-2025, they appeared on roughly a quarter of all search queries, with the heaviest concentration in health, science, and informational searches. For those queries, the percentage ran as high as 90 percent.
The effect on click behavior is measurable. A study by the research firm Seer Interactive, published in late 2025, found that when an AI Overview appeared on a search result, organic click-through rates dropped 61 percent, falling from 1.76 percent to 0.64 percent. Paid ad click-through rates fell 68 percent. The searches still happened. The clicks often did not.
This is not limited to Google. ChatGPT now processes roughly one billion queries per day, according to usage data compiled by multiple analytics firms. Perplexity, an AI-native search engine, handles 780 million monthly queries and has grown its user base 66 percent year over year. Microsoft’s Copilot, which integrates AI search into Bing and the entire Microsoft 365 ecosystem, saw a 90 percent increase in weekly sessions during 2025. Together, these platforms represent an entirely new category of search behavior: people asking questions in natural language and receiving synthesized answers rather than lists of websites to visit.
Meanwhile, the volume of traditional Google searches in the United States dropped nearly 20 percent per user year over year in 2025, according to clickstream data analyzed by Datos and SparkToro. Europe saw only a 2 to 3 percent decline, suggesting the shift is most advanced in the U.S. market.
The Rise of the Answered Question
There is a term in the search industry for what happens when someone types a query, gets the information they need, and never clicks through to any website: a zero-click search. It is not a new phenomenon. Google has been answering simple factual questions directly in the search results for years. What is new is the scale.
SparkToro’s 2024 study, using clickstream data from Datos, found that 58.5 percent of Google searches in the United States ended without a single click to any website. In the EU, the figure was 59.7 percent. These numbers actually represent a modest improvement from earlier years (the figure was 65 percent in 2021), but the composition of what counts as a zero-click search has changed. Where it once meant a user glancing at a phone number or weather forecast, it now increasingly means a user reading a multi-paragraph AI-generated summary that draws on multiple sources and answers a complex question in place.
For organizations that have spent years optimizing their websites to attract search traffic, this is a meaningful evolution. The website still matters. But the moment of discovery, the point at which someone first encounters your organization or learns what you do, is increasingly happening inside the search engine itself, inside an AI chatbot’s response, or inside a voice assistant’s spoken answer. The content that gets surfaced in those moments is not chosen by the same rules that governed traditional SEO rankings.
How AI Decides What to Cite
This is where the shift becomes most interesting for mission-driven organizations. When a traditional search engine ranks a website, it evaluates factors like keyword relevance, backlinks, page speed, and domain authority. When an AI engine synthesizes an answer, it evaluates something subtly different. It looks for content that is clearly structured, authoritative, well-sourced, and easy for a language model to parse and restate with confidence.
Researchers at Princeton University, Georgia Tech, and IIT Delhi published a study in late 2023 (later presented at the ACM’s KDD conference) that formalized this as Generative Engine Optimization. Their key finding: content optimized for AI visibility could see up to a 40 percent increase in how often it was cited in generative search responses. The methods that produced the largest gains were not tricks or hacks. They were fundamentals: adding relevant statistics, citing authoritative sources, using clear and direct language, and structuring content so an AI model could extract and restate the key points.
In other words, the content that performs best in AI search is content that is genuinely useful, clearly organized, and backed by evidence. The days of gaming a ranking algorithm with keyword density and link schemes were already fading. In the age of generative search, they are all but done.
The content that performs best in AI search is content that is genuinely useful, clearly organized, and backed by evidence. The era of gaming algorithms is effectively over.
The data on structured markup reinforces this. Pages with clean heading structures and schema markup (a standardized format that helps machines understand what a page is about) are cited in AI responses at 2.8 times the rate of poorly structured pages, according to analysis compiled by SEO researchers in 2025. FAQ-formatted content is 3.2 times more likely to appear in Google AI Overviews. The pattern is consistent: AI search engines reward clarity, structure, and substance.
Who Gets Cited, and Why That Matters
One of the most revealing data sets to emerge from this shift is the list of which domains AI search engines cite most frequently. According to research by BrightEdge and Ahrefs, the top five cited domains control roughly 38 percent of all AI citations. The top 20 control 66 percent. The pattern is not random. It is concentrated.
YouTube is the single most cited domain in Google AI Overviews, appearing in 29.5 percent of all AI-generated summaries, according to BrightEdge’s analysis of data from May 2024 through September 2025. Across all AI platforms combined, YouTube holds a 20 percent citation share and is cited 200 times more frequently than any other video platform. Wikipedia, Reddit, and major institutional sites round out the top tier.
The implications of this are worth sitting with. In the traditional search model, any website could theoretically rank for a query if it played the SEO game well enough. In the AI model, the citations are fewer, more concentrated, and disproportionately favor content that AI systems consider authoritative. Video content hosted on YouTube has a structural advantage that text-only websites do not.
For nonprofits and government agencies, this creates an interesting dynamic. A well-produced video about your organization’s work, hosted on YouTube with proper titles, descriptions, and transcripts, has a measurably better chance of being cited by an AI engine than a blog post buried on page three of your website. The AI does not care about your site’s domain authority in the traditional sense. It cares about whether your content clearly and authoritatively answers the question someone asked.
What This Looks Like in Practice
A Conductor survey of more than 250 digital leaders, published in January 2026, found that 97 percent of CMOs and senior marketing executives reported that AEO and GEO strategies had a positive impact on their marketing funnels in 2025. Ninety-four percent said they planned to increase their investment in these strategies in 2026. These are not niche adopters. These are enterprise-level decision-makers who have seen the data from their own analytics dashboards and drawn the same conclusion.
The traffic patterns tell the same story from the other direction. AI-sourced website traffic grew 527 percent year over year in the first half of 2025, according to the Previsible AI Traffic Report. That number sounds dramatic, and it is, but context matters: AI-sourced traffic still accounts for a small fraction of total web traffic. The more significant finding is qualitative. Visitors arriving from AI recommendations convert at 4.4 times the rate of visitors from traditional organic search. They spend more time on the page, view more pages per session, and engage more deeply with the content. The audience is smaller but considerably more intentional.
For nonprofits, this is particularly relevant. A potential donor or volunteer who discovers your organization through an AI-generated answer to the question “What nonprofits in Richmond do community development work?” is further along in their decision-making process than someone casually browsing search results. They asked a specific question and received your organization as part of the answer. The trust transfer is built into the format.
Why Video Sits at the Center of This
It is not a coincidence that YouTube is the most cited domain in AI search. Video content occupies a unique position in the AEO and GEO landscape for several reasons. First, it is inherently multimodal: a single video can contain spoken language, visual demonstrations, on-screen text, and emotional nuance that a text article cannot replicate. AI systems can now parse all of these elements when deciding what to cite. Second, YouTube videos generate transcripts, which give AI models a rich text layer to analyze alongside the video itself. Third, video content tends to be more specific and situational than written content, which makes it easier for an AI to match to a particular query.
The practical effect is that an organization with a library of well-produced, clearly titled, properly described video content on YouTube has a significant structural advantage in the new search ecosystem. A 90-second explainer about your nonprofit’s housing program, a three-minute behind-the-scenes look at how a government agency processes permit applications, a testimonial from a program beneficiary: each of these is a discrete, citable asset that an AI can surface in response to a relevant question.
This is where the traditional content categories that video producers work in; advertisements, explainers, testimonials, brand stories, event coverage, training materials; map directly onto the new discovery mechanics. Each format answers a different kind of question. Each one, if produced with clarity and quality, becomes a candidate for AI citation. The organizations investing in a sustained library of this kind of content are building, whether they realize it or not, an AI-discoverable archive of their work and their value.
SEO Is Not Dead. It Is the Foundation.
One clarification matters here, because the conversation around AEO and GEO can sometimes suggest that traditional SEO no longer matters. The data says the opposite. Analysis of Google’s AI Mode responses has found that 99 percent of the URLs appearing in AI-generated answers also appear in the top 20 traditional organic search results. In other words, AI engines are not pulling content from obscure corners of the internet. They are pulling from the same pool of high-quality, well-optimized content that already ranks well in traditional search.
What has changed is the layer on top of that foundation. Having a well-optimized website is still necessary. It is just no longer sufficient by itself. The organizations that show up in both traditional search results and AI-generated answers are the ones that combine strong SEO fundamentals (fast site, clean architecture, relevant content) with the structured, authoritative, multimedia content that AI engines prefer to cite.
For nonprofits and government agencies, this actually simplifies the strategic picture rather than complicating it. The same investment in quality content, clear messaging, and professional video production that serves your website and social media channels is the investment that positions you for AI discovery. The difference is that in the AI model, the quality bar is higher and the reward for meeting it is more concentrated. The organizations that get cited in an AI answer do not share that space with ten blue links. They share it with three or four other sources, sometimes fewer.
The Shape of the Opportunity
What does all of this mean in concrete terms for an organization that wants to be visible in 2026? The data points toward a few consistent themes.
Video content, hosted on YouTube with accurate titles, thorough descriptions, and proper metadata, is the single highest-performing content type across AI citation platforms. Organizations producing a steady cadence of professional video; explainers, testimonials, advertisements, event coverage, training content, and social media edits optimized for each platform; are building a discoverable library that serves both traditional and AI search simultaneously.
Structured, clearly organized written content with proper schema markup performs measurably better in AI citations. Content that includes relevant statistics, cites authoritative sources, and answers specific questions in clear language is what generative search engines are designed to surface.
Website architecture still matters. Fast load times, clean heading structures, accessible design, and mobile optimization remain foundational. The AI layer rewards these things more, not less, than traditional search did.
And there is one more dimension worth noting. AI search engines are still learning which organizations to trust. The entities that establish a consistent, authoritative presence across their website, their YouTube channel, their social media, and the broader web are building the citation history that AI models will draw on for years. The organizations doing this well now are, in effect, training the algorithms on who they are.
The way people find information has changed before. The shift from phone books to websites felt seismic at the time. The shift from websites to Google felt equally large. What is happening now is a change of similar scale, and the data is still early enough that its full shape is not yet visible. But the direction is clear, and the numbers are consistent across every major research firm tracking it.
The question for any organization; nonprofit, government agency, or otherwise; is not whether to abandon SEO. It is whether to keep treating search visibility as a game played exclusively on a website, or to recognize that the playing field has expanded to include every platform where an AI might go looking for an answer about who you are and what you do.
SourcesSeer Interactive. “AIO Impact on Google CTR: September 2025 Update.” November 2025. seerinteractive.com/insights/aio-impact-on-google-ctr-september-2025-updateSparkToro / Datos. “2024 Zero-Click Search Study.” sparktoro.com/blog/2024-zero-click-search-studyBrightEdge. “YouTube’s Growing Presence in AI Search.” 2025. brightedge.com/resources/weekly-ai-search-insights/youtube-presence-ai-searchMurahari, V. et al. “GEO: Generative Engine Optimization.” Princeton University, Georgia Tech, IIT Delhi. arXiv:2311.09735. Presented at KDD ’24.Conductor. “The State of AEO / GEO in 2026: CMO Investment Report.” January 2026. conductor.com/academy/state-of-aeo-geo-reportPrevisible. “2025 AI Traffic Report.” Reported via Search Engine Land. searchengineland.com/ai-traffic-up-seo-rewritten-459954Datos / SparkToro. “Google Searches Per U.S. User Decline.” Reported via Search Engine Land. searchengineland.com/google-searches-per-us-user-fall-report-468051Ahrefs. “Most Cited Domains in AI Mode.” 2025. ahrefs.com/blog/most-cited-domains-ai-modeGartner. “Gartner Predicts Search Engine Volume Will Drop 25% by 2026.” Press Release. February 19, 2024. gartner.com/en/newsroom/press-releases/2024-02-19Semrush / ALMCorp. “AI Overviews Study 2026: Complete Analysis.” almcorp.com/blog/semrush-ai-overviews-study-2026-complete-analysis