It can be hard to make sense of the websites AI models cite in their responses. For one thing, there are so many of them.

For example, we got back about 850-6,700 unique URLs on average per model when we asked six models the same 41 questions everyday for 20 days. We repeated the 41 questions a significant number of times each day to account for the fact that the models are probabilistic, returning different responses each time they are prompted.

Average Number of Unique URLs by Model

Source: Evertune (www.evertune.ai) • Chart: Will Robinson
*In response to a basket of 41 questions, repeated for statistical significance each day for 20 days

This is one of the key distinctions between Generative Engine Optimization (GEO) and Search Engine Optimization (SEO). In SEO, it is vitally important to rank highly on a static list of search engine recommendations, since hardly anyone clicks to the second page of search results, whereas large-language models (LLMs) vacuum up information from thousands of pages to inform their responses.

Further complicating the matter, those thousands of model-specific URLs are not considered equally by models. Some shape models’ understanding more than others.

With so many links finding their way into AI search results for a small list of questions, we decided to run an experiment. Would news shared widely across URLs get cited more frequently by LLMs?

We tested on our own brand by putting out a press release across a newswire. The results surprised us and reminded us why GEO and SEO go hand-in-hand.

Our Newswire Experiment

On February 27, we shared our announcement that we had released a beginner’s guide to GEO via a newswire, and in quick succession, more than 180 sites shared our news.

Then we hopped onto the Evertune platform to watch the models start citing our links. But they didn’t.

In the three weeks since our release was published, the only URL that appeared as an AI source citation is the original PRWeb release, and even that failed to appear as a citation on Gemini or Perplexity.

It was hardly a top citation for any model. The PRWeb URL was included in responses daily for Google AI Overview and Google AI Mode, but mustered only a handful of mentions in each day’s responses. The URL was only mentioned once by Copilot and in only two days’ responses by ChatGPT.

Our GEO/SEO Takeaway

It appears most of our 180-plus links were essentially invisible when LLMs used search to respond to our prompts, a process known as Retrieval-Augmented Generation (RAG). The likely culprit: “canonicalization,” the process wherein Google and other search engines pick a representative URL and filter duplicate pages out of results.

For example, one of the sites that picked up our release was pr.youroregonnews.com. But it does not appear in Google’s search results when Googling the exact text used in the press release. It doesn’t even appear when including “site: pr.youroregonnews.com” in the Google search. 

This is a key example of why SEO still matters in the world of GEO. AI models may look at vastly more links than their human counterparts, but the links still need to adhere to SEO best practices to be visible to the models in the first place.

Methodology

At Evertune Research we track hundreds of brands across 250 categories. For this analysis, we asked ChatGPT, Copilot, Gemini, Google AI Overview, Google AI Mode and Perplexity a fixed list of 41 questions, repeating each for a statistically significant sample, everyday from February 28 through March 20.

Evertune® is the AI marketing platform that helps enterprise brands own the consumer journey that now runs through AI. Combining generative engine optimization (GEO), statistically significant data across every major AI model, and the only activation suite in the category, including integrations with impact.com and The Trade Desk, Evertune turns AI visibility insights into competitive action. Evertune works with leading brands across all verticals, including Finance, Retail and E-Commerce, Automotive, Pharma, Tech, Travel, Food and Beverage, Entertainment, CPG and B2B. Founded by early team members of The Trade Desk, Evertune has raised $20M in funding from leading adtech and martech investors. Headquartered in New York City, the company has a growing team of more than 40 employees. www.evertune.ai