Comparison Of AI Citation Patterns Offers Strategic SEO Insights via @sejournal, @martinibuster

BrightEdge published new data showing the different kinds of sites five AI search surfaces tend to show in generated answers. The data makes it possible to see how those differences shape which types of sites each AI engine shows, with strong implications for how to promote to each one.

The research focused on five AI search surfaces:

1. ChatGPT
2. Google AI Overviews
3. Google AI Mode
4. Google Gemini
5. Perplexity

AI Engines Cite Different Sources But Recommend The Same Brands

The BrightEdge research compared the top cited website sources across AI engines to measure how much they overlap (Source Overlap). What the data shows is that there was a wide discrepancy across the five AI search engines tested, with the lowest level of overlapping source citations between any two AI search surfaces at 16% and the highest level of agreement between any two engines at 59%.

• Lowest level of agreement: 16%
• Highest level of overlap: 59%

Significant Agreement In Brand Citations

BrightEdge also measured brand name overlap between the five AI search surfaces and found that there was more agreement between all five. The lowest overlap between any two AI surfaces was 36% and the highest level of overlap between any two surfaces was 59%.

• Lowest level of overlap: 36%
• Highest level of brand citation overlap: 55%

This suggests that name brands that are tightly associated with products and services tend to perform similarly across most of the tested AI search surfaces and may also reflect how widely brands are cited by trusted websites and possibly user intent and expectations.

In my opinion, the takeaway here is that associating a brand with a product or service in a consumer’s mind is a powerful way to influence user expectations which can then translate into branded search. This is something that the SEO community has been slow to pick up on, even though Google has been hinting at user signals playing a strong role in rankings. I say that the SEO community has been slow to pick it up because Google’s been doing this since at least 2004 (Navboost) and most directly with the brand navigation signals in search (Google’s brand signals patent).

Wide Divergence Of Cited Sources

BrightEdge analyzed citations from the five AI surfaces across three types of websites (Institutional, Commercial and Editorial, and User Generated Content) and discovered wide variance between all five engines, despite the convergence on citing strong brands.

Three Categories Of Sites Analyzed

1. Institutional sites, including government, academic, and big brand industry leaders
2. Commercial and editorial sites, including media, reviews, and listings
3. User Generated Content (UGC), including forums, video platforms, and social content

The data shows that every engine draws from all three categories, but weights the mix differently: institutional sources range from a low citation rate of 10% to a high of 26% of citations. Citations of UGC sites range from a low of 0.2% to a high of 18% of citations.

The largest category overlap across all five search engines are found in citations of corporate brand, commercial, and editorial sites, with a low end of 37% on Gemini to as high as 51% on AI Overviews.

BrightEdge offers this takeaway about that data:

“Review sites, comparison content, trade press, retailer listings, and finance data are the sources AI most frequently reaches for. Investment in PR, trade coverage, review site visibility, and category comparison content translates into visibility across every engine, not just one.”

Something that BrightEdge doesn’t mention is that AI search engines surface sponsored articles from trusted websites that are clearly labeled to conform with FTC guidelines on native advertising and Google’s guidelines on sponsored posts. This enables companies to tightly associate their brands with specific products and services and increase the likelihood of being cited in AI search surfaces.

Gemini And AI Overviews Differ On Website Authoritativeness

The difference between the kinds of websites Gemini and Google AI Overviews uses as sources shows that Gemini is more conservative, tending to show more trust toward institutional sites at a higher rate than user generated content (UGC). Institutional sites are academic, government, academic, and big brand sites.

AI Overviews, on the other hand, trusts both institutional and UGC sources of information, with nearly twice as many citations going to UGC websites.

• Authoritativeness Of Institutional Versus UGC Content
• Gemini: 26% institutional, 0.2% community
• AI Overviews: 10% institutional, 18% community

Another revealing finding is that there is a wide variance in thhe top level domains that are cited by each AI search surface. Gemini tended to link out to only the very most trustworthy and authoritative websites. For example, Gemini tended to cite .gov and .org websites at higher rates than any of the other AI engines.

Gemini: 13% .gov, 23% .org

Gemini’s answers tend to trust institutional websites more than user generated content, citing them 26% of the time but distrusts UGC sites, only citing them a fraction of a percentage point. AI Overviews trusts UGC content to a vastly greater extent. Why is that?

It could be that the technologies underlying Gemini and AI Overviews differ. For example, it could be that Google’s FastSearch, which prioritizes speed over other ranking signals, may be a reason why UGC sites are sources more often than they are in Gemini. It’s an interesting question.

I did an informal experiment by asking both Gemini and AI Overviews to compare the use of a specific op-amp (an electrical part) in a specific amplifier.

• Gemini’s answer cited institutional sources (Texas Instruments and the amplifier’s manufacturer).
• AI Overviews cited the two institutional websites but also multiple user generated content (UGC) sites.

Gemini’s answer was typically conservative, citing the institutional website (Texas Instruments, the manufacturer).

AI Overviews citations of various UGC sites were useful in the context of this question because actual users shared their experiences with this op-amp as well as actual electronic measurements of the op-amp and comparisons to other ones.

.Edu Sites Not Authoritative?

Another interesting finding is that all of the AI search engines don’t often cite .edu websites. Perplexity cited .edu sites at a higher rate than any of the other AI engines, citing .edu websites 3.2% of the time.

Those results contradict a longstanding belief in SEO circles that .edu sites are more authoritative. BrightEdge’s research shows that .edu sites are not authoritative for the kinds of questions that users are asking AI search engines.

ChatGPT Cites A Higher Diversity Of Sources

The data also shows that ChatGPT shows a more diverse variety of website sources, relying on its top ten sources only 18.5% of the time, with Google AI Mode right behind it with 19.4%. Gemini (26.3%) and Perplexity (26.7%) show a greater amount of the same sites drawn from their top ten.

Percentage Of Top 10 Sources

• ChatGPT: 18.5%
• Google AI Mode: 19.4%
• Gemini: 26.3%
• Perplexity: 26.7%

Gemini And Perplexity Rely On Authoritative Sites

Gemini and Perplexity tended to rely the most on authoritative websites. As already noted, Gemini trusted institutional sites the most and Perplexity cited .edu sites more than any of the other AI engines.

Perplexity showed a similar pattern of conservatively linking out to the most trusted and authoritative sites. BrightEdge’s report explains:

“Perplexity concentrates more of its citations in institutional medical, government, encyclopedic, and medical publisher sources than any other engine. Combined, those four categories account for approximately 30% of Perplexity’s citations.”

Five AI Engines, Five Distinct Citation Profiles

Here is the breakdown showing the citation distribution for each AI search surface, with Gemini and Perplexity showing a strong preference for authority sites.

Gemini

• 26% institutional sites
• 23% .org
• 13% .gov
• 0.2% UGC

Perplexity

• 86% of brand mentions appear in position 5 or earlier
• 30% of citations from institutional medical, government, encyclopedic, and publisher sources
• 22% institutional sites
• 3.2% .edu
• 1.5% UGC sites

ChatGPT

• Top 10 sources account for 18.5% of citations
• 20% .org
• 12% .gov
• 0.5% UGC

Google AI Mode

• Top 10 sources account for 19.4% of citations
• 14% institutional sites
• 7% UGC

Google AI Overviews

• 18% UGC
• 10.6% of citations from a single video platform
• 10% institutional sites
• 2.9% from a forum platform

Google AI Is Not One System

Google’s AI Mode and Ai Overviews show almost the same websites, with a 59% rate of overlap of cited websites. Gemini has the least amount of overlap.

• Gemini vs AI Overviews: 34%
• Gemini vs AI Mode: 27%

These differences show that the Google’s AI systems rely on different mixes of sources, with Gemini showing the widest amount of difference.

Takeaways

The data makes it easy to view each AI search surface with a shorthand description of what kinds of sources each AI engine tends to cite. There is a wide variance in source citations with clear preferences of which kinds of sites each engine prefers to link to. If there is one big takeaway from the data, in my opinion it would be the importance of establishing a brand connection to products and services.

Other Takeaways

• Gemini and Perplexity rely on high authority brand and institutional websites.
• ChatGPT cites a broader range of sources, showing a higher mix of websites.
• Google’s AI Overviews cites UGC sites more than any other AI search.
• Gemini shows the least amount of overlap among the three Google AI systems.
• AI Overviews and AI Mode show the highest level of overlap.
• Citation overlap varies widely across all five AI engines, indicating major differences in source selection.

Read the BrightEdge report: Why AI Engines Cite Different Sources but Recommend the Same Brands

Featured Image by Shutterstock/Toey Andante