BrightEdge AI Catalyst analysis reveals that ChatGPT, Perplexity, Gemini, Google AI Mode, and Google AI Overviews operate with fundamentally different editorial personalities when selecting the sources they cite. At the same time, the brands named in AI-generated answers remain far more consistent across engines than the sources those engines use to construct those answers. The gap between how engines source and what engines recommend is the single most important pattern for any brand building an AI search strategy.

The prevailing assumption in AI search is that each engine requires its own playbook because each engine behaves differently. The data confirms the engines do behave differently, in some cases by close to two orders of magnitude. But the consistency on the output side, which brands get named in the final answer, tells a different story. The playbook does not need to be fragmented by engine. It needs to be organized by source layer.

This is the latest installment in our BrightEdge AI Catalyst research series. We analyzed citations and brand mentions across ChatGPT, Perplexity, Gemini, Google AI Mode, and Google AI Overviews, drawn from prompts spanning ten industries including B2B technology, education, entertainment, finance, healthcare, insurance, restaurants, travel, and ecommerce. The patterns that emerged are directly relevant to any brand planning AI visibility at scale.

Data Collected

Key Finding

AI search engines are often discussed as if they behave similarly because they produce a similar kind of output: a synthesized answer with citations. The BrightEdge AI Catalyst data shows that behind the surface, the five engines pull from meaningfully different parts of the web. The share of citations coming from authoritative sources ranges from 10% to 26%, depending on the engine. The share coming from user-generated content ranges from 0.2% to 18%, roughly a 90x spread across engines answering the same categories of questions. Despite that divergence in sourcing, the brands those engines recommend cluster in a much tighter range. Pairwise top-100 overlap in named brands across engines falls between 36% and 55%, a 19-point spread, while pairwise top-100 overlap in cited sources ranges from 16% to 59%, a 43-point spread. Source agreement between any two engines varies widely and inconsistently. Brand agreement is consistently steady. The implication for brand strategy is that the path AI takes to reach its answer matters less than most strategies assume, but being present across the three distinct source layers that feed those paths matters more than strategies typically account for.

Five AI Engines, Five Sourcing Personalities

Gemini functions as a formal institutional recommender. Gemini shows the strongest bias toward authoritative sources of any engine in the dataset. Approximately 26% of Gemini's citations come from government domains, academic institutions, and major industry institutional bodies combined. UGC and social content makes up only 0.2%. The authority-to-UGC ratio is roughly 130 to 1, the highest in the study. Gemini also shows the highest .gov share of any engine at roughly 13%, paired with a .org share of 23%. The engine behaves as a conservative, list-oriented recommender that leans on trusted institutional voices and tends to produce longer, more inclusive brand lists than other engines.

ChatGPT acts as a long-tail editorial engine. ChatGPT cites the flattest source distribution of any engine in the study. Its top 10 most-cited domains account for only 18.5% of total citations, meaningfully lower than Perplexity (26.7%), Gemini (26.3%), or AI Mode (19.4%). ChatGPT also has almost no UGC presence (0.5%) and pulls heavily from government and .org domains (12% and 20% respectively). The engine reads as a formal editorial assistant with a long, diverse tail of corporate, institutional, and government sources.

Perplexity behaves like a research librarian. 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. Perplexity shows the highest share of .edu citations (3.2%) and the highest share of international country-code domains (4.4%) in the dataset, reflecting a more formal and globally sourced material mix. It also names brands earliest of any engine, with 86% of its brand mentions landing in position 5 or earlier. Perplexity behaves like an engine that commits to a short, authoritative shortlist rather than producing an exhaustive list.

Google AI Mode operates as a broad commercial aggregator. Google AI Mode pulls from a wider catalog of unique domains than most other engines, with a long-tail distribution that spreads citations across far more sources than its siblings. It also distributes its citations more evenly across source types than any other engine in the study, showing the strongest mix of review aggregators, finance data sources, and news media citations in the dataset. UGC exposure is moderate at roughly 7%, well above ChatGPT or Gemini but well below AI Overviews. AI Mode's top 10 citation concentration is among the lowest at 19.4%, reinforcing its identity as a long-tail, balanced commercial surface.

Google AI Overviews is a UGC-first engine. Google AI Overviews stands apart from every other engine in the study. Approximately 17.5% of its citations come from user-generated content platforms, 35x higher than ChatGPT (0.5%) and 87x higher than Gemini (0.2%). A single video platform accounts for roughly 10.6% of all AI Overviews citations on its own, and a single forum platform adds another 2.9%. Authoritative sources, including government, academic, and major institutional bodies, account for only 9.5% of AIO citations combined. AI Overviews is the only engine in the dataset where UGC citations outweigh authoritative citations.

Authority Share Versus UGC Share, by Engine

The Two Google Engines Are Not the Same Engine

Among the five engines studied, the two most similar are Google AI Mode and Google AI Overviews, with a top-100 citation overlap of roughly 59%. But Gemini, also a Google product, behaves very differently from its siblings. Gemini's top-100 citation overlap with AI Mode is only 27%, and with AI Overviews only 34%. Gemini actually has more in common with ChatGPT (39% overlap) than with the Google search-embedded surfaces. In practical terms, "Google AI" is not one thing. The search-embedded surfaces lean heavily on commercial and UGC content, while standalone Gemini behaves like a conservative, authority-heavy reference engine. Any brand strategy that treats all three Google AI surfaces as interchangeable will miss the actual sourcing patterns driving visibility on each.

The Brand Convergence Signal

The most consequential finding in the study is not the divergence in sources. It is the convergence in brand recommendations despite that divergence. Pairwise top-100 overlap in cited sources across engines ranges from 16% to 59%, a 43-point spread. Pairwise top-100 overlap in named brands ranges from 36% to 55%, a 19-point spread. In every pairwise comparison, brand overlap falls in a tighter, more predictable range than source overlap. The engines disagree substantially and inconsistently about where to pull information from. They agree more consistently about which brands belong in the final answer. That pattern is what makes a unified strategy viable across all five engines, rather than five separate playbooks.

Sentiment Is Overwhelmingly Positive Across Every Engine

Brand sentiment in AI-generated answers skews positive on all five engines, but not uniformly. Gemini is the most positive at roughly 96% positive sentiment, with only 0.3% negative. ChatGPT sits at 94% positive with effectively zero negative mentions. Perplexity shows the highest neutral share at 11%, consistent with its more journalistic, reference-oriented posture. The Google search-embedded surfaces (AI Mode at 93% and AI Overviews at 89%) show slightly higher negative sentiment (1.7% and 2.1%), which reflects their deeper pull from UGC and commercial commentary sources where critical framing more commonly appears. Across the dataset, negative brand mentions remain a marginal share of total volume, which reinforces that visibility in AI answers is almost always presented in a positive or neutral frame.

What Marketers Need to Know

AI engines pull from three distinct source layers, and every engine uses all three. Authoritative sources include government, academic, and major industry institutional content. Commercial and editorial sources include review sites, comparison content, trade press, news media, finance data, and retailer listings. UGC includes video content, forum threads, community discussion, and creator coverage. No engine uses only one layer. The engines differ in how they weight each layer, not in whether they use it. A brand visibility strategy built around only one layer, no matter which layer, will underperform on engines weighted toward the other two.

Authority is category-relative. "Authoritative" does not mean .gov or .edu for every brand. Not every company can or should aim to be cited by federal agencies or academic institutions. Every category has its own authoritative layer: trade associations, analyst firms, expert publishers, standards bodies, professional associations, and institutional voices trusted within the vertical. The strategic question is which authoritative sources serve as the backbone of AI citations in your specific category, and whether your brand is covered by those sources.

Commercial and editorial presence is the widest visibility lever. Across all five engines, the brand/corporate and commercial/editorial source layer accounts for the largest share of citations, ranging from roughly 37% on Gemini to 51% on AI Overviews. 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.

UGC is non-negotiable for AI Overviews and still meaningful elsewhere. The AI Overviews surface draws roughly 18% of its citations from user-generated content, but UGC is not zero on other engines either. Perplexity pulls 1.5% of its citations from UGC, AI Mode pulls 7%, and both represent real retrievable impressions in categories where community and creator content is strong. A UGC strategy does not mean "produce short-form video." It means understanding which videos, forum threads, and community discussions AI is already citing in your category, and being present in that conversation with authority.

Weight investment based on which engines matter most to your buyers. The three-layer framework is universal. The emphasis is not. A B2B SaaS brand whose buyers rely heavily on ChatGPT and Perplexity will prioritize authority and commercial coverage, with UGC as a supplemental layer. A consumer brand whose buyers use AI Overviews heavily will prioritize UGC and commercial presence, with authority as reinforcement. Brand tracking at the engine level, not just in aggregate, is how those priorities get set and validated.

Engine overlap patterns should inform where you measure first. The two Google search-embedded surfaces share roughly 59% of their top-cited sources, so visibility gains on one frequently translate to the other. Gemini behaves more like ChatGPT than like its Google siblings, so brand teams should not assume that a Gemini strategy is a Google strategy. These overlap patterns are not intuitive, and brands that map their measurement plan against actual engine behavior will catch gaps that aggregate tracking hides.

Technical Methodology

Key Takeaways