Why the same social citation reads as the crowd on one engine and a credible authority on the other, and what that means for AEO strategy

Marketers have known for a while that AI search leans on social platforms. Reddit and LinkedIn in particular show up again and again in AI-generated answers. That much is not news. The useful question is not whether these channels get cited. It is where they get cited, what they get cited for, and which sources they get cited alongside. Those answers turn out to be very different depending on the engine.

We used BrightEdge AI Hyper Cube to pull the full universe of prompts where Reddit and LinkedIn earn citations on Google AI Overviews and ChatGPT, then classified each prompt by topic, query intent, functional job, and the other sources cited alongside the social channel. The pattern is clear: ChatGPT and Google AI Overviews do not use these two channels the same way. One engine treats a Reddit citation as the voice of the crowd. The other treats it as a credible reference, sitting it beside the most authoritative publishers on the web.

This is the kind of nuance an engine-agnostic view of AI search will miss. Earlier research in this series showed that AI engines assign functional roles to the biggest sites on the internet, citing Reddit less as a forum and more as a consumer-opinion and product-research layer. This installment goes one level deeper, into exactly how two engines diverge in the way they deploy the two social channels marketers ask about most.

What We Analyzed

We analyzed the prompt and citation universe for Reddit and LinkedIn across Google AI Overviews and ChatGPT, spanning both consumer and professional topics. Every prompt was classified four ways: by topical cluster, by query intent, by functional job-to-be-done (the kind of question being asked), and by co-citation neighborhood (the other brands and sources cited in the same answer). The goal was to understand not just that these channels get cited, but the specific conditions under which each engine reaches for them.

Data Collected

Key Finding

The same social channel plays a different role on each engine. On Google AI Overviews, Reddit is cited as part of a social pack: YouTube appears alongside it in roughly 36% of citations, with Facebook, TikTok, and Instagram close behind, while editorial authorities appear next to it only about 6% of the time. On ChatGPT, the pattern nearly inverts. Reddit is cited beside Healthline, Mayo Clinic, Cleveland Clinic, and Encyclopedia Britannica, with authoritative publishers flanking it about 36% of the time and other social barely registering. Same channel, opposite standing.

The functional picture reinforces it. Both engines cite these channels mainly for how-to, definitional, and verification questions, but ChatGPT leans on them far harder for procedural how-to and causal why answers, while Google AI Overviews is the engine that surfaces them for head-to-head comparison queries. The implication for marketers is that a Reddit or LinkedIn presence is not one asset with one value. It is an asset whose value depends entirely on which engine is reading it and what job the user is doing.

The Same Reddit Citation Lives in Two Different Neighborhoods

The clearest signal in the data is the company Reddit keeps. We measured how often a Reddit citation appears next to other social and UGC platforms versus next to editorial authorities, and the two engines come out as near mirror images of each other.

On Google AI Overviews, Reddit sits inside a crowd. YouTube is the dominant neighbor, and the rest of the pack is Facebook, TikTok, Instagram, and Quora. The engine is effectively grouping Reddit with other places where people post, treating it as one more voice in the user-generated layer.

On ChatGPT, Reddit keeps very different company. Its most frequent co-citations are Healthline (around 12% of Reddit-cited answers), Mayo Clinic (around 9%), Cleveland Clinic (around 8%), and Encyclopedia Britannica, with Medical News Today, Verywell Health, WebMD, and the CDC close behind. The engine is slotting Reddit into the same answers as the most trusted reference publishers on the web. For a marketer, that is the difference between background noise and borrowed credibility.

LinkedIn Keeps Professional Company on Both Engines

LinkedIn does not show the same dramatic flip, and that is itself a finding. On both engines its co-citation neighbors are professional: career and education platforms like Indeed (roughly 11% of LinkedIn citations on AI Overviews), ZipRecruiter, Coursera, Udemy, and LinkedIn Learning. Editorial authorities sit next to LinkedIn rarely on either engine, around 3% on AI Overviews and 5% on ChatGPT. The role is consistent rather than inverted: both engines file LinkedIn as a professional and career source.

One detail stands out. On ChatGPT, the single most common co-citation inside LinkedIn-topic answers is Reddit itself, appearing in roughly 15% of those answers. ChatGPT reaches for Reddit to round out professional answers far more than Google AI Overviews does, which means the two channels are not always competing for the same slot. Sometimes they share it.

What These Channels Get Cited For

Looking only at prompts that carry a clear question or intent, the functional jobs these channels perform are mostly shared, with a few sharp differences. The table below shows the share of intent-bearing citations by functional job.

How-to is the swing job, and ChatGPT leans on social much harder for it. Reddit how-to citations roughly double from AI Overviews to ChatGPT, and LinkedIn climbs from about 22% to 33% of intent-bearing prompts. ChatGPT also pulls Reddit for causal why questions (why something happens, why a symptom appears) more than twice as often as AI Overviews does.

Comparison is a Google AI Overviews specialty. Around 10% of social citations on AI Overviews are head-to-head comparison prompts (americano vs latte, premium vs Sales Navigator). On ChatGPT it is about 1%, because the engine tends to synthesize the comparison itself rather than pointing to the thread where humans debated it.

Verification is everywhere, and the two channels do it differently. It accounts for 14% to 24% of citations across the board. On LinkedIn it is platform-capability checking (Can I unsend a LinkedIn message?). On Reddit it is consumer permission and reassurance (Can dogs have corn? Is this normal?).

One caution on reading this table: the labels capture the shape of the question, not always the reason the social source was pulled. A how-to prompt about a home remedy or a product setup is procedural on its face, but the reason Reddit gets cited for it is often the lived experience in the thread underneath. The experiential value of these channels is real, and much of it hides inside the how-to and verification buckets.

The Topics Each Channel Owns

The topical split is the most intuitive part of the picture and it holds across both engines. LinkedIn earns its citations in professional contexts: careers and recruiting, professional skills and online learning, platform how-to, business-to-business and sales topics, and term definitions. Reddit earns its citations in broad consumer contexts, but the consumer mix shifts by engine. On ChatGPT, Reddit skews toward health and medical questions, money and finance, definitions, and food. On Google AI Overviews, it skews toward entertainment and media, gaming, food, and tech. The health and finance concentration on ChatGPT is what produces the authority-publisher neighborhood described above. When the question is medical, ChatGPT pulls Reddit and Mayo Clinic into the same answer.

Intent and Tone

Informational intent dominates everywhere, and ChatGPT leans into it harder, accounting for roughly 80% to 85% of its citations versus about 65% to 70% on AI Overviews. The more commercially interesting band is consideration intent, which runs roughly 9% to 14% across all four cuts. That is the slice closest to a buying decision and the one marketers should care most about. Transactional intent is thin everywhere, in the low single digits, so neither channel is earning citations at the point of purchase. They are upper and mid-funnel assets.

There is also a tone difference worth noting. When LinkedIn is named as a brand in an answer, ChatGPT speaks about it positively far more often than AI Overviews does, roughly 46% of the time versus about 31%. Reddit is cited more neutrally on both engines, as a reference point rather than an endorsement. ChatGPT, in other words, is more willing to frame LinkedIn as a recommendation.

What Marketers Need to Know

Reddit is your highest-leverage credibility play on ChatGPT. Because the engine cites it next to Mayo Clinic and Healthline, a strong, well-upvoted Reddit thread can punch at the weight of an editorial citation there, especially in health, finance, and other research-heavy categories. That same thread on Google AI Overviews mostly buys a seat in a crowded social pack. Different engines, different value from the exact same content.

Match the channel to the job, not to the logo. LinkedIn earns citations for professional how-to and capability questions. Reddit earns them for consumer how-to, comparison, and lived experience. Decide which channel to invest in based on the question you are trying to win, then build the asset that answers it.

Win the question, not just the brand name. Citations flow to content that answers how do I, can you, and is it worth it, not to a bare brand mention. Build for the underlying job and the brand mention comes with it. A thread or post that resolves the actual question is far more citable than one that simply names the product.

Audit by engine, not in aggregate. The same channel is an authority on one surface and background noise on another. A blended, cross-engine view averages that difference away and hides it. Look at ChatGPT and Google AI Overviews separately to see the real role each channel plays in your category.

Treat comparison content as a Google AI Overviews opportunity. If you produce head-to-head comparison content, AI Overviews is where the social version of that conversation surfaces. Seeding credible comparison discussion where the crowd debates pays off disproportionately on that surface.

Technical Methodology

Key Takeaways

The conventional wisdom said AI search would force marketers to build a parallel optimization discipline. AEO, GEO, llms.txt, content chunking, AI-specific rewrites. A whole new playbook for a whole new surface. The data tells a different story. Across seven major industries on Google organic search and AI search engines, the underlying jobs users are trying to accomplish are largely identical. Surface mechanics differ. The job universe does not.

This finding aligns directly with Google's updated AI Optimization Guide, published just before Google I/O. The guide states plainly: "From Google Search's perspective, optimizing for generative AI search is optimizing for the search experience, and thus still SEO." Google's own one-line summary of how to succeed: "Focus on what your visitors would enjoy, find helpful, and feel satisfied with after visiting your website." A user-centered content strategy is the strategy. The surfaces sort themselves out.

BrightEdge Data Cube X organic keyword data and AI Hyper Cube prompt data, analyzed across seven dominant brands in seven major industries, show that the underlying user jobs are present on both surfaces. When we collapse the traditional five-bucket intent taxonomy into three jobs-to-be-done | Learn, Decide, and Act | 85% of queries fall into the same job-to-be-done bucket whether the user is on organic search or AI search. When we go deeper to an 11-job taxonomy that travels across categories, 8 of the 11 specific jobs appear meaningfully on both surfaces. Users are users. Same jobs. Different doors.

This is the third installment in our funnel-shape research series. The first installment examined how the four classical query intent categories survived into AI search, each reshaped to fit the medium. The second installment looked at the same data through the lens of the consumer journey and found that brand-owned content dominates the middle of the funnel. This installment steps back to the system level: across the entire query universe and both surfaces, do users come to organic and AI search for the same underlying reasons? The answer is yes, and the implications for content strategy are significant.

We analyzed the full keyword universe and AI prompt universe for top brands across seven industries: Retail, SaaS, Healthcare, Insurance, Finance, Travel, and Education. Each keyword and prompt was classified into one of 11 generalized jobs-to-be-done, then grouped into three job buckets. The findings are directly relevant to any brand building a unified content strategy that needs to perform across organic search and AI surfaces.

Data Collected

Key Finding

The job universe is the same on organic search and AI search. Across seven industries pooled, 85% of queries fall into the same job-to-be-done bucket whether the user is on organic search or AI search. Eight of the 11 specific jobs in the unified taxonomy appear meaningfully on both surfaces. The Learn bucket dominates both. The Decide and Act buckets are present on both. The implication for marketers is that there is no separate AI playbook to build. The same user-centered content strategy that wins on Google organic also makes a brand eligible across AI Overviews, AI Mode, ChatGPT, Perplexity, and Gemini. What changes between surfaces is the grammar of the query (keyword-shaped on organic, conversation-shaped on AI) and the mechanics of how each engine retrieves and presents the answer. The underlying user need is the constant.

What the 11-Job Taxonomy Looks Like

Users come to search to do one of three things: learn something, decide between options, or get something done. Within those three buckets, eleven specific jobs travel across every industry we measured. Healthcare's "find a clinic near me" and Finance's "atm near me" are the same job. Retail's "what is back to school sales" and SaaS's "what is CRM software" are the same job. The taxonomy generalizes.

Learn (the largest bucket on both surfaces).

Decide (the evaluation moment).

Act (the conversion moment).

The Learn bucket is the dominant share of activity on both surfaces. The Decide bucket is meaningfully present on both. The Act bucket is present on both, often expressed through different specific jobs (organic search carries more direct-transactional language, AI search carries more planning and managing language). In every case, the underlying user need is identifiable in both query sets.

What Changes Across Surfaces: Grammar, Not Goals

The traditional five-bucket intent taxonomy (informational, navigational, commercial, transactional, consideration) makes the two surfaces look more different than they are. Organic data is heavy with navigational queries because people type "walmart" into Google to get to the site. AI has near-zero navigational queries because nobody asks ChatGPT to navigate them anywhere. The standard taxonomy interprets this as divergent user behavior. It is not.

"Walmart hours" typed into Google and "what time does Walmart close" asked of ChatGPT are the same job: Get Hours / Specs / Status. The user is doing the same work. The grammar of the query is different because the surface is different. Organic search has trained users to drop articles, verbs, and natural language because the keyword-matching paradigm rewarded brevity. AI search has trained users to write full sentences because the conversational paradigm rewards specificity. The keywords look different. The job is the same.

This grammar difference is the single largest source of apparent misalignment between the two surfaces. When we collapse navigational and informational queries into a unified Learn bucket (because they answer the same underlying user need), the surfaces snap into alignment. 85% of queries fall into the same job bucket across both.

Decide-Stage Behavior Shows Up Differently

Within the strong alignment, one pattern stands out: Decide-stage queries are more visible on AI surfaces than on organic. Across the seven industries pooled, the Decide bucket accounts for a meaningfully larger share of AI prompt activity than it does of organic keyword activity. This is consistent with what BrightEdge's prior funnel-shape research found: the consideration stage of the funnel is real on both surfaces, and AI consolidates evaluation behavior in ways organic search distributes.

This is not a sign that user goals have changed. It is a sign that the consideration journey has changed shape. The buyer asking "what's the cheapest car insurance for a young driver" or "best CRM for small business" used to do that work across review sites, comparison aggregators, and forum threads. Now it consolidates into a single prompt and a single answer. The job is the same. The path through it is shorter and more visible.

For marketers, this is the most actionable finding from the cross-surface analysis. The brands that organize content around the full job taxonomy | including the Decide-stage jobs of Compare Options, Validate Choice, and Get Recommendation | are positioned to be retrieved by AI surfaces when that consolidation happens. The brands that have historically optimized only for the Learn-stage and Act-stage queries on organic, leaving consideration content to third-party reviewers and aggregators, have a gap to close.

Industry-Specific Patterns

The taxonomy travels across industries, but the mix of jobs varies by category in predictable ways. Some industries are Learn-dominant. Some carry meaningful Decide-stage volume. The shape of the user journey differs by what the user is shopping for.

Healthcare and Education. These categories are overwhelmingly Learn-dominant on both surfaces. Users come to search for symptoms, definitions, treatments, courses, degree programs, and how-to guidance. The Decide and Act buckets are small. The implication is that the user-centered content strategy here is depth and breadth of educational coverage. The brands that win are the ones whose content actually teaches.

Finance and Insurance. These categories carry meaningful Decide-stage volume on both surfaces, with AI carrying a larger share than organic. The user journey involves significant comparison and evaluation before commitment. Compare Options ("best high yield savings accounts"), Validate Choice ("is Otto insurance legit"), and Get Recommendation ("what credit card should I get") are core jobs. Brand-owned content that addresses these jobs directly | comparison tables, transparent product details, defensible claims about coverage or rates | is the leverage point.

Retail and Travel. These categories show a more even distribution across Learn, Decide, and Act. Users research, compare, and transact, often within the same session. Plan a Trip / Activity is a significant Act-stage job that is largely unique to Travel. The implication is that content needs to serve the full journey, from category exploration to specific destination planning to booking-stage information.

SaaS. B2B software shows a mix of Learn-stage definitional content ("what is CRM software," "what is account management") and Decide-stage evaluation content ("best project management tools," "small business CRM software"). The user journey is research-heavy and consideration-heavy, with the Act stage often deferred to a sales conversation outside the search session. Brand-owned product education and buyer's guide content does most of the work.

What Marketers Need to Know

The jobs are the same across surfaces. The brands that organize content around what users are trying to do, rather than around the surface they happen to do it on, build the most durable AI search strategy. Google's updated guide says it directly. The data confirms it across seven industries.

Your strategy does not bifurcate. There is one user, one job universe, and multiple surfaces. Foundational SEO is the cost of entry to AI visibility. A page that cannot be crawled, indexed, and retrieved by Google Search cannot be retrieved by an AI surface that draws from the same index. The technical and content fundamentals that make a brand visible on organic are the same fundamentals that make it eligible across AI surfaces.

Cover the full job taxonomy. The 11 jobs above are durable across categories. The brands that win in AI search are the ones whose content actually serves all of them, not just the head terms or the brand keywords. Audit your content coverage against Learn, Decide, and Act. Where you have gaps, you are invisible to AI surfaces when users do those jobs.

Build for users, not for surfaces. Google's own guidance is clear: focus on what visitors find helpful and satisfying. The same content that satisfies the job on organic search is what gets retrieved by AI surfaces. There is no separate AI playbook to build. Skip the llms.txt files, the content chunking, the AI-specific rewrites. Write for the user. The surfaces will follow.

Measure across surfaces, not against them. Yes, there are things you can do to make content more visible in AI experiences. But the fundamental content strategy does not change. What changes is your ability to see how that strategy performs everywhere it lives. That requires one platform connecting organic search, AI Overviews, AI Mode, ChatGPT, Perplexity, and Gemini into a single view of how users are finding you. This is what BrightEdge is built for.

Expect surface-specific grammar, not surface-specific intent. Users phrase queries differently on AI than on organic, but the underlying need is the same. Content that answers the job comprehensively, in plain language, with clear structure and entity clarity, satisfies the user regardless of how they ask. Optimizing for the job is the optimization.

Technical Methodology

Key Takeaways

The conventional wisdom said AI search would compress the funnel and shift citations toward third-party reviewers and aggregators in the consideration phase. The data tells a different story. Across eight industries on Google AI Overviews and ChatGPT, the middle of the funnel still represents meaningful volume, and brand-owned content takes the dominant share of citations there.

BrightEdge AI Hyper Cube analysis across Google AI Overviews and ChatGPT shows that the consideration stage of the funnel is alive, measurable, and varies dramatically by industry. The size of the middle ranges from 4% to 26% of AI search demand depending on the category and the engine. Inside that middle, brand-owned pages account for 42% to 79% of citations across every industry studied, while review and comparison aggregators account for 1% to 7%. The opportunity for brands is on their own domains.

This is the second installment in our funnel-shape research series. In the prior installment we examined how the four classical query intent categories survived into AI search, each reshaped to fit the medium. That analysis mapped consideration-stage queries to the classical commercial intent bucket. This installment looks at the same data through the lens of the consumer journey: top, middle, and bottom of the funnel. The terminology shifts from "commercial intent" to "consideration stage" because the questions we are answering this time are about consumer-journey shape, not classical SEO intent taxonomy. The underlying data is consistent with the prior piece.

We analyzed the full prompt universe across eight industries on both engines: B2B, Ecommerce, Education, Entertainment, Finance, Healthcare, Insurance, and Travel. Each prompt was classified by the BrightEdge Generative Parser into a funnel stage. Citations from consideration-stage prompts were categorized by source type. The findings are directly relevant to any brand planning content strategy for AI search across multiple industries.

Data Collected

Key Finding

The middle of the AI search funnel is real volume in every industry, and brand-owned content owns it. Volume-weighted, the consideration stage represents between 4% and 26% of AI search demand across the eight industries studied. Travel, Ecommerce, B2B, and Finance show the largest middles. Healthcare and Entertainment show the smallest. Inside that consideration stage, brand-owned commercial pages take between 42% and 79% of all citations across every industry on both engines. Review and comparison aggregators, the source type marketers most often assume dominates the research phase, take between 1% and 7% of citations in most categories. The implication for marketers is that the buying-guide content, category explainers, and comparison pages on a brand's own domain are the highest-leverage AI search assets in the middle of the funnel. Outsourcing the consideration phase to third-party reviewers leaves the dominant citation channel uncovered.

The Size of the Middle, by Industry

Volume-weighted share of AI search demand classified as consideration stage:

Two patterns stand out. First, the size of the middle of the funnel varies by a factor of six between the largest category and the smallest. Marketers who assume the consumer journey looks the same across industries are missing where the opportunity actually concentrates. Second, AI Overviews consistently shows a larger consideration stage than ChatGPT in commercial categories. In Travel, Ecommerce, B2B, Finance, and Insurance, AIO's middle is meaningfully bigger than ChatGPT's. In Education and Entertainment, the two engines look roughly the same. The pattern suggests AIO is woven into the buying journey in commercial categories in a way ChatGPT is not, despite the popular narrative that consumers have moved their research behavior to conversational AI.

What a Consideration Prompt Looks Like

Consideration prompts capture users who are comparing options or evaluating a category without committing to a specific brand or product. Examples drawn from the data, generalized for clarity:

In Travel: "best beach vacations for families," "top all-inclusive resorts in Mexico," "cheapest time to fly to Europe."

In Ecommerce: "best treadmill for home use," "treadmill vs elliptical for cardio," "best mattress for back pain."

In B2B: "small business CRM software," "best project management tools for remote teams," "top cloud storage providers for enterprise."

In Finance: "best high yield savings accounts," "Roth IRA vs traditional IRA," "top robo-advisors."

In Insurance: "term vs whole life insurance," "best homeowners insurance companies," "cheapest car insurance for new drivers."

These queries differ from branded queries (which name a specific product or company) and from transactional queries (which signal readiness to act). The defining characteristic is comparison and evaluation. The user is figuring out what to want, not which one to click.

Brand-Owned Content Dominates Consideration-Stage Citations

Across every industry studied and both engines, brand-owned commercial pages take the largest share of citations in the consideration stage. The range is 42% on the low end (Healthcare, where authority sites take a meaningful slice) to 79% on the high end (Travel ChatGPT, where the engine routes consideration queries heavily to brand domains and bypasses online travel agencies). Most industries land between 50% and 70%.

This finding pushes back on a widely held assumption in the AEO and GEO community. The assumption was that AI engines, when faced with a comparison query, would lean on third-party reviewers and aggregators to make the recommendation. The data shows the opposite. The brand's own buying guide, category explainer, or comparison page is more often the cited source than a review aggregator.

The pattern is consistent across engines, with one nuance. AIO concentrates citations across a small number of brand-owned domains. ChatGPT distributes citations across a wider set of brand-owned domains. The dominance of brand-owned content holds in both cases, but the competitive dynamics are different. On AIO, winning a consideration-stage citation in a given category means displacing a small number of established players. On ChatGPT, winning a citation is more accessible, but the citation share per win is smaller.

Review and Comparison Aggregators Are Not the Dominant Source

The source type that conventional AEO wisdom positioned as the natural winner of the consideration stage, review and comparison aggregators, accounts for between 1% and 7% of citations in most industries on both engines. The two exceptions are B2B on ChatGPT, where software review and comparison sites take a slightly larger share, and Finance on AIO, where financial product comparison sites cluster around the high end of the range. Even in those exceptions, brand-owned pages still take more citations than aggregators.

This does not mean third-party reviews are irrelevant. They influence the brand recommendations AI engines surface and they remain important for trust signals. But the citation slot, the actual source AI engines link to in the consideration stage, more often belongs to a brand's own domain. Marketers who have built their AI search strategy primarily around earning third-party reviewer mentions are competing for a small share of the citation channel.

Citation Concentration: AIO Concentrates, ChatGPT Distributes

Citation concentration in the consideration stage differs substantially between engines. On Google AI Overviews, a small number of domains accounts for the majority of citations in any given industry. On ChatGPT, citations spread across a much larger set of domains for the same industries.

The pattern means winning consideration-stage citations on AIO requires going deeper on a smaller number of pages within a category. The competitive set is narrow. Once a brand earns a citation slot, it tends to hold it across many related prompts. ChatGPT is the opposite. The citation pool is more democratic. Breadth of topical coverage, distinctive perspectives on a category, and content depth across many comparison angles matter more than dominance on a single page.

For content strategy, this means the optimization approach differs by engine. On AIO, the priority is identifying the small number of pages that win the highest-volume consideration queries in a category and concentrating optimization investment there. On ChatGPT, the priority is breadth, coverage across the full comparison landscape, and content depth that signals authority across many sub-topics within a category.

A Note on Google AI Overviews and When They Trigger

Google AI Overviews do not appear on every search. AIO is triggered only when Google decides an AI Overview is the right response format for a given query. Many consideration-stage searches return a traditional organic results page with no AIO at all. The analysis in this study measures the subset of consideration queries where Google has chosen to deploy an AIO.

This caveat actually strengthens the finding rather than weakening it. Even on the consideration queries Google has decided merit an AI Overview, the citation slots are not spreading across third-party reviewers and aggregators. They are concentrating on brand-owned pages. Whatever combination of signals Google uses to decide when to deploy AIO and what to cite inside it, the result is that brand-owned content is the dominant beneficiary in the consideration stage.

ChatGPT does not have an equivalent trigger condition. Every prompt receives a response. The full consideration-stage volume on ChatGPT is measured directly. The fact that brand-owned content dominates on both surfaces, despite the very different mechanics of how AIO and ChatGPT decide what to cite, reinforces the strength of the underlying pattern.

Industry-Specific Patterns

Healthcare. Healthcare authority sites (major medical centers, government health agencies, established medical reference sites) take a larger share of citations than in any other industry, between 26% on AIO and 36% on ChatGPT. Even so, brand-owned commercial pages still take the largest single share. The takeaway for healthcare marketers is that competing for citation slots means competing against highly credentialed authority sources, which makes E-E-A-T signals (expertise, experience, authoritativeness, trust) even more important in this category than elsewhere.

Travel. Travel shows the most divergent engine behavior. On AIO, online travel agencies and booking aggregators take a meaningful slice of consideration citations (around 24%). On ChatGPT, the engine bypasses OTAs and routes consideration citations directly to brand-owned destinations (79% brand-owned). For travel brands, this means a ChatGPT optimization strategy that targets brand-owned travel content can win significant citation share, while an AIO strategy needs to plan for OTA competition in the citation slot.

B2B. B2B shows the cleanest gap between AIO's larger middle of the funnel (22%) and ChatGPT's smaller middle (9%). The implication is that B2B buyers are using AIO for category exploration more than they are using ChatGPT for the same purpose, at least in the consideration stage. Software review aggregators have a slightly more prominent role here than in other categories, but brand-owned product pages and buyer's guides still take the largest share.

Education and Entertainment. These are the only two industries where ChatGPT's middle of the funnel is larger than or equal to AIO's. Both categories also show meaningful citation share for video platforms (10% to 17% on AIO) and UGC sources (10% to 13% on both engines). The pattern suggests that for educational and entertainment decisions, users are pulling in more diverse source types than in commercial categories.

What Marketers Need to Know

The middle of the funnel is real volume in AI search. The size varies by industry and by engine, ranging from 4% to 26% of total demand. In Travel, Ecommerce, B2B, and Finance, the consideration stage represents a meaningful share of AI search demand on both engines and should be a primary focus for content strategy.

Your own content is the opportunity. Brand-owned commercial pages take 42% to 79% of consideration-stage citations across every industry studied. The category guides, comparison pages, and buying guides on your own domain are doing the work. Investing in this content is more leveraged than chasing third-party reviewer placements.

Do not outsource the middle to third parties. Review and comparison aggregators take 1% to 7% of consideration citations in most categories. They remain important for trust signals and indirect influence on what AI engines recommend, but the citation channel itself belongs to brand-owned content.

Optimize for both engines differently. AIO concentrates citations across a small number of pages. ChatGPT distributes citations across a wider set. The same brand-owned content strategy serves both engines, but the tactical priorities differ. On AIO, win the small number of pages that own the highest-volume consideration queries. On ChatGPT, build breadth and topical depth across the full comparison landscape.

Audit your consideration coverage by industry. Some industries have much larger middles than others. If you compete in Travel, Ecommerce, B2B, or Finance, the consideration stage deserves significant share of your AI search investment. If you compete in Healthcare, Education, or Entertainment, the middle is smaller, but the source-type dynamics in those categories require category-specific strategy (authority signals in Healthcare, video and UGC presence in Education and Entertainment).

Expect engine-specific behavior, not engine-specific intent. The underlying user behavior in the consideration stage is the same across engines. The way each engine surfaces and cites sources for that behavior differs. A unified content strategy organized around the consumer journey, with execution tuned for each engine's citation dynamics, is more durable than separate engine-specific playbooks.

Technical Methodology

Key Takeaways

BrightEdge AI Hyper Cube analysis across Google AI Overviews and ChatGPT shows that the four classical query intent categories all still exist in AI search. Each one has been reshaped to fit the medium, and the differences between engines reveal where AI is encroaching on the consumer journey beyond research.

BrightEdge AI Hyper Cube analysis reveals that the question of "what is search intent in the AI era" has a more nuanced answer than the prevailing narrative suggests. Informational intent still dominates, as it always has, but navigational, commercial, and transactional intent are all present in measurable share. More importantly, each of the four classical intent buckets has taken on a different shape depending on which AI engine the user is querying. The same underlying user behavior produces fundamentally different query syntax, different content requirements, and different citation patterns across engines.

The prevailing assumption is that AI has flattened query intent into one mode: people ask AI questions, AI answers them, the end. The data shows the four-bucket model from classical search is still alive, just reformulated. AI is also no longer just a research tool. Navigational, commercial, and transactional intent all appear in cited prompt volume across both engines, including on pure-play conversational AI like ChatGPT. That has direct implications for any brand thinking about where AI search fits in the consumer journey.

This is the latest installment in our BrightEdge AI Hyper Cube research series. We analyzed prompts that cited the most-referenced websites on the internet across multiple industries, including ecommerce, healthcare, finance, social, video, encyclopedic reference, and community content. The findings are directly relevant to any brand planning AI content strategy at scale.

Data Collected

Data Collected

Key Finding

The four classical query intent categories that have defined search for two decades are all still present in AI search, but each has been reshaped to fit the medium it operates in. Informational intent has deepened, accounting for 71% of cited volume on Google AI Overviews and 92% on ChatGPT. Navigational intent has split into two completely different behaviors: terse keyword fragments on AI Overviews (53% of AIO prompts are three words or fewer) versus branded questions on ChatGPT ("Is United Airlines good?", "How much is Kindle Unlimited?"). Commercial intent, the research-and-comparison phase before buying, makes up 8% of cited volume on AIO and 3% on ChatGPT. Transactional intent remains the smallest bucket today at 2% to 3% across both engines, with commerce activity in AI search still mostly upper-funnel. The implication for marketers is that AI search is no longer just a research tool. People are using it for navigation, comparison, and even purchase intent, and each engine reshapes those behaviors differently.

Volume-Weighted Intent Distribution by Engine

Four Intents, Reshaped to Fit the Medium

Informational intent has deepened, not diminished. The conventional wisdom that AI is killing informational search has it backwards. On ChatGPT, 92% of cited prompt volume is informational. Users phrase actual questions and expect synthesized answers. On AI Overviews, 71% of cited prompt volume is informational, with the remainder distributed across the other three buckets. The classical "look it up" behavior didn't shrink in the AI era. It concentrated, especially on conversational engines where the interface is purpose-built for question-answering.

Navigational intent changed shape, depending on the engine. On AI Overviews, 53% of cited prompts are three words or fewer. People use AIO as a sophisticated address bar: "tv app," "play music," "amazon prime free shipping," "iphone 14." The intent is to surface a specific known thing, not to ask a question. On ChatGPT, the same intent shows up reformulated as a branded question: "Is United Airlines good?" "How much is Kindle Unlimited?" "How many followers does MrBeast have?" The intent didn't go away. The syntax did. This split has direct implications for content strategy. The same underlying user behavior produces two completely different content requirements depending on which engine they use.

Commercial intent is the research phase before buying. Commercial intent captures users who are comparing options or evaluating categories without being ready to act. Examples drawn from the data include "home pregnancy test," "basic bread maker," "cloud storage service," and "dumbbells for home workout." The query expresses interest in a category or product type without committing to a specific purchase action. The defining test is whether the user is trying to decide what to buy versus buy something specific. Commercial intent makes up 8% of cited AIO volume and 3% of cited ChatGPT volume. It is meaningfully present on both engines today.

Transactional intent remains the smallest bucket. Pure transactional intent (the user is ready to act and the prompt names a specific action) accounts for 2% of cited volume on AIO and 3% on ChatGPT. Examples include "shop holiday decor on sale," "amazon prime video subscription," and "free trial to amazon prime." Where commerce shows up in AI search today, it is still mostly upper-funnel. This is the smallest intent bucket on both engines and represents the part of the consumer journey that AI search has the least share of so far.

AI Engines Have Functional Uses for the Biggest Sites on the Internet

One of the more striking patterns in the data is how AI engines have functionally re-categorized the most-cited websites on the internet. The label that defined these sites for years isn't how AI cites them today.

YouTube on AI Overviews: 81% informational. To the engine, YouTube isn't a "video site." It's a how-to and educational utility. The cited prompts are dominated by "how to" content, tutorials, and explainers where video is the most useful format for the answer, not because YouTube is a video destination.

Amazon on ChatGPT: 80% informational. Even the canonical commerce site on the internet is being cited primarily as a product information source rather than a transaction destination. Users ask ChatGPT product questions, and Amazon listings become a reference layer in the synthesized answer.

Reddit on ChatGPT: 18% commercial, 5.6% transactional. To the engine, Reddit is no longer "a forum." It is the consumer-opinion layer for product research and local intent. Cited prompts include "thai restaurant near me," "buy here pay here," "flights to new york," and "casual dining near me." Local commerce, dining recommendations, and product comparison queries route through Reddit threads on ChatGPT in volumes that would not be predicted by Reddit's brand identity as a discussion platform.

The implication for marketers is direct. The label your site carries based on what it sells or hosts is not the label AI engines apply when they decide whether to cite you. Auditing how AI engines actually use your site, rather than how your site categorizes itself, is the first step in any AI search content strategy.

Why AI Search Looks More Informational on ChatGPT Than on AIO

Two structural reasons explain why ChatGPT shows 92% informational versus AIO's 71%.

First, ChatGPT users phrase intent explicitly. Question-format prompts (those beginning with how, what, why, when, where, who, which, can, does, is) account for 96% of cited prompt volume on ChatGPT, compared with 21% on AI Overviews. When users phrase their query as a question, the intent parser has clear signal to classify the prompt as informational. When users type a two-word fragment, the parser often can't assign an intent, and those fragments end up in a "Not-Applicable" bucket. On AIO, that bucket is meaningfully large because keyword-fragment behavior is much more common.

Second, the classical "navigational" query is largely absent on ChatGPT. You can't ask a conversational engine to "take you" somewhere. The navigational impulse on ChatGPT gets reformulated into branded questions, which the parser usually classifies as informational rather than as a separate navigational category. The result is a higher informational share on ChatGPT not because users have fundamentally different intent, but because the medium forces them to express intent through complete sentences.

Sentiment in Cited Responses Skews Positive Across Both Engines

Brand sentiment in cited prompts is overwhelmingly positive or neutral on both engines. Volume-weighted, ChatGPT shows roughly 55% positive sentiment and 45% neutral sentiment, with negative sentiment below 1%. AI Overviews shows roughly 24% positive sentiment and 76% neutral sentiment, with negative sentiment also below 1%. ChatGPT is meaningfully more opinionated than AIO. The conversational engine treats cited sources as recommendations more often than as neutral references, while AIO treats most citations as neutral information lookups. For brands that win citations on either engine, the framing is rarely negative. Earning the citation correlates strongly with not being criticized.

What Marketers Need to Know

Audit how AI engines actually treat your site, not how you categorize yourself. The biggest sites on the internet have all been functionally re-categorized by AI engines. YouTube is cited as a how-to utility, not a video site. Amazon is cited as a product information source, not a store. Reddit is cited as a consumer-opinion layer, not a forum. The first step in any AI search content strategy is understanding which job AI engines are actually using your site to do.

Plan content for two intent expressions of the same user behavior. On AI Overviews, navigational behavior shows up as three-word keyword fragments. On ChatGPT, the same behavior shows up as a branded question. Brand-anchored content needs to answer both syntactic shapes. A page that wins "iphone 14" on AIO is not necessarily the page that wins "Is the iPhone 14 still worth buying?" on ChatGPT, but the underlying user is the same.

Build content that answers brand questions on ChatGPT, not just brand mentions. On ChatGPT, the classical navigational query has been replaced by branded informational questions. "Is the Toyota Corolla a good first car?" matters more than "Toyota Corolla." Content that earns ChatGPT citations is content that answers the question a user would ask about your brand, not content that merely mentions your brand.

Treat commercial and transactional as separate content jobs. Commercial intent (the research and comparison phase) shows up in real volume across both engines today. Transactional intent remains small. Marketers who collapse "commerce" into a single bucket miss the larger of the two opportunities. Build comparison content, category guides, and "which one should I get" pages to win commercial intent before optimizing for transactional capture.

Stay measured about transactional intent in AI search. Pure transactional behavior in cited AI prompts is 2% to 3% of cited volume across both engines today. Where transactional intent does show up on ChatGPT, it tends to be local commerce and dining queries that pull from Reddit and community sources, not from retailer pages. Marketers planning AI commerce strategy should weight investment toward the part of the funnel where AI is already meaningfully cited (commercial investigation) rather than toward the part it hasn't yet earned (transactional conversion).

Expect engine-specific shape, not engine-specific intent. The underlying intent buckets are the same across engines. The way users express each intent, and the way each engine surfaces and cites sources for it, differs substantially. A unified content strategy organized around the four classical intents, with content shaped twice (once for keyword-fragment surfacing on AIO and once for question-format surfacing on ChatGPT), is more durable than five separate engine-specific playbooks.

Technical Methodology

Key Takeaways

BrightEdge AI Catalyst analysis across five AI search engines shows that the brands AI engines recommend converge tightly in some categories and diverge sharply in others. Where consumers transact, the engines agree. Where consumers research, the engines develop distinct preferences for which brands belong in the answer.

BrightEdge AI Catalyst analysis reveals that ChatGPT, Perplexity, Gemini, Google AI Mode, and Google AI Overviews recommend a strikingly similar set of brands at the aggregate level. Pairwise overlap in top-named brands across the dataset falls in a tight 36% to 55% band. But that aggregate average masks meaningful category-level variance. In retail, travel, and tech, brand agreement across all five engines runs between 88% and 97%. In healthcare and finance, it drops to 60% and 71%. The same engines that converge on the same retailers, hotels, and consumer electronics brands disagree substantially about which hospitals, financial institutions, and authoritative publishers to recommend.

This is the follow-up to our recent BrightEdge AI Catalyst research on cross-engine source and brand convergence. The original study found that AI engines pull from wildly different sources but recommend the same brands. That finding holds at the dataset level. This installment splits the same dataset by category to show where the agreement is real, where the divergence concentrates, and what each engine's editorial signature looks like inside the categories where the engines part ways.

Data Collected

Key Finding

The aggregate brand convergence finding is real, but it is not evenly distributed across categories. When the same dataset is split by industry vertical, the spread of pairwise top-30 brand overlap looks very different from category to category. Retail shows 97% average pairwise brand agreement across all five engines. Travel shows 94%. Tech shows 88%. Finance drops to 71%. Healthcare drops to 60%, with a low-to-high range of 40% to 77% across engine pairs. The categories where consumers transact directly with brands (retail, travel, tech) draw from a smaller universe of well-established brands, and the engines pick from that same shared pool. The categories where consumers conduct deep informational research (healthcare, finance) have a much wider universe of credible brands, and each engine has developed its own preferences for which brands belong in the answer. The implication for brand strategy is that one playbook works across all five engines in transactional categories, while research-driven categories require engine-by-engine measurement and category-specific positioning.

Brand Agreement Sits at Different Points by Category

The pairwise top-30 brand overlap across all five engines varies materially by category. Where the buying decision is the dominant query intent, the engines converge. Where the research process is the dominant query intent, they diverge.

In retail, every engine pulls from a small, stable universe of major retailers and consumer brands. Amazon, Walmart, Target, Best Buy, Home Depot, and a handful of category leaders dominate every engine's top 30, leaving little room for divergence. Travel behaves the same way: a tight cluster of online travel agencies, hotel groups, and airlines (Expedia, Booking, Marriott, Hilton, Delta, United) carries across all five engines. Tech sits one step behind at 88% because the pool of relevant tech brands is slightly larger and the engines start to differentiate on which platforms, tools, and services they elevate. Finance and healthcare are the two categories where the divergence becomes the story.

Same Brand. Same Study. Different Treatment.

Inside the high-divergence categories, the same brand often receives dramatically different treatment from one engine to another. The gap is not subtle. Several flagship brands show share-of-mention differences of 9x to 22x across engines.

Mayo Clinic. Mayo Clinic is the most universally recognized healthcare brand in the dataset. Gemini directs 13.1% of its healthcare brand mentions to Mayo Clinic. Google AI Overviews directs only 1.5%. The same flagship authority commands roughly 9x more share on Gemini than on AI Overviews.

Cleveland Clinic. Cleveland Clinic captures 6.5% of Perplexity's healthcare brand mentions. On ChatGPT, that figure drops to 0.3%, a roughly 22x gap. Two engines querying the same healthcare prompts treat the same major medical institution very differently.

Goldman Sachs. Goldman Sachs ranks in the top 10 most-mentioned finance brands on Google AI Mode. On ChatGPT, Gemini, and Perplexity, Goldman Sachs does not crack the top 100. Institutional banking presence on AI Mode is materially different from institutional banking presence on the other engines.

Nasdaq. Nasdaq holds 4.5% share of finance brand mentions on ChatGPT. On AI Mode, that figure drops to 0.4%, an 11x gap. Engines that draw from exchange and regulatory sources elevate Nasdaq strongly. Engines that draw from institutional banking sources do not.

Bloomberg. Bloomberg captures 2.9% of finance brand mentions on Gemini. On AI Mode, Bloomberg has zero presence in the top 100. Editorial finance brands are heavily represented on Gemini and largely absent from AI Mode.

These are not measurement noise. They are systematic patterns that reflect the editorial signature each engine has developed inside high-divergence categories.

Each Engine Has a Personality Inside High-Divergence Categories

Inside the categories where the engines part ways, each engine favors a recognizable type of brand. The patterns hold up across the dataset and offer a usable mental model for predicting how each engine will treat brands in any high-divergence category.

ChatGPT favors specialty associations and exchanges. ChatGPT's healthcare top brands include the American Academy of Orthopaedic Surgeons (AAOS), the American Urological Association (AUA), the American College of Gastroenterology (GI), the American Academy of Pediatrics (AAP), and the American Academy of Family Physicians (AAFP). Its finance top brands lead with Nasdaq, the SEC, and major brokerages (Fidelity, Schwab, Vanguard). The pattern is consistent: ChatGPT elevates professional associations, regulatory bodies, and exchanges that carry institutional weight within their specific subcategory.

Perplexity leans on consumer health publishers and trading research tools. Perplexity is the only engine where consumer health publishers (Healthline, WebMD, Medical News Today) appear prominently in healthcare mentions. In finance, Perplexity uniquely elevates trading research tools like Stock Analysis, http://Investing.com , and Marketbeat alongside the standard exchanges. The pattern reflects Perplexity's broader posture as a research-oriented engine that surfaces both institutional and consumer-facing reference sources.

Gemini hyper-concentrates on flagship authorities and editorial finance. Gemini directs 13.1% of healthcare brand mentions to Mayo Clinic and 9.5% to NIH, two brands that account for nearly a quarter of all Gemini healthcare mentions. In finance, Gemini elevates Bloomberg, Wall Street Journal, Reuters, Forbes, and Investopedia, the editorial finance media set. The pattern is consistent: Gemini behaves like a concentrated authority recommender that leans heavily on a small set of flagship brands rather than producing broad lists.

Google AI Mode quietly favors institutional banks. AI Mode's finance top brands lead with JPMorgan, Wells Fargo, UBS, Goldman Sachs, Barclays, and Citigroup. The institutional banking presence is unique to AI Mode and is not mirrored on any other engine in the dataset. This pattern is not visible in aggregate share-of-voice reports, but it is clearly visible when finance brand mentions are isolated by engine.

Google AI Overviews spreads thin across providers. AI Overviews has the lowest brand concentration in high-divergence categories. No single brand commands more than 1.5% of healthcare mentions, and no finance brand exceeds 0.8%. The engine's top-15 lists in these categories include a wider variety of brands at lower share-of-voice levels, consistent with AIO's broader UGC-first sourcing posture.

What Marketers Need to Know

Brand agreement is real, but it is category-dependent. The aggregate finding that AI engines mostly agree on brands holds up. But the agreement is not evenly distributed. Retail and travel converge at 94% to 97% pairwise overlap. Healthcare and finance drop to 60% and 71%. Where your category sits on that spectrum determines whether one strategy works across all five engines or whether you need to look engine by engine.

Where consumers buy, engines converge. Where consumers research, engines diverge. Transactional categories pull from a small pool of well-established brands, and the engines pick from the same pool. Research-heavy categories have a much wider universe of credible brands, and each engine has developed its own preferences inside that universe. The strategic implication is straightforward: a single AI search strategy is portable across engines if you operate in retail, travel, or tech. If you operate in healthcare, finance, or other research-driven categories, you need to think harder about which engine is treating your brand how.

Each engine has a recognizable personality inside high-divergence categories. ChatGPT favors specialty associations and exchanges. Perplexity favors consumer health publishers and trading research tools. Gemini favors flagship authorities and editorial finance media. AI Mode favors institutional banks. AI Overviews spreads thin across providers. Knowing the editorial signature of the engines that matter to your buyers is half the work of building visibility in a high-divergence category.

Audit by engine within your category, not just in aggregate. The aggregate share-of-voice number can hide a lot if you are operating in a high-divergence category. A flagship competitor "missing" from one engine but dominant on another is not a measurement error. It is that engine's editorial signature in your space, and it is actionable. Brand teams operating in healthcare, finance, or any research-driven category should be measuring share of voice engine by engine and treating the divergence as a strategic input rather than a reporting inconvenience.

Match your authority strategy to the engines that matter. If your buyers rely heavily on Gemini, your strategy should prioritize being covered by the flagship authorities and editorial media that Gemini elevates. If your buyers use ChatGPT, your priority is presence in the specialty associations, regulatory bodies, and exchanges relevant to your subcategory. If your buyers use AI Mode in finance, institutional banking visibility is the lever that moves the needle. The three-layer source framework from the prior study still applies. The category-level engine personalities tell you where to put the emphasis.

A flagship competitor missing from one engine is signal, not noise. When a major brand in your category dominates one engine and is absent from another, that is meaningful information about how each engine has built its editorial signature. Mayo Clinic at 13.1% on Gemini and 1.5% on AIO is not random. Goldman Sachs in AI Mode's top 10 and absent from ChatGPT, Gemini, and Perplexity is not random. These patterns reflect each engine's source layer weighting and editorial posture. They are also the most actionable input for prioritizing PR, content, and authority-building investment.

Technical Methodology

Key Takeaways

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