How ChatGPT Cites Sources: A 2025 Data Study
TL;DR
Our September 2025 analysis reveals a shift in how AI engines determine authority. ChatGPT shows different citation patterns compared to traditional search. We analyzed how ChatGPT answers non-branded, top-of-funnel B2B questions and compared it to Google Search. The key finding is a change in what AI considers an authoritative source.
The data shows ChatGPT favors content directly from branded domains. It cites competitor websites 11.1 points more and a company's own website 3.0 points more than Google. It also shows increased preference for encyclopedic (+3.1 points) and academic (+1.4 points) sources. Forums maintain relevance but serve a different function. ChatGPT appears to use community discussion as an authenticity signal to support primary claims from corporate sites.
Introduction: Changes in Authority Signals
For two decades, SEO professionals built authority using backlinks, media mentions, and high-ranking articles in industry publications. AI models use different signals to establish trust. As generative models become common entry points for information, the signals they use to establish trust are changing. This represents more than an algorithm update. It represents a change in how authority is determined.
In our main guide on Generative Engine Optimization (GEO), we introduce new principles for building trust with AI. This post provides specific data showing this new model of authority. This is the fourth data study in our series, where we analyze how different AI engines source their information. Here, we focus on ChatGPT to understand its specific behavior patterns.
Our Experiment: Testing for Authoritative Sources
To understand ChatGPT's behavior, we designed an experiment to compare its citations against traditional Google Search.
- Data Collection Period: September 13-14, 2025.
- The Queries: We generated 80 high-intent, non-branded questions for each of our three test companies: Mindtickle (Horizontal SaaS), MotherDuck (Developer Tools), and Prodigal (Vertical FinTech). These questions simulated users in the interest stage of the buyer's journey, seeking information about a category rather than a specific brand.
- The Engines: We ran each query through Google's Search API and the ChatGPT web interface with its web search feature enabled.
- The Analysis: We extracted every source URL cited by both engines, normalized them to their root domain, and categorized each one to map the citation landscape. This allowed direct comparative analysis of sources each engine used to build its answers.
The Data: Branded Websites as Primary Sources
Our analysis revealed a pattern: ChatGPT builds its understanding by looking directly at market participants, treating their websites as primary sources.
Key Finding: ChatGPT Prioritizes Branded Content
| Content Category | Google Share (%) | ChatGPT Share (%) | Change (ChatGPT vs Google) |
|---|---|---|---|
| Self (your site) | 2.9% | 5.9% | ▲ +3.0 pts |
| Competitors | 17.7% | 28.8% | ▲ +11.1 pts |
| Ecosystem Partners | 35.5% | 30.7% | ▼ -4.8 pts |
| Discussion Forums | 18.4% | 10.5% | ▼ -7.9 pts |
| Industry Publications | 12.6% | 7.8% | ▼ -4.8 pts |
| User Generated | 9.3% | 6.5% | ▼ -2.8 pts |
| Encyclopedic | 0.1% | 3.1% | ▲ +3.0 pts |
| Academic | 2.1% | 3.5% | ▲ +1.4 pts |
| Other | 1.6% | 3.2% | ▲ +1.6 pts |
Note: This table reflects the overall share of citations across our entire dataset. This differs from citation frequency, as a source category can appear in a high percentage of answers while comprising a smaller overall share of total links.
Here are the key trends we identified.
AI Triangulates Authority Between Market Players
The most significant finding is that ChatGPT constructs answers by consulting main vendors in a space. The +11.1 point increase in competitor citations shows that the AI views the competitive landscape as the source of truth for a topic. It learns what a category is, who the key players are, and how products compare by reading what companies themselves publish.
Primary Sources and Factual Grounds Define Authority
The growth in "Self" (+3.0 pts) and "Encyclopedic" (+3.1 pts) citations points to a new authority model. For ChatGPT, authority is proven in two ways: by being a primary source (a vendor in the space) or by being a neutral, factual source (like an encyclopedia). The AI prefers getting information directly from the source or from content dedicated to objective fact, rather than from intermediary commentators.
Rebalancing of Authority Signals
The largest drops in citation share come from industry publications (-4.8 pts) and discussion forums (-7.9 pts). This might initially suggest these traditional signals are becoming less relevant. However, closer examination reveals a change in role rather than elimination.
While our data shows clear rebalancing of citation share toward branded domains, this doesn't mean abandoning community platforms. This data measures the proportion of citations, not the frequency of their appearance. Forums remain a frequent source for the AI, particularly for technical and problem-solving queries. The key insight is that their function within AI-generated answers is evolving.
Why would an AI reduce citation share from forums while still citing them frequently? This may be a strategy to balance authority with authenticity. The model appears to use branded domains or encyclopedic sources to establish core facts. Then it uses forum links to add lived experience through real-world examples, specific solutions, or human perspectives. This makes the answer more comprehensive and grounded.
Future research should explore the prominence of these citations. The decline in share may result from forums shifting from primary, top-ranked sources to supplementary citations within AI-generated answers, which would explain decreased share alongside maintained frequency.
How ChatGPT's Behavior is Evolving: A Comparative Analysis
This study is our second comparison of a large language model to Google Search. Adding these findings to our previous research on Gemini reveals that AI engines behave differently.
The Rise of Encyclopedic and Academic Sources
This trend is specific to ChatGPT. Our Gemini study showed minimal change in encyclopedic citations (+0.2 points). In contrast, this ChatGPT study reveals a +3.1 point increase for encyclopedic sources and a +1.4 point increase for academic sources.
Insight: ChatGPT relies heavily on neutral, factual sources to ground its answers for non-branded, top-of-funnel questions. Creating content that aligns with the factual, unbiased tone of these sources could be effective for GEO.
A More Moderate Decline in Forum Citations
In our Gemini study, forum citations dropped by -16.5 points. Our new ChatGPT data shows a drop of -7.9 points.
Insight: This suggests ChatGPT has a developed strategy for using community content. While both models reduce reliance on forums as primary sources, ChatGPT retains them as frequent, supplementary signals for authenticity. This behavior appears less prevalent in Gemini.
Industry Deep Dive: Authority in Different Markets
The aggregate data reveals patterns that become clearer when examining specific industries.
Prodigal (Vertical FinTech)
Prodigal demonstrates the new authority model clearly. The AI determines that vendors themselves are the primary authorities. It largely bypasses industry media, with competitor citations increasing from 13% on Google to 51% on ChatGPT. Industry publications, which accounted for 32% of Google's sources, dropped to 8% for ChatGPT. In specialized markets, ChatGPT trusts the companies building the technology to define it.
MotherDuck (Developer Tools)
The developer tool space shows a hybrid authority model. The AI recognizes authority in community discussion, with forums comprising 18% of citations (down from 27% but still significant). ChatGPT supplements this community knowledge with foundational, factual authority from encyclopedic sources, which increased from 0% to nearly 5%. For developer marketers, community engagement remains important but should be paired with clear, factual content that defines core technology concepts.
Mindtickle (Horizontal SaaS)
In the broader SaaS market, the AI gives authority to vendors over individual voices. It trusts content from competitor and self domains more than user-generated blogs. Citations from these blogs, which comprised over 10% of sources for Google, decreased to 2% for ChatGPT. Authority has shifted from individual thought leaders to official company voices.
Understanding the Shift: Why ChatGPT Favors Branded Authority
These citation patterns aren't random. They reflect deliberate choices in how ChatGPT evaluates information quality.
The Training Data Effect ChatGPT's training included vast amounts of web content where company websites and academic sources were likely tagged as high-quality. The model learned to associate certain content patterns (structured information, technical documentation, official statements) with reliability.
Solving the Content Farm Problem Traditional SEO created an ecosystem where intermediary sites could rank highly by aggregating and repackaging information. ChatGPT's preference for primary sources represents a structural solution to this problem. Rather than trusting the site that talks about the technology, it goes directly to the site that builds it.
The Verification Challenge AI models face unique challenges in verifying information. By defaulting to primary sources (companies) and neutral references (encyclopedias), ChatGPT reduces its exposure to opinion, speculation, and outdated information. Forums serve as a reality check rather than a primary source.
Legal and Ethical Considerations Citing official sources provides clearer attribution and reduces potential liability issues. When ChatGPT says "According to Microsoft," the attribution is clear and verifiable.
The GEO Playbook: How to Become an Authoritative Source for AI
Understanding these trends enables action. Here is a playbook for building the type of authority that AI models like ChatGPT recognize.
1. Publish Authoritative Comparisons
The data shows that if you don't publish authoritative comparisons, the AI will use your competitors' content as the source of truth about you. Creating detailed "Alternative To" and comparison pages is both a defensive necessity and an important strategy for establishing authority with AI.
2. Align with Factual and Encyclopedic Sources
ChatGPT's reliance on encyclopedic sources provides direction. Create foundational content that defines your category, your technology, and your core concepts. This content should mirror the factual, unbiased, and structured tone of encyclopedic sources that ChatGPT favors. Consider "What is X?" and "How does Y work?" pieces that serve as definitive references.
3. Maintain a Strategic Community Presence
The goal is to create high-signal content about lived experience that ChatGPT seeks to balance its answers. By providing clear solutions and real-world context on platforms like Reddit and StackOverflow, you provide the AI with authenticity signals it needs to trust and cite perspectives.
Conclusion: Continuous Experimentation is Essential
The age of AI brings changes to authority signals, and our September 2025 data shows ChatGPT has unique priorities. Currently, its strategy appears to be conferring primary authority on branded websites, using encyclopedic sources for fact-checking, and leveraging community forums to provide authenticity and real-world context.
These are current patterns, not permanent rules. Change is the only constant in this era. Success in generative engine optimization comes from building a culture of continuous, data-driven experimentation. Success no longer comes from mastering one algorithm but from adapting to multiple intelligent systems. B2B teams that succeed will treat their content strategy as a hypothesis they continuously test, measure, and improve.
Frequently Asked Questions (FAQ)
Does this mean backlinks are no longer important for SEO?
Not at all, but their role is evolving from a direct ranking factor to a foundational signal of trust. In traditional SEO, backlinks are a primary signal for ranking a specific webpage. For generative engines, which synthesize answers from multiple sources, a single backlink doesn't guarantee your content will be used in a generated paragraph.
However, a strong backlink profile contributes to your domain's overall authority. This authority makes your site a more trustworthy source for the AI's underlying search system to pull from in the first place. Think of it as a shift in focus. While link authority remains important for being discoverable, content authority—the clarity, structure, and factual nature of your information—is becoming the most direct signal for getting cited in an AI-generated answer.
How is this different from Google's AI Overviews?
This is a great question because while they seem similar, they are different systems. Google's AI Overviews are an extension of its core search engine, deeply integrated with its existing web index and traditional ranking signals. It's search, evolved.
ChatGPT, on the other hand, is a conversational AI that uses web search as a tool to fetch real-time information. It's an answer engine that uses search as an input. This architectural difference means they weigh sources differently. For example, our data shows that while both systems are shifting towards branded content, ChatGPT has a uniquely strong preference for encyclopedic and academic sources to fact-check its answers. They are both part of the broader trend toward generative answers, but they are distinct products with unique behaviors that need to be analyzed separately.
How can I track my brand's citations in ChatGPT?
Tracking citations manually is difficult and doesn't scale. Answers in ChatGPT can vary between sessions, and to get a clear picture, you need to ask thousands of questions that your customers are asking and analyze how you, your partners, and your competitors are being mentioned over time.
This is precisely the problem Zenith was built to solve. Our platform provides exhaustive tracking of your share-of-voice across all major AI platforms. We monitor citations, rankings, and sentiment changes in a unified dashboard, eliminating the complexity of manual tracking. This turns a noisy, unreliable process into a clear, actionable dataset that allows you to build, measure, and adapt your Generative Engine Optimization strategy effectively.