Why We Ran This 500-Query Study

Generative Engine Optimization — GEO — is no longer a niche concern for early adopters. It has become a core part of digital visibility strategy for anyone whose business depends on content traffic. And yet, most guidance on the topic is either too vague to act on or too focused on other AI systems like ChatGPT to be directly applicable to Claude.

Here’s what surprised us when we started researching: almost no published studies focused specifically on how Claude makes source selection decisions. There was plenty of content about AI citations generally, a lot about Google AI Overviews, and some useful material on Perplexity. But Claude? Largely untouched.

That gap matters. Claude is used by millions of people for research, purchasing decisions, health questions, financial queries, and more. If you are a publisher, marketer, or content strategist and you are not thinking about Claude citation behavior, you are leaving a significant visibility opportunity on the table.

We designed this study to give practitioners something concrete. Not broad principles. Not vague guidance. Actual data — directional but actionable — on what types of content, domains, and formats earn citations from Claude across real-world query categories.

Research Methodology

Our methodology was designed to replicate real-world user behavior as closely as possible while maintaining enough consistency to draw comparisons across categories.

Query Categories

We tested 500 queries distributed evenly across 10 topic categories — 50 queries per category. Categories were selected based on high commercial intent and existing SEO competition, so results would be relevant to publishers making real optimization decisions.

• Finance — investment, banking, credit, debt
• Insurance — auto, health, life, homeowners
• Healthcare — symptoms, treatments, medications, mental health
• Technology — software, hardware, SaaS tools, cybersecurity
• Travel — destinations, booking, visas, packing
• Education — degree programs, online courses, student loans
• Home Improvement — renovation, contractors, DIY, tools
• Ecommerce — product comparisons, buying guides, reviews
• Legal — contracts, rights, small business law, family law
• Local Services — plumbers, HVAC, pest control, cleaning

What We Tracked

For every query, we recorded the cited domains, classified each domain by type, noted the primary content format of the cited page, and assessed visible authority signals including author bios, citations and references, publication dates, and structured formatting.

We also tracked whether the cited content included original data, third-party references, expert authorship, or proprietary research — all signals we suspected might influence citation selection.

Limitations

The Biggest Findings

Here’s what the data told us — and where it surprised us.

Which Domain Types Claude Cites Most Often

Below is the full breakdown of citation share by domain type across all 500 queries. Note that these percentages reflect share of all citations recorded — not unique domains — so a single high-authority government page cited across multiple queries could represent several citation instances.

The numbers tell a clear story: trust architecture matters. Claude appears to use domain type as a proxy for credibility, particularly for factual claims where a wrong answer carries real-world consequences.

The interesting exception is the commercial category. At 14%, commercial .com domains punched above their weight compared to what many publishers expected. The difference between commercial sites that got cited and those that did not came down to editorial quality, author credentials, and the presence of original data.

Content Formats That Earn Citations

Domain type creates the ceiling for citation potential. Content format determines whether you reach it. Here is what we found by format.

Original Research

Nothing beat original research. If your site publishes a proprietary survey, a dataset from real customers, or a benchmark study from your own platform, you are creating something that no other page can replicate. Claude cannot source that claim anywhere else. That uniqueness is citation gold.

Think about it this way: if you run a budgeting app and publish an annual survey of spending habits across demographics, that data becomes a primary source. Government sites do not have it. Wikipedia does not have it. You do.

Statistics Pages

Dedicated statistics pages — the kind that compile data from multiple sources into one well-organized reference — performed exceptionally well. They match the exact pattern of how AI systems like Claude anchor factual claims: they need a number, they need a source, and they need to find it quickly.

If you are in any data-rich vertical — finance, health, SaaS, insurance — a regularly updated statistics page is one of the highest-ROI GEO investments you can make.

Long-Form Guides

Comprehensive guides of 2,500 words or more consistently earned citations when they demonstrated genuine depth, not just word count. Claude appears to reward coverage breadth combined with structural clarity. Guides that answered multiple related questions under well-organized headings outperformed those that answered one question at length.

Comparison Content

Comparison pages — best-of lists, head-to-head breakdowns, tool comparisons — were cited regularly for decision-oriented queries. When someone asks Claude to compare credit monitoring services or explain the difference between term and whole life insurance, comparison pages that provided structured, balanced analysis were consistently surfaced.

FAQs

FAQ sections within longer content performed strongly. Their structure maps almost perfectly to how AI systems extract direct answers: a clear question, a direct answer, and enough context to make that answer complete. If your content already has FAQs, make sure they are schema-ready and placed within clearly structured sections.

Product Reviews

Product review content can earn citations — but the bar is high. Reviews that cited manufacturer specifications, included real testing methodology, disclosed affiliate relationships transparently, and avoided promotional language consistently outperformed those that read like sponsored content. The closer a review looked to editorial journalism, the better it performed.

Authority Signals Claude Appears to Value

Authority is not just domain rating or backlink count. From our analysis, Claude evaluates authority through several visible signals — signals you can actually optimize for.

• Author Credentials Pages that displayed clear author bios with relevant professional credentials outperformed anonymous content consistently. A page about tax deductions written by a CPA who is named and credentialed performs better than the same page published under a faceless brand account.
• Expert Review Content that displayed “reviewed by” labels — particularly in healthcare, legal, and finance — showed higher citation rates. This is the GEO equivalent of E-E-A-T signals for traditional search.
• Citations and References The irony is real: pages that cite their own sources earn more citations from AI systems. Including links to government reports, peer-reviewed studies, or credible third-party data signals that your content is part of the broader knowledge ecosystem, not isolated opinion.
• First-Party Data As noted in the format analysis, original data is one of the strongest authority signals available. It signals that your organization has done primary research — a quality most content producers skip.
• Transparency About pages, editorial policies, disclosure statements, and business information all contribute to the trust architecture that AI systems use to evaluate credibility. Sites with clear ownership and transparent business models were cited more reliably.
• Freshness For time-sensitive verticals, keeping content updated matters. A healthcare page with a 2019 publication date and no update notice is a weaker citation candidate than one updated in the current year, even if the underlying content is similar.
• Structured Content Clear H2/H3 hierarchy, data tables, bullet points, and FAQ-style formatting all helped Claude locate and extract relevant information. Structure is not just user experience — it is AI readability.

Industry-by-Industry Results

Citation behavior is not uniform across industries. Here is what we found broken down by vertical.

The finance and healthcare categories showed the highest concentration of government citations. Legal followed a similar pattern, with state and federal government pages dominating factual query responses. Technology and SaaS showed the most opportunity for commercial publishers, with original research being the primary differentiator.

What Affiliate Marketers Can Learn

Affiliate content has a reputation problem with AI systems — and some of that reputation is earned. Pages that are thin, promotional, keyword-stuffed, and lacking in genuine expertise are exactly what AI citation systems seem designed to deprioritize.

But here’s the important part: affiliate content that mirrors editorial quality can and does earn citations.

What Works

• Product Reviews with Real Methodology Explain how you tested or evaluated products. Include specifications, comparison data, and clear evaluation criteria. Affiliate reviews that read like Consumer Reports journalism significantly outperformed standard “top 5” listicles.
• Comparison Pages with Neutral Framing Insurance comparison tools, credit monitoring service breakdowns, and budgeting software head-to-heads all performed well when they presented information neutrally and cited pricing from official sources.
• Buyer Guides with Original Insights Guides that answered specific use-case questions — “best identity theft protection for families,” “best credit monitoring for small business owners” — outperformed generic “best credit monitoring” pages when they demonstrated genuine understanding of the audience’s situation.
• Statistics Pages as Lead-in Content Several of the highest-performing affiliate-adjacent pages in our study were statistics pages that established credibility before linking to product recommendations. Publishing a comprehensive data page first, then referencing your review content, creates an authority bridge that improves both pages.

What Hurts

Heavy disclosure language at the top of a page, product-first framing without editorial context, and the absence of author information were the three clearest predictors of poor citation performance among affiliate-style content. These are also the easiest things to fix.

Claude vs ChatGPT vs Gemini vs Perplexity: Citation Behavior Compared

Not all AI answer engines cite sources the same way. Understanding the differences helps you prioritize your GEO efforts across platforms.

The key takeaway: Perplexity offers the most citation opportunities for diverse content types, but Claude’s citation behavior carries particularly strong weight for complex research and advisory queries. If your content targets high-trust, high-stakes decisions — financial planning, healthcare, legal questions — Claude optimization should be a priority.

Step-by-Step GEO Optimization Framework

Here is the framework we would recommend for any publisher looking to increase AI citation visibility, based on the patterns from this study.

1. Publish Original Research Identify a question your audience asks that no one in your industry has answered with primary data. Survey your customers. Pull data from your platform. Commission a study. Publish it with full methodology and make it citable.
2. Show Your Expertise Add named authors with relevant credentials to every piece of content. Include professional titles, relevant certifications, and links to professional profiles. If content requires specialized knowledge, add a “reviewed by” credential from a relevant expert.
3. Add References Cite your sources. Link to government reports, academic studies, and authoritative third-party sources within your content. AI systems look for evidence that your content is part of a larger knowledge network.
4. Create Statistics Pages Build at least one authoritative statistics page per major content cluster. Keep it updated. Use clear headings for each statistic or data point. This becomes a reference page that AI systems return to repeatedly.
5. Improve Entity Coverage Make sure your brand, your authors, and your core topics are represented consistently across your website, your About page, LinkedIn, and any professional profiles. Entity clarity helps AI systems understand who you are and what you cover.
6. Update Regularly Add update dates to all time-sensitive content. Create a content refresh calendar. For evergreen content, a note that it was reviewed in the current year signals maintenance that AI systems can detect.
7. Use Structured Headings Organize all long-form content with clear H2/H3 hierarchy. Use headings that mirror the actual questions your audience asks. This structure makes it easy for AI to extract the relevant portion of your page.
8. Answer Questions Directly The first sentence after every heading should directly answer what that heading asks. AI systems that extract answers for citation look for direct, declarative statements — not build-ups or preambles.

Real-Life Examples of the Framework in Action

Common Mistakes Publishers Make

If your content is not getting cited, one of these is probably why.

• Thin affiliate content with no methodology, no author, and no original insight — the most common citation killer we observed
• No author profiles or author information buried in bios with no credentials listed
• Outdated statistics with no update date visible — content from 2020 with no indication of review
• Missing citations within the content — pages that make claims without linking to supporting sources
• Weak or absent entity signals — no clear About page, no named business entity, no consistent author presence
• Keyword stuffing that destroys readability — AI systems extract and evaluate natural language, not keyword density
• Ignoring FAQ structure — pages that contain answers but do not format them for easy extraction
• Publishing content that is promotional in tone even when the topic is informational — this is the biggest trust signal problem for affiliate publishers

Future Trends in GEO and AI Source Selection

The landscape is moving fast. Here is where we see things heading in 2026 and beyond.

Frequently Asked Questions

Final Thoughts

The question “how does Claude choose its sources?” has a deceptively simple answer: it chooses sources that demonstrate trust, authority, originality, and structure — and it does so in patterns that are consistent enough to optimize for.

This study confirms that GEO is not a guessing game. Publishers who invest in original research, transparent expertise, well-structured content, and credible references are building exactly the kind of content that AI systems are trained to surface. The principles are not new. The application to a new visibility channel is.

If you take one action after reading this: start with your statistics page. Create one authoritative, well-structured, regularly updated data page in your core topic area. That single piece of content — done right — has the potential to become your most-cited asset in AI systems for years to come.

The publishers who understand AI citation behavior now are building competitive advantages that will compound as AI-driven search continues to grow. Start optimizing for the citation, not just the click.