TLDR: A new research paper introduces the GEO-16 framework, a 16-pillar auditing system that quantifies page quality signals relevant to AI answer engine citation behavior in B2B SaaS. The study, which analyzed 1702 citations from Brave, Google AIO, and Perplexity, found that engines differ significantly in the quality of pages they cite, with Brave citing the highest quality. Key factors strongly associated with citation include Metadata & Freshness, Semantic HTML, and Structured Data. Pages with a GEO score of 0.70 or higher and at least 12 pillar hits achieve a 78% cross-engine citation rate. The findings provide actionable guidance for publishers to prioritize recency metadata, semantic structure, valid structured data, and earned media to improve AI discoverability and citation likelihood.
In today’s digital landscape, AI answer engines are rapidly becoming a primary way for businesses to distribute their knowledge, especially in the B2B SaaS sector. These engines, such as Brave Summary, Google AI Overviews, and Perplexity, synthesize information and provide direct answers, often citing external web content as their sources. However, until recently, there hasn’t been a clear understanding of what makes a particular web page more likely to be cited by these powerful AI systems.
A new research paper, “AI Answer Engine Citation Behavior: Bringing the GEO-16 Framework in B2B SaaS,” by Arlen Kumar and Leanid Palkhouski, addresses this critical gap. The researchers introduce a novel framework called GEO-16, which is a 16-pillar auditing system designed to quantify the quality signals of a web page that are most relevant to how AI engines select their citations. This framework helps publishers understand and optimize their content for better visibility in generative search.
The GEO-16 Framework and Methodology
The study involved a large-scale audit, using 70 industry-targeted prompts to collect 1702 citations from the three major AI answer engines: Brave, Google AIO, and Perplexity. A total of 1100 unique URLs were then audited using the GEO-16 framework. Each page was scored from 0 to 3 for each of the 16 pillars, and these scores were aggregated into a normalized GEO score, ranging from 0 to 1. The framework is grounded in six core principles that link human-readable quality to machine parsability and retrieval behavior:
- People-first content: Emphasizes clear, answer-first structures, compact paragraphs, and descriptive headings to enable easy extraction of snippets.
- Structured data: Focuses on proper HTML hierarchy (like single h1, logical h2/h3), valid JSON-LD (e.g., Article, TechArticle, FAQPage schemas with dates and author info), and canonical URLs.
- Provenance: Highlights the importance of citing primary sources, including reference sections, favoring authoritative domains, and maintaining healthy links.
- Freshness: Stresses visible timestamps, machine-readable dates, notes on substantive revisions, and up-to-date sitemaps.
- Risk management: Involves editorial review, fact-checking, disclosures, and scope limits to reduce potential inaccuracies.
- RAG optimisation: Encourages well-scoped pages, descriptive internal anchors, contextual anchor text, and avoiding duplicate URLs.
Key Findings and Engine Differences
The research revealed significant differences in the quality of pages cited by each engine. Brave Summary consistently cited higher-quality pages (mean GEO score of 0.727) with the highest citation rate (78%). Google AIO followed closely (mean GEO score of 0.687, 72% citation rate), while Perplexity cited pages of considerably lower quality (mean GEO score of 0.300, 45% citation rate).
The study also identified the specific GEO-16 pillars most strongly associated with citation likelihood. These include Metadata & Freshness, Semantic HTML, and Structured Data. This reinforces the idea that machine-readable structure and up-to-date information are crucial for AI discoverability.
A practical operating point emerged from the analysis: pages with a GEO score of 0.70 or higher and at least 12 pillar hits achieved an impressive 78% cross-engine citation rate. Furthermore, URLs cited by multiple engines exhibited 71% higher quality scores than those cited by only one engine, indicating that high-quality content tends to gain broader recognition.
Also Read:
- Navigating the New Era of AI Search: A Guide to Generative Engine Optimization
- Geoptie Unveils Comprehensive AI-Powered Generative Engine Optimization (GEO) Dashboard
Actionable Guidance for Publishers
For B2B SaaS publishers aiming to improve their visibility in generative search, the findings provide clear, actionable benchmarks. The authors recommend prioritizing:
- Recency Metadata: Ensure both human-visible timestamps and machine-readable dates (via JSON-LD) are consistently updated and accurate.
- Semantic Structure: Implement a clear and logical HTML hierarchy (h1, h2, h3 tags) and ensure schema completeness.
- Valid Structured Data: Provide accurate and relevant JSON-LD (e.g., Article, TechArticle, FAQPage) that matches the visible content.
- Diverse References: Include authoritative sources and maintain accessible page structures.
Beyond on-page optimization, the research also highlights the importance of securing coverage on authoritative third-party domains, often referred to as ‘earned media.’ Generative engines tend to favor these domains over brand-owned blogs or social content. Therefore, a dual strategy of achieving on-page excellence and cultivating earned media relationships is recommended.
This paper offers the first empirical link between structured page quality signals and AI answer engine citation outcomes, providing practical guidance for improving discoverability in the evolving landscape of generative search. You can read the full research paper here.
