The research behind AI Recommendation Infrastructure.

Princeton / IIT Delhi Generative Engine Optimisation Study

The landmark study in the GEO field. Tested nine optimisation tactics across 10,000 search queries using a controlled experimental framework.

Key findings:

• Citing credible sources produced a 115.1% visibility increase for lower-ranked sites. The single strongest tactic tested.
• Adding specific statistics delivered up to a 40% visibility improvement.
• Adding expert quotations improved visibility by 22 to 37%.
• Keyword stuffing performed worse than baseline.
• Lower-ranked sites benefited more than already-dominant ones. This is the structural finding that makes GEO especially relevant for challenger brands.

How we apply it: every AI Knowledge Centre we set up implements citation density, statistical frequency, expert quotation, and answer-first structure as core requirements.

RAG architecture research (multiple papers)

RAG (Retrieval-Augmented Generation) is the fundamental architecture behind how ChatGPT, Perplexity, and Google AI Overviews retrieve and synthesise information. Understanding RAG mechanics is essential for building content AI systems can reliably extract.

RAG systems use cosine similarity in vector space to match content to queries. Content that closely mirrors how users phrase their questions to AI is more likely to be retrieved. AI systems break pages into individual passages for retrieval and re-ranking, which means each section of a page must stand alone as a self-contained answer. Typically, LLMs pull from 5 to 16 sources per answer. Different models favour different source ecosystems: Perplexity leans heavily on community platforms, while Gemini relies on them far less.

How we apply it: Knowledge Centre content is structured to semantically match the natural language patterns consumers use when asking AI systems for recommendations. Each section is modular and extractable.

OpenAI ChatGPT Search documentation

OpenAI's official documentation confirms that ChatGPT Search uses specific crawlers (OAI-SearchBot, GPTBot) to discover content, and rewrites user prompts into search queries sent to Bing. This makes Bing indexing a prerequisite for ChatGPT citation. OpenAI states: "Ranking in ChatGPT Search is based on a number of factors designed to help users find reliable, relevant information. There is no way to guarantee top placement." 80% of top publishers block AI crawlers, creating a significant opportunity for brands that allow crawl access.

How we apply it: every AI Knowledge Centre is indexed by Bing Webmaster Tools. All AI crawlers are explicitly allowed in robots.txt. No AI crawler is blocked.

Google Search Central structured data documentation

Google's documentation recommends putting Product markup in initial HTML rather than relying solely on JavaScript rendering. Schema markup (Product, Organisation, FAQ, HowTo, Review, Article) increases the probability of inclusion in AI Overviews.

How we apply it: every AI Knowledge Centre page includes JSON-LD structured data in pre-rendered HTML that AI crawlers receive directly, without requiring JavaScript execution.

Lily Ray, VP SEO, Amsive (March 2026)

Lily Ray identified the foundational constraint: 100% of sites that lost Google organic traffic also lost AI citations across ChatGPT, AI Mode, and Gemini. Strong organic SEO remains a prerequisite for AI recommendation.

Ray has also flagged the following tactics as dangerous: scaled AI content production (triggers Google's "Scaled Content Abuse" spam policy), Reddit and forum astroturfing (detectable and carries brand risk), and self-promotional listicles.

How we apply it: AI Knowledge Centres are built on genuine human expertise and verified research. We do not use AI-generated content at scale, forum astroturfing, or self-promotional listicles.

Ahrefs AI citation analysis

Ahrefs analysis found that Google AI Overviews pull 74% of citations from top-10 organic results. This confirms that organic search performance feeds directly into AI recommendation probability.

How we apply it: AI Knowledge Centres are designed to rank organically as well as serve AI systems, because organic ranking is a primary input to AI recommendation.

ZipTie.dev ChatGPT citation pattern analysis

ZipTie.dev analysed ChatGPT citation patterns and found:

• Sites with Trustpilot presence average 4.6 to 6.3 citations from ChatGPT.
• Domain trust scores of 97 to 100 average 8.4 ChatGPT citations versus 1.6 for scores below 43 (a 5.25x gap).
• Content updated within 30 days receives 3.2x more ChatGPT citations than stale content.

How we apply it: AI Knowledge Centres strengthen cross-platform brand signals and maintain content freshness with visible update timestamps on every page.

Growth Memo ChatGPT citation positioning analysis

Growth Memo analysis found that 44% of ChatGPT citations come from the first third of a page's content.

How we apply it: every AI Knowledge Centre page puts the direct answer in the first 40 to 60 words.

First Page Sage algorithm research

First Page Sage research found that being featured in independent "best [category]" articles is more valuable for AI recommendation than any single on-site optimisation.

How we apply it: AI Knowledge Centres include comparison and "best of" content that positions the brand within broader category discussions, rather than only self-promotional product pages.

Brandlight GEO research (2026)

Brandlight analysis suggests the overlap between top Google links and AI-cited sources has dropped from 70% to below 20%. Organic SEO dominance no longer guarantees AI recommendation, and the playing field for challenger brands is more open than it has been in a decade.

How we apply it: our methodology treats AI recommendation as a distinct channel from organic SEO, with its own requirements for structure, citation, and schema, while maintaining the organic SEO foundations that remain a prerequisite.