Last updated 13 May 2026
How AI Recommendation Infrastructure works.
When a buyer asks ChatGPT "what is the best [your category]", AI does not give them ten links. It gives them one answer with two or three named businesses. Your business is either in that answer or it is not. AI Recommendation Infrastructure is the system that gets you into the answer, measured weekly, grounded in peer-reviewed research.
AI is the new front door of business discovery.
AI-referred web sessions grew 527% year-over-year in the first half of 2025 (Previsible, 2025 AI traffic report). Gartner predicts traditional search engine volume will drop 25% by the end of 2026, with AI chatbots absorbing that share (Gartner, 2024). According to McKinsey (2025), 44% of consumers now use AI as a primary information source for purchasing decisions. For B2B buyers, the figure is higher: Walker Sands reports 90% of B2B buyers integrate generative AI at some point in their buying journey.
This is not a future trend. It is the current reality for challenger brands competing against incumbents with larger marketing budgets and deeper brand recognition. When a buyer asks an AI system for a recommendation, the system does not show a list of ten blue links. It synthesises a single answer, drawing from sources it trusts most. Brands structured for AI extraction get recommended. Brands that are not, disappear from the conversation entirely.
The metric that matters is not search ranking. It is recommendation rate: how often your business appears when AI systems generate answers in your category, measured across ChatGPT, Claude, Gemini, and Perplexity.
We follow the research, not the hype.
The AI recommendation field is full of unverified claims and self-reported vendor data. AI See You builds its methodology on the published research that has survived peer review and independent scrutiny. The single strongest piece of evidence is the Princeton and IIT Delhi Generative Engine Optimisation study (Aggarwal et al., KDD 2024, arxiv.org/abs/2311.09735), which tested nine optimisation tactics across 10,000 search queries.
Lily Ray, VP SEO at Amsive, identified the foundational constraint in March 2026: 100% of sites that lost Google organic traffic also lost AI citations across ChatGPT, AI Mode, and Gemini. Strong organic foundations remain a prerequisite. The "novel" tactics promoted by many vendors are either rebranded SEO or unproven shortcuts.
OpenAI states explicitly in its own documentation: "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." No AI company has published its ranking algorithm. Honest practitioners work from inference backed by evidence, not from guarantees.
We apply the tactics with the strongest evidence base. We explicitly avoid the tactics flagged as risky or unproven. The difference matters because AI systems are getting better at detecting manipulation, not worse.
"100% of sites that lost Google organic traffic also lost AI citations across ChatGPT, AI Mode, and Gemini."
What the Princeton GEO study actually proves.
The Princeton/IIT Delhi GEO study is the most rigorously tested evidence in the field. Here is what it found, and how each finding maps to what we build into every AI Knowledge Centre.
Citing credible sources increases AI visibility
The single strongest tactic in the Princeton study. For lower-ranked sites (the category that includes most challenger brands), embedding credible source citations within content produced a 115.1% visibility increase in AI-generated responses (Aggarwal et al., KDD 2024). Every AI Knowledge Centre we set up includes verified citations to peer-reviewed studies, regulatory bodies, clinical data, and industry sources. Every factual claim maps to a traceable, timestamped origin.
Statistics and quantitative data boost citation
Embedding specific statistics and quantitative data delivered up to a 40% improvement in AI visibility (Aggarwal et al., KDD 2024). AI Knowledge Centres target one verifiable statistic every 150 to 200 words. Not filler. Real data that helps consumers make better purchasing decisions and gives AI systems the factual specificity they need to confidently cite a source.
Expert quotations move the score
Adding quotations from recognised authorities improved AI visibility by 22 to 37% across tested domains (Aggarwal et al., KDD 2024). The strongest effects appeared in categories where trust is a decision factor, which includes most product and service categories where buyers compare options. Knowledge Centres include named expert perspectives where they are contextually relevant.
What the research says does NOT work
The Princeton study tested keyword stuffing. It performed worse than baseline. The most commonly used "optimisation" tactic actively reduces AI visibility. Lily Ray (VP SEO, Amsive) has warned that scaled AI content production triggers Google's "Scaled Content Abuse" spam policy. Reddit and forum astroturfing is detectable and carries brand risk. AI See You uses none of these tactics.
Want every source? See The Research.
Six layers of AI Recommendation Infrastructure.
AI Recommendation Infrastructure is not a single optimisation. It is a system of six layers, each addressing a specific requirement that AI systems use when deciding which sources to trust and cite. Removing any layer weakens the others.
AI crawlers (GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot, Google-Extended) must be able to read your content before anything else matters. According to OpenAI's documentation, ChatGPT Search uses these crawlers to discover and index web content. 80% of top publishers currently block AI crawlers (OpenAI documentation), creating an immediate opportunity for brands that allow crawl access. Every Knowledge Centre is fully crawlable by all major AI systems, pre-rendered from React into static HTML (AI crawlers do not reliably execute JavaScript), and indexed by both Google Search Console and Bing Webmaster Tools. ChatGPT uses Bing for its search retrieval, making Bing indexing a prerequisite for ChatGPT citation.
How a Knowledge Centre is set up.
Step 1. Measure.
We run your business through the same AI queries your buyers are asking. You receive a baseline Recommendation Score and a gap analysis showing exactly where you are losing recommendations and why.
Step 2. Set up.
We set up your AI Knowledge Centre at knowledge.yourbusiness.com. Structured content, schema, proof layer, FAQ architecture. Everything AI needs to evaluate and recommend you with confidence.
Step 3. Track.
Weekly automated testing across ChatGPT, Claude, Gemini, and Perplexity. Your Recommendation Score updates every week. You know when it is moving and why.
We sit above your Shopify store.
For ecommerce and DTC brands on Shopify, the AI Knowledge Centre is the data layer above your store. Your store is the buying experience. The Knowledge Centre is what AI reads when it decides which brands to recommend. The buyer hears your name from AI, lands at your Shopify store, buys. The Knowledge Centre integrates directly with Shopify via Admin API, so product changes, pricing updates, and inventory data flow into the recommendation layer automatically.
See the Product Brands page for full Shopify integration detail →
One number. Updated weekly.
The Recommendation Score is a single number from 0 to 100 that measures how likely AI systems are to recommend your business when a buyer asks. It is calculated from four signal layers: Knowledge Centre Health, Directory Consistency, Proof Layer, and Technical Readiness. The score updates weekly based on automated testing across ChatGPT, Claude, Gemini, and Perplexity, run against the same query sets every time so the trend over weeks is genuine signal, not noise.
Each signal layer maps to specific work in the six-month phased plan. Each piece of work moves the layer. The layer movement moves the score.
Knowledge Centre Health
Measures whether your AI Knowledge Centre contains the structured content AI needs to evaluate and recommend you. Assessed across content coverage, content depth, schema implementation, and citation readiness.
70+ indicates strong structured content coverage with well-implemented schema.
Directory Consistency
Measures whether your business information is consistent across the directories and platforms AI references. Inconsistencies reduce recommendation confidence.
75+ indicates consistent NAP data and service descriptions across major directories.
Proof Layer
Measures the strength of your evidence layer including reviews, credentials, case studies, and certifications.
65+ indicates a strong proof layer with diverse evidence types.
Technical Readiness
Measures whether your digital infrastructure is structured for AI crawler access including schema markup, page architecture, and content accessibility.
70+ indicates well-structured technical infrastructure for AI access.
Why this methodology favours challenger brands.
The Princeton study's most important finding for challenger brands is structural: the GEO tactics that produce the strongest results deliver the largest improvements for lower-ranked sites, not for already-dominant ones (Aggarwal et al., KDD 2024).
This means a challenger brand with 200 pages of structured, citation-rich, comparison-driven content on its own domain can close the recommendation gap against an incumbent with 10,000 pages of unstructured marketing copy. AI systems do not count pages. They evaluate the quality, verifiability, and structure of what those pages contain. A smaller brand that implements the Princeton findings systematically can outperform a larger brand that relies on volume and brand recognition alone.
The gap between traditional search overlap and AI citation sources has widened significantly. Research from GEO firm Brandlight 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. The playing field is more open than it has been in a decade, and the brands that structure their content for AI extraction now will compound that advantage as AI adoption accelerates.
This is why AI See You focuses on challenger brands. The methodology works best where the gap is largest and the potential improvement is greatest.
The Aussie Man, 18% to 34% in three weeks.
Same questions, same four AI platforms, independently verified. The Recommendation Score for The Aussie Man, an Australian DTC men's skincare brand on Shopify, moved from 18% to 34% in three weeks after the AI Knowledge Centre went live. The same methodology applies to every brand on the platform. Recommendation Score timing varies by category, market, and existing domain authority. Most clients see measurable change within 4 to 8 weeks.
Frequently asked questions
Most clients see measurable changes in their Recommendation Score within 4 to 8 weeks of an AI Knowledge Centre going live. The timeline depends on existing domain authority, content depth, and how quickly AI crawlers index the new content. The reference brand, The Aussie Man, moved from 18% to 34% in three weeks. Results vary by category and market.
See how we apply this methodology to your business.
Book a 15-minute call. We will run your baseline Recommendation Score across ChatGPT, Claude, Gemini, and Perplexity, then walk you through the methodology applied to your specific category.