Research

How AI platforms decide who to recommend

Built on 180+ primary research sources, forensic analyses across multiple industries, and the discovery of universal patterns in how AI recommendation engines work.

Foundational

How AI Recommendations Actually Work

The foundational explainer. Plain English breakdown of the Two-Brain Architecture, source chains, and the 14-signal hierarchy that determines who AI platforms recommend.

Read the research

Key findings from real intelligence reports

Patterns we have found across forensic analyses of AI recommendations in competitive categories. All findings are from real reports with details anonymized.

Being cited is not being recommended

One company was cited 103 times by AI platforms but ranked only #3 in recommendations. The sources that actually drive AI recommendations — listicles, review sites, authoritative directories — were where this company was absent.

From a real intelligence report

The Technical Optimization Paradox

The #1 recommended company in the category blocks 18 AI crawlers and scores lowest on technical signals. It wins because third-party sources — press, reviews, listicles — are what AI platforms actually consult. Technical optimization accounts for about 5% of the formula.

From a real intelligence report

Platform divergence is the norm

A company's visibility dropped from 54.5% to 0% on ChatGPT alone — the platform with 60% of AI users — because ChatGPT uses Bing for search, and the company's SEO was Google-focused.

From a real intelligence report

74% recommendation churn from single-word query changes

Adding one word to a buyer query ('best CRM' → 'best CRM for remote teams') changes AI recommendations 74% of the time. Static monitoring scores hide massive instability beneath the surface.

From a real intelligence report

Category positions harden at 7/10

AI recommendation positions harden over time. At 5/10 on the maturity scale, positions are fluid and displacement is cheap. At 7+, it costs 10x more. Early movers get 10x leverage.

From a real intelligence report

Our methodology: 11 phases, 6 platforms, 2,000+ citations

Every intelligence report follows the same rigorous process: discovery, site analysis, AI querying across 6 platforms, response analysis, competitive positioning, accuracy auditing, and synthesis. We trace every citation back to its source and test recommendation stability across thousands of query variations.

More resources

Want these insights for your company?

Our intelligence reports apply this research to your specific category, competitors, and market position.