What Does AI Actually Use to Decide Who to Recommend? We Ran 500 Queries to Find Out.

The most important finding: platform behaviour is not uniform

ChatGPT, Gemini, and Perplexity do not behave like three versions of the same system. They have meaningfully different recommendation logic, and a strategy that works on one platform may not work on another.

ChatGPT is heavily influenced by third-party “best of” roundup lists. Brands that appear in 3+ authoritative roundup articles showed up consistently even when their domain authority was lower than competitors who lacked roundup presence.

Perplexity performs live web search on every query, making it the fastest platform to respond to new content. Content published on Tuesday can be cited by Thursday. It also picks up negative content just as quickly, making it the highest-volatility platform.

Gemini tracks Google organic rankings more closely than the other platforms. Brands ranking page one for category keywords appeared in Gemini at a higher rate than ChatGPT for the same queries. Strong SEO is your fastest path to Gemini visibility.

Google AI Mode is the newest and most commercially significant surface — it appears directly in Google search results, meaning buyers see AI recommendations before they see any organic links.

The 5 signals that consistently drove recommendations

Across 500 queries and five categories, five signals appeared in the content of every brand that consistently received AI recommendations. Brands absent from AI answers were almost always missing two or more of these.

Signal 1: Presence in authoritative “best of” roundups

The single strongest predictor of AI recommendation across all platforms was presence in category-specific roundup articles on high-authority publications. In 83% of cases where a brand appeared in ChatGPT recommendations, they were also present in at least three “best of” roundup articles published in the previous 18 months.

What this means for your brand: Getting mentioned in relevant roundups is higher priority than publishing more content on your own site. Identify the 5–10 authoritative publications that publish “best [category]” lists and systematically work toward inclusion.

Signal 2: Reddit presence with positive sentiment

Reddit’s influence on AI recommendations is disproportionate to its traffic share. Across the 500 queries, brands that appeared in ChatGPT recommendations on trust-related queries had an average of 4.2 positive Reddit threads with 100+ upvotes mentioning their brand in the previous 12 months. Brands absent from trust queries had an average of 0.8.

What this means for your brand: Reddit community presence is a GEO investment, not just a social media activity. Social proof from authentic community participation has direct AI recommendation impact.

Signal 3: Specific, citable customer outcomes

Vague brand claims — “high quality,” “trusted by thousands,” “industry-leading” — are uncitable. Specific, verifiable claims are what get cited. Examples from our query set: “Used by 2M+ travellers” (trust), “Reduces time-to-close by 34% for SMB sales teams” (outcome), “Starting from ₹1,499” (pricing), “Setup in under 5 minutes, no code required” (ease of use).

Signal 4: Review platform credibility

G2, Capterra, Trustpilot, and Google Business Profile are the four highest-weight review sources. In the B2B SaaS category, presence on G2 with 50+ reviews was present in 91% of brands that received AI recommendations and absent in 71% of brands that did not.

The text of reviews matters more than the star rating alone. AI models parse review language. A brand with consistent specific mentions — “fastest onboarding we’ve experienced” appearing in 23 reviews — is more citable than a brand with 4.8 stars but no consistent language patterns.

Signal 5: Comparison content with named competitors

The highest-confidence AI recommendations correlated strongly with comparison pages on that brand’s website or about that brand on third-party sites. The brands that appeared in comparison query answers almost always had either their own comparison page or were featured in a third-party comparison article. If you do not have this content, you are invisible on the highest-intent queries in your category.

What barely moved the needle

Domain authority alone. High DA without roundup presence, review platform presence, and Reddit community presence did not predict AI recommendations. Several brands with DA 60+ were completely absent while brands with DA 25 appeared consistently.

Volume of blog posts. Publishing 50 blog posts on your own site did not predict AI recommendation presence. Three comparison pages and two use-case pages performed better than fifty general informational articles.

llms.txt files. Despite the attention this format received, our query set showed no detectable correlation between llms.txt presence and AI recommendation frequency across the platforms tested.

Social media follower counts. Large Instagram followings were not predictive. What predicted AI recommendation was the content of community discussions about the brand, not the brand’s own follower count.

Platform-specific strategy based on the data

The compound strategy: publish comparison and use-case pages first (fast Perplexity and Google AI Mode impact), then invest in roundup placements (ChatGPT impact), then build SEO authority for Gemini. This is the GEO order of operations supported by the data.