State of AI Search: Stats & Insights 2026
This report contains data from B2B companies' Chosenly dashboards they're using to perform GEO (Generative Engine Optimization) for themselves.
Methodology and Data Sets
This data is collected by the Chosenly platform. We've anonymized the data our clients see and are sharing it with the larger community to further the GEO experiments others can also run.
Key considerations about the data:
- 1.Chosenly is built for B2B companies, so 100% of this data is based on searches from our B2B clients.
- 2.We guide our clients to only track searches that recommend companies—we're specifically tracking prompts with purchase intent.
Monthly session counts from top LLM search platforms (ChatGPT, Claude, Perplexity). Based on UTM parameters.
Key Insight:
Traffic grew 150% in the last six months Slight dip in August September will show if its a trend or just a fluctuation
Percentage breakdown of total traffic by source
Traffic Evolution:
SEO traffic has been gradually declining, shifting the overall traffic balance. LLM traffic has increased from 1.5% in January to 3.7% today, showing steady upward momentum. If current linear trends continue, LLM traffic could reach 50% market share by late 2036.
Conversion performance by traffic source
Key Insight:
LLM traffic converts significantly better than SEO traffic, showing higher user intent. Despite this, LLM traffic converts lower than paid Google Ads traffic, which is highly targeted. Paid ads outperform because they focus on specific bottom-funnel users, while LLM traffic grows naturally. Overall, LLM traffic shows strong intent, even compared to SEO, but doesn’t match the precision of paid campaigns.
Shows citation distribution across viral platforms (Reddit, LinkedIn, YouTube, Twitter, Review Sites)
Platform Diversity:
Despite some variation across YouTube, LinkedIn, and review sites, less than 0.5% of all citations are influenced by them. This shows that for B2B searches, these platforms have minimal impact. Doubling down on social media to drive LLM search traffic is unlikely to yield significant results.
Each dot represents the total citations for websites with the same Domain Authority score
Domain Authority:
We observed references across all kinds of websites, not just high-authority ones and domain authority does not have a significant impact on being referenced. The spike we see is largely due to sites like Wikipedia, but overall, references are widely distributed.
Content Preferences:
Listicles outperform all other content types for LLM B2B purchase-intent searches. Wikipedia and landing pages (including homepages) are frequently referenced, but mainly as general citations rather than for decision-making. Key insight: LLMs rely on listicles to make purchase-intent decisions, while Wikipedia and landing pages serve more as supporting references.
Criteria: The factors LLMs explicitly mention when recommending vendors (e.g., integrations, compliance, pricing, customer support).
Decision Framework: How LLMs Evaluate Vendors
LLMs often reference multiple decision criteria, usually at least five, when picking which companies to recommend. Chosenly is the only tool that actually shows these criteria for your industry, so you can see where you stand.
Variability: The structural differences in LLM responses for the same prompt across multiple generations, measured on a scale of 0-100.
Response variability: Searching the same thing multiple times shows different companies almost 50% of the time. This means you can’t rely on tools that run a prompt just once since the results aren’t consistent. Chosenly runs multiple prompt variations, which costs more but gives you reliable, actionable data you can actually trust.
Adoption: The percentage of cited websites that have implemented llms.txt files to guide LLM behavior and content usage.
LLMs.txt Implementation:
9.4% of referenced websites use LM.txt. But based on our data, it doesn’t really impact visibility, so there’s no need to rush adding it to your site.
How different content types perform over time
Evolution Insights
Biggest Gainer: Listicles (+1.9%)
Biggest Decline: Blog Posts (-5.8%)
Analysis Period: 2 months
Most Consistent: Blog Posts
Evolution of content on llms over the time:
From July onwards, listicles have become even more important for LLM B2B purchase-intent searches.
Dive deeper into specific months for detailed citation analysis