A quarterly study built from 500+ anonymised enterprise domain audits. Benchmarks, structural shifts, cross-model data. Free of charge, but available in limited supply — 600 slots per edition.
No newsletter. Only the report. Delivered on publication date.
Last reviewed:
Our own studies show that visibility in AI answers follows its own rules and can be measured. State of AI Search 2026 analyses 12,000 prompts across six AI systems, eight industries and 150+ brands: the correlation between Google position and LLM citation rate is 0.42, and 41 percent of brands in Google's top three are underrepresented in LLMs. The strongest predictor is entity maturity (r = 0.67), not the backlink profile (0.24 to 0.31).
The LLM Citation Benchmark defines the measurement framework: seven primary and five secondary metrics, 500 to 2,000 prompts per week as a robust base and a 95 percent confidence interval as the publication threshold. The entity resolution pilot from August 2026 checked 54 answers about 18 people in the DACH SEO scene: only 18.5 percent were error-free, and 46.3 percent contained false claims. With SEO context in the prompt, misattributions dropped to zero.
18–24 pages of dense analysis — no filler, no marketing pitch. Every number from our own audit cohort, documented methodology, reproducible KPIs.
Absorption rate by industry and query type. Quarterly delta, 12-month trend.
The most-cited domains in GPT, Claude, Gemini and Perplexity — by industry.
Share of Model across models, markets and languages. DACH, UK/US, TR, ES, AR.
RDI median values by industry. Negative peaks, recovery periods, patterns.
GPTBot, ClaudeBot, PerplexityBot — frequency, status codes, 429 dead zones.
Which schema types correlate with which citation rates — adjusted for domain authority.
Wikidata coverage of top brands by industry. Gaps, opportunities, timelines.
Three prioritised actions per industry, based on the quarter's data.
Every prompt is executed against GPT-4o, Claude Sonnet 4.6, Gemini Pro and Perplexity Sonar — five times per model to control for statistical variance. Brand mentions are extracted with a spaCy-based NER pipeline; sentiment with a verified classifier. All raw data is stored in BigQuery, analysis in Python notebooks with reproducible versioning. Every published number has a documented source.
The AI Search Index is published quarterly. Every edition remains permanently accessible to existing readers.
A one-year retrospective since rollout. Which industries lost the most traffic, which gained through citations — and why.
Wikidata coverage of the DAX 40 and its correlation with LLM citation. The brands with anchored entities get cited more.
MVG values across 15 language regions. Which markets are most underrepresented and why the corpus asymmetry is structural.