AI search statistics 2026.

Usage: the shift to answer search

• ChatGPT processes over 2.5 billion prompts per day with more than 800 million weekly users — a growing share with web search enabled. (Source: OpenAI figures, as of 2025)
• Google AI Overviews absorb 28–41% of clicks in informational query clusters — measured via the AI Overview Absorption Rate across a cohort of 500+ enterprise domains. (Source: own audit cohort, SUMAX Research 2026)
• AAR benchmarks by intent: informational 0.28–0.41 · navigational (brand) below 0.10 · transactional 0.05–0.18. Click loss is intent-dependent, not evenly distributed. (Source: own audit cohort, SUMAX Research 2026)
• Model usage weights, EU market (working model 2026): ChatGPT ~52% · Google Gemini ~22% · Claude ~11% · Perplexity ~10% · Bing Copilot ~5%. Basis of share-of-model weighting; calibrated quarterly. (Source: own calibration, SUMAX Research 2026)

Citation benchmarks: how often sources actually get cited

• Perplexity: 2–8% citation rate for domain authorities on subject-matter queries — the most citation-friendly system. (Source: LLM Citation Benchmark, SUMAX Research 2026)
• ChatGPT with Search: 1–5% — fewer source slots per answer, stronger consolidation. (Source: ibid.)
• Google AI Overviews: 0.5–3% for top domains — smallest source selection, highest competition. (Source: ibid.)
• Measurement basis for robust values: 500–2,000 prompts per week across several models and runs; single prompts are worthless as measurement. (Source: ibid.)

Entity resolution: what AI models get wrong about people

All figures from the Entity Resolution Pilot DACH SEO 2026 (54 answers, 3 standardized prompts × 18 people, against ground truth verified via web research; open aggregate data):

• Only 18.5% of answers error-free (10 of 54) — in a sample drawn largely from the industry's most visible figures.
• 46.3% contained false claims (25 of 54) — from invented podcast names to attributions to the wrong companies.
• Largest error class “partially correct”: 31.5% — right person, plausible but invented details. The most insidious class, because it spreads unchecked.
• 38.9% of the people (7 of 18) were never correctly resolved in any answer.
• Domain context eliminates misattributions: context-free prompts produced 4 misidentifications/confabulations each; the prompt with SEO context produced zero.

Ranking ≠ LLM visibility: the decoupling

All figures from State of AI Search 2026 (12,000 prompts, six AI systems, 150+ brands, eight industries; open aggregate data):

• 41% of brands dominant in Google SERPs have below-par visibility in ChatGPT and Perplexity (the LLM gap).
• Correlation Google position ↔ LLM citation rate: 0.42 — brand visibility in LLMs is largely decoupled from SERP position.
• Industry baselines of LLM citation rate (median): Legal Professional Services 31% · Travel/Hospitality 28% · Automotive 26% · B2B software 24% · E-commerce retail 22% · Finance 19% · Healthcare 18% · Industrial manufacturing 14%. Full table as CSV.

Google surfaces: the documented state in 2026

• GSC generative AI report in all properties since August 11, 2026: for the first time, official impressions from AI Overviews, AI Mode and generative Discover features — without clicks, without queries. Announced June 3, 2026; for data from August 13, 2026 onward the report carries a known logging bug (falsely declining impressions). (Source: Google Search Central; analysis: deep dive)
• Preferred Sources acts into AI Overviews and AI Mode: users mark preferred sources (domains/subdomains only); official button embed since August 20, 2026. Effect strictly personalized, effect size unquantified by Google. (Source: Google documentation; analysis: deep dive)
• FAQ and HowTo rich results are discontinued (for most sites in 2023, remaining cases until May 2026). The markup remains usable as a semantic declaration; a citation effect in AI Overviews is not proven. (Source: Google documentation)

Transparency case: this domain's own measurement

Because statistics pages love aggregating other people's numbers while hiding their own — here is this domain's baseline from the GSC generative AI report (collected August 26, 2026, period May 24 to August 23, 2026, old site version before the relaunch):

• 9,449 AI impressions in three months, rising trend (August peaks ~250–300/day versus ~100 in June).
• 135 pages appeared in generative AI features; 9 of the top 10 pages were English-language glossary entries.
• Complete AI fan-out prompts appeared as queries in the data — visible evidence of the decomposition machinery.

This baseline will be continued with future measurements — including the effects of the August 26, 2026 relaunch. (Source: own Search Console property)

Methodology & how to cite

Inclusion criteria: every figure on this page is either an own, methodologically documented survey (SUMAX Research, raw or aggregate data linked) or an officially documented vendor figure with a date. Estimates and working models are labeled as such. Outdated values are replaced, not silently overwritten — substantial changes are documented in the page's update state.

How to cite (CC BY 4.0): “Ulusoy, M. (2026): AI Search Statistics. SUMAX Research. https://www.muratulusoy.de/en/ai-search-statistics.html” — for individual figures, please also name the primary source given.