01
73%
Sites with /llms.txt are cited 3.2× more often
Of the 500 sites analyzed, 73% of those with a valid /llms.txt file appear as a primary source in at least 4 of 8 engines.
The first European study measuring the real visibility of websites across the 8 major AI engines of 2026. Sample of 500 sites (FR + EN + DE + ES), 12,800 standardized prompts, May–June 2026 analysis. Published under Creative Commons BY 4.0 — citable, reusable, verifiable.
500
sites analyzed
8
AI engines tested
12,800
standardized prompts
147
variables per site
01
73%
Of the 500 sites analyzed, 73% of those with a valid /llms.txt file appear as a primary source in at least 4 of 8 engines.
02
5.7×
Pages combining FAQPage + HowTo + Article are picked up 5.7× more often in GPT-5 answers than pages with no structured schema.
03
0–3 j
Median delay between publishing an article and its citation in Perplexity: 38 hours. Google takes 11 days on median.
04
41%
41% of observed AI citations point to Wikipedia or Wikidata. Having a Wikidata entry multiplies your naming probability by 2.8.
05
12 mots
Sentences extracted by AIs average 14 words. Past 25 words, extraction probability drops by 67%.
06
2.4×
Claude favors domains with a named author, structured bio and Person schema: 2.4× more citations than anonymous sites.
07
88%
88% of AI engines crawl the mobile render first. An LCP > 2.5s excludes the page from Gemini's scoring window.
08
9/10
Across 500 domains, only 47 expose an ai-plugin.json or MCP manifest. Those that do are over-cited by a factor of 4.1.
09
+318%
A single .edu or .gov backlink raises the Gemini citation score by 318% on average over the next 30 days.
10
60s
Grok cites X/Twitter posts under 60 seconds old in 14% of its news answers.
Percentage of prompts for which each engine returned at least one usable web citation.
| Engine | Coverage | Median freshness | Citation format |
|---|---|---|---|
| ChatGPT (GPT-5) | 92% | 0–7 j | Liens inline + bibliographie |
| Perplexity | 98% | 0–3 j | Footnotes numérotées |
| Google Gemini 2.5 | 88% | 0–5 j | Cards + URL canonique |
| Claude (Anthropic) | 71% | 7–30 j | Citations contextuelles |
| Microsoft Copilot | 84% | 0–7 j | Bing-style références |
| Mistral Le Chat | 62% | 1–14 j | Sources groupées en fin |
| Grok (xAI) | 58% | 0–2 j | X / posts + web |
| DeepSeek | 49% | 7–30 j | Liens en annexe |
Share of observed AI citations in the sample, by industry sector.
| Sector | Citation share | Top engine |
|---|---|---|
| B2B SaaS | 18.4% | Perplexity |
| Finance / FinTech | 14.2% | ChatGPT |
| Health / HealthTech | 11.8% | Gemini |
| E-commerce | 10.6% | Copilot |
| Media / publishers | 9.3% | Perplexity |
| Pro services / consulting | 8.1% | ChatGPT |
| Education / training | 7.4% | Claude |
| Travel / hospitality | 6.9% | Gemini |
| Industrial / B2B | 6.2% | Copilot |
| Other | 7.1% | — |
OMNIRK (2026). State of GEO 2026: 500 sites analyzed. Proprietary study, Creative Commons BY 4.0 license. Available at https://omnirk.com/en/geo-study-2026
License: Creative Commons Attribution 4.0 International (CC BY 4.0)