Updates and references

This is the living version of the references and figures from Get Recommended by AI. AI search changes monthly, and print does not. Anything that has moved since the book went to press is recorded here.

Current figures

Last updated 2026-07-27

The book quotes the State of AI Visibility Index as it stood when it went to print. These are the current values from the same benchmark.

Websites analysed	1,000,000+
Sites reaching AI-Ready (GEO 80 or above)	216
Share of the corpus	0.022%
Expressed as a ratio	1 in 4,600
Sites reaching Verified (GEO 90 or above)	0
Highest GEO score measured anywhere	86.6

The tier bands used throughout the book are unchanged: 80 to 100 AI-Ready, 60 to 79 Strong, 40 to 59 Emerging, 20 to 39 Low Visibility, 0 to 19 Invisible, with Verified awarded at 90 and above. Full method at searchscore.io/savi/methodology.

Corrections

References

Every external claim in the book, sourced by name. In print these links cannot be clicked, so this is the working version.

Academic research

• GEO's founding paper, and the 30–40% lifts from quotations, statistics, and cited sources (with keyword stuffing underperforming baseline): Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan & Deshpande, GEO: Generative Engine Optimization, KDD 2024 (arXiv:2311.09735, first posted November 2023).
• Answer volatility (same-day, same-prompt source overlap of roughly 0.32–0.43; minimum 7–8 runs per prompt and multi-day windows for stable measurement): Schulte, Bleeker & Kaufmann, Don't Measure Once: Measuring Visibility in AI Search, University of St. Gallen, April 2026 (arXiv:2604.07585).
• Prompt repeatability (two runs of the same brand-recommendation prompt returning the same brand list less than 1 in 100 of the time; identical ordering under 1 in 1,000): SparkToro, 2,961-run study, December 2025.
• Citation selection vs citation absorption, and ChatGPT citing fewer sources with higher per-citation influence: Zhang, He & Yao, April 2026 (arXiv:2604.25707).
• Structural optimisation improving generative visibility while body-text-only tweaks degrade it: Kim et al., SAGEO Arena, Yonsei University, February 2026 (arXiv:2602.12187).
• Earned media supplying 82–92% of GPT citations in tested consumer verticals (Google's surfaces more brand-tolerant): Chen, Wang, Chen & Koudas, University of Toronto, September 2025 (arXiv:2509.08919).

Citation-source studies

• ChatGPT referral traffic after the May 2026 links update (52 sites; sessions up 97.9% over six weeks, with the top five sites capturing 91% of the net increase and 14 sites declining): Botpresso, 2026.
• Most-cited domains (Reddit first, then YouTube, LinkedIn, Wikipedia): Peec AI analysis of 30 million citations across ChatGPT, Gemini, Perplexity, Google AI Mode, and AI Overviews, March 2026. Engine differences (ChatGPT historically Wikipedia-first): Profound's 680-million-citation dataset, 2024–2025; Semrush's 100-million-citation study, 2025, also places Reddit first with a different tail order.
• LinkedIn as the most-cited domain for professional/business queries: Profound, early 2026.
• Brand mentions vs backlinks (correlations of 0.664 vs 0.218 with AI visibility; YouTube mentions strongest of all): Ahrefs correlation studies, May and December 2025. Backlinks weakest and page count near-zero as predictors: same studies; Seer Interactive, January 2025.
• Third-party sources dominating evaluative citations (~4 in 5 in professional services): Semrush, 2025. Vendors' own "best of" listicles leading to competitor recommendations ~69% of the time: Ahrefs/Lily Ray analysis, June 2026.
• Zero-click behaviour (clicks falling from 15% to 8% of results pages when an AI summary appears): Pew Research Center, July 2025. AI referral growth (+357% year on year to 1.13B monthly visits): Similarweb, June 2025; AI referrals at roughly 0.1-0.5% of site traffic: Ahrefs and Similarweb analyses, 2025.
• Sources cited per AI answer (ChatGPT ~4, Gemini ~11, Perplexity ~9): SurfacedBy, 127,000-citation analysis, 2026.
• 44% of ChatGPT citations drawn from the top third of the cited page: Kevin Indig, Growth Memo, 1.2-million-citation analysis, 2026.
• Earned media supplying ~84% of AI-cited links (~25 million links analysed): Muck Rack, 2025.
• ChatGPT ads appearing in roughly half of US free-tier replies by mid-2026: industry rollout reporting, 2026.

Platform documentation

• OpenAI bot semantics (OAI-SearchBot governs ChatGPT search inclusion; GPTBot is training-only; ChatGPT-User handles live fetches): OpenAI crawler documentation, current 2026.
• Anthropic bot semantics (Claude-SearchBot for search indexing; ClaudeBot training; Claude-User fetches): Anthropic support documentation, current 2026.
• Google-Extended controlling Gemini training/grounding only, with no effect on Search, AI Overviews, or AI Mode (which are fed by Googlebot), and having no request user-agent: Google Search Central crawler documentation and "AI features and your website," current 2026.
• Query fan-out in AI Mode and AI Overviews: Google (Elizabeth Reid), May 2025, and Google Search Central documentation.
• Structured data helping Bing/Copilot's AI systems: Microsoft (Fabrice Canel), March 2025. Google stating no special structured data is needed for AI features: Google AI-optimisation guidance, 2026. A controlled Ahrefs experiment (2026) found no significant citation lift from adding schema alone.
• ChatGPT ads rollout: OpenAI announcements, from February 2026.
• ChatGPT shopping results and Instant Checkout (Etsy and Shopify merchants; the open Agentic Commerce Protocol, published with Stripe): OpenAI announcements, 2025.
• Google AI Mode product answers drawing on the Shopping Graph (tens of billions of listings): Google I/O and Shopping documentation, 2025–2026.
• Perplexity merchant programme and in-chat purchasing: Perplexity announcements, 2024–2025.
• AI-referred traffic to US retail sites up 14× October 2024 to May 2026; holiday AI traffic up ~700% year on year: Adobe Analytics, 2026.

llms.txt

• Adoption at roughly one in ten sites: SE Ranking analysis of ~300,000 domains, November 2025.
• 97% of existing llms.txt files receiving zero requests in a month (137,210 domains): Ahrefs, June 2026.
• Google Search does not support llms.txt: John Mueller and Gary Illyes statements, 2025, and Search Central guidance, 2026.
• llms.txt check in Chrome Lighthouse 13.3 ("Agentic Browsing" category, experimental; missing file marked not-applicable): Chrome for Developers documentation, May 2026.

Schema and FAQ rich results

• Google deprecated FAQ rich results fully on 7 May 2026 (Search Central changelog, 8 May 2026; documentation removed June 2026), having restricted them to government and health sites in August 2023. FAQPage remains a valid Schema.org type.

AI crawler access

• Share of sites blocking AI crawlers: close to four in ten in the SearchScore SAVI dataset, versus ~6% of the broad web blocking GPTBot in Ahrefs' ~140-million-site analysis; a fifth to a third of top-1,000 sites and roughly half of major news publishers block at least one AI crawler (Originality.ai and publisher robots.txt trackers, 2025–2026). Cloudflare began blocking AI crawlers by default for new zones in July 2025.
• Microsoft Copilot is grounded in the Bing index; Bing Webmaster Tools added an "AI Performance" report in February 2026. ChatGPT search launched on Bing (October 2024) but OpenAI has since built its own index and reportedly supplements it with other providers, 2025 reporting.

SearchScore SAVI (State of AI Visibility Index)

• State of AI Visibility Index, Q3 2026 edition (published August 2026), the source of the whole-web figures in Chapters 1-4: average GEO Score 43.2/100; 63.8% scoring below 50; 0.022% AI-Ready (GEO ≥ 80), about 1 in 4,600, being 216 sites; 0% Verified (GEO ≥ 90), with the highest score measured anywhere at 86.6; around nine in ten sites carrying no Organization schema. The edition scored more than 1,000,000 websites. Source: searchscore.io/savi-report/august-2026/. The next edition is due Q4 2026; current live figures are at searchscore.io/data/.
• UK edition (2026): 31,152 UK sites; average 42.6/100. Source: searchscore.io/savi-report/uk-april-2026/.
• UK Accountancy edition (published 25 June 2026): 1,038 UK accountancy firms; average GEO Score 53/100; top score 80.2 (chacc.co.uk), the single firm reaching AI-Ready (GEO ≥ 80) and none reaching Verified (GEO ≥ 90); zero firms publishing answers in an AI-quotable format; 94% display qualifications and 84% name an author, but only 8% label that author in a machine-readable way (the "Author Paradox"); adding structured-data labels worth nearly 20 points, with roughly four in ten firms still lacking them. Source: searchscore.io/savi-report/uk-accountancy-2026/.
• UK Dentistry edition (2026): 3,399 practices; average 52.8/100; two practices reaching AI-Ready (80+), none Verified; 0% publishing AI-quotable answers. Source: searchscore.io/savi-report/uk-dentists-2026/.
• UK Law Firms edition (2026): 9,487 firms; average GEO Score 51.4/100; three firms reaching AI-Ready (80+). Source: searchscore.io/savi-report/uk-law-firms-2026/.
• UK Aesthetic Clinics edition (2026): 1,567 clinics; average GEO Score 56.2/100; one clinic reaching AI-Ready (80+). Source: searchscore.io/savi-report/uk-aesthetic-clinics-2026/.
• UK Care Homes edition (2026): 2,307 homes; average GEO Score 48.3/100; none reaching AI-Ready (80+). Source: searchscore.io/savi-report/uk-care-homes-2026/.
• UK Veterinary edition (2026): 1,977 practices; average GEO Score 52.4/100; none reaching AI-Ready (80+). Source: searchscore.io/savi-report/uk-veterinary-2026/.
• Six-sector aggregate cited in Chapter 3 (19,775 businesses; 7 AI-Ready; about 1 in 2,825): the sum of the six UK sector editions listed above. Additional sector editions at searchscore.io/savi-report/.

Case study

• FileSeal vs DocuSign in ChatGPT: SearchScore case study, searchscore.io/case-studies/fileseal-chatgpt-docusign/. Figures and quote as published.

Where the book stops

Two areas are deliberately out of scope in the first edition: enterprise governance across multi-site and multi-brand organisations, and international and multilingual AI search. Both are planned for a future edition. Anything published on either in the meantime will be linked here.