Every source behind the scores.
Evidence-backed findings from controlled studies, peer-reviewed research, and official platform documentation — tiered by credibility, updated as new research lands. No opinion pieces. No tips roundups. Only hard evidence.
September 2026
189 branded prompts generated 1,797 sub-queries, averaging approximately 9.5 queries per prompt.
MJ Cachón · · Controlled experimentCover fan-out zero-volume queries
Google introduced multi-account support allowing media companies to manage multiple brand Search profiles from a single Google Account login.
Google · · Official documentation· ✓ OfficialEstablish your brand entity
Across 120 annotated responses to incorrect tool returns spanning five models, none of the models explicitly flagged or acknowledged conflict with contradictory information.
arXiv · · Controlled experimentCite authoritative sources and expert quotes
95% of deals go to vendors already on the initial shortlist formed before formal buying activity.
6sense · · Observational studyEstablish your brand entity
Google's systems may retain negative assessments of a site based on its older low-quality pages, even after the site is updated with better content.
Google (via Bluesky) · · Official documentation· ✓ OfficialAvoid keyword stuffingDemonstrate E-E-A-T signals
Uberall analyzed 120,000 AI mentions across 3,793 multi-location businesses in five AI models to identify recommendation patterns.
Uberall · · Controlled experimentDemonstrate E-E-A-T signalsEstablish your brand entityCite authoritative sources and expert quotesImplement JSON-LD schema markup
YouTube was the second most-cited social media platform in an analysis of 100+ million AI citations.
Otterly · · Observational studyBuild YouTube presence for AI citationCite authoritative sources and expert quotes
Cloudflare published a reproducible formula for the crawl-to-refer ratio with full methodology and limitations disclosed.
Cloudflare · · Official documentationAdd statistics and quantitative data
Retrieval does not equal citation: 1.4M ChatGPT prompts analyzed; Reddit URLs retrieved at scale but cited in only 1.93% of cases, showing AI systems access content without necessarily crediting it in final responses.
Ahrefs · · Controlled experimentOptimise for Perplexity citations
10% of over 1.2 million sites that respond to llms.txt requests have mistakenly placed robots.txt at the llms.txt path instead, indicating widespread confusion about the two file types.
Common Crawl · · Controlled experimentAdd an llms.txt file
Google states that recovery from algorithmic ranking penalties takes months, not weeks.
Reddit · · Official documentationKeep content current and updated
Of 1,851 AI citations analyzed, brand-owned pages received only 2.8% overall, but when the brand was recommended by name, its own page was cited 31% of the time, meaning 69% of brand mentions resulted in third-party citations.
Shero Commerce · · Observational studyOptimise for Perplexity citationsEstablish your brand entityOptimise for Google AI Overviews
Community and UGC sources (Reddit, YouTube, forums, Quora, social media) represent approximately 50% of leading citations across IT services (51%), consumer goods (50%), and communication services (50%) verticals.
Trendos · · Controlled experimentCite authoritative sources and expert quotes
Counted listicles (with explicit item numbers in title) win 56.8% of direct matchups against editorial roundups and rank in top 3 55.9% of the time versus 41.4% for roundups.
Growth Memo · · Controlled experimentStructure comparisons as semantic HTML tables
Google disclosed four explicit factors used to evaluate site reputation abuse in manual reviews, providing measurable criteria for content assessment.
Google · · Official documentation· ✓ OfficialEstablish your brand entityDemonstrate E-E-A-T signalsDisplay author bio and credentials
Google deployed link carousels for developing topics in AI Mode to increase clickability and surface relevant articles alongside AI-generated answers.
Google (X/Twitter) · · Official documentation· ✓ OfficialOptimise for Google AI OverviewsCite authoritative sources and expert quotes
.com domains show AI-authorship markers at 9.35%, compared to 4.59% on .org, 1.03% on .edu, and 0.76% on .gov domains.
Pew Research Center · · Controlled experimentUse domain-specific technical language
August 2026
When brands were recommended by AI systems, their own website received the citation only 31% of the time; third-party sites received the remaining 69% of citations despite the brand being named in the recommendation.
Shero Commerce · · Observational studyEstablish your brand entityOptimise for Perplexity citationsOptimise for Google AI Overviews
AI agent traffic grew 45% quarter-over-quarter to 17.7 billion requests in Q2 2026, measured across DataDome's bot-defense network.
DataDome · · Observational studyEnsure AI crawlability and indexing
Facebook content was cited in 19.5 million Google AI Overviews, Instagram in 877,000, and TikTok in 78,000 across 300 million U.S. searches analyzed.
BrightEdge · · Observational studyCite authoritative sources and expert quotesOptimise for Google AI OverviewsEstablish your brand entityEarn digital PR and brand mentions
Reddit's ChatGPT Search citation share fell from 3.83% to 0.52% between July 18–Aug 17, representing an 86.4% relative decline, with distinct drops on Aug 8 and Aug 14.
Promptwatch · · Observational studyOptimise for Perplexity citations
Across three LLM models and 250 category queries, models showed 91.6% majority agreement on which brands to include in top recommendations, even when disagreeing on rank order (41.6% exact agreement on #1).
arXiv · · Controlled experimentEstablish your brand entityAdd statistics and quantitative data
Anthropic confirmed its watermark is based on Google DeepMind's peer-reviewed SynthID-Text method from a 2024 Nature publication, using randomness-source modification rather than embedded characters.
Anthropic · · Official documentation· ✓ Official
Analysis of 1,249 ChatGPT answers found no difference in format, length, or freshness between pages served from OpenAI's in-house index regardless of whether publishers had licensing deals.
Resoneo · · Controlled experimentKeep content current and updatedCite authoritative sources and expert quotes
Gemini app reached 1 billion monthly users as of August 2026, with 63% actively using voice interaction.
Google · · Official documentation· ✓ OfficialAdd statistics and quantitative data
64% of agencies rank Google AI Overviews as the top industry concern; 59% cite AI search disruption as a major concern.
AgencyAnalytics · · Controlled experimentOptimise for Google AI Overviews
Define Media Group analyzed Search Console data from 64 websites and measured a 42% decline in organic clicks since Google AI Overviews scaled.
Define Media Group · · Observational studyOptimise for Google AI Overviews
AI detectors show systematic bias against non-native English writers, producing higher false-positive rates for their work compared to native English writers.
Stanford HAI · · Observational study
Frontier LLMs encode 95-98% of facts during training but fail to directly recall 26-34% of encoded facts in response to queries; recall failures account for over 70% of errors.
Google Research · · Official documentation· ✓ OfficialDemonstrate E-E-A-T signalsUse definitive language and entity echo
86% of Google clicks originate from discovery behavior rather than direct intentional searches.
iPullRank · · Observational studyAdd statistics and quantitative data
Of all citations used by LLMs for branded and bottom-funnel queries, 48% originate from earned media, 30% from third-party commercial content, and only 22% from brand-owned websites.
Be Omniscient · · Observational studyCite authoritative sources and expert quotesDemonstrate E-E-A-T signalsEarn digital PR and brand mentions
Researchers analyzed 1,200 ChatGPT answers and 26,900 distinct pages to identify the retrieval pipelines ChatGPT uses for citations.
Suganthan (independent blog) · · Controlled experimentOptimise for Perplexity citationsEnsure AI crawlability and indexing
Model accuracy on facts causally correlates with volume and variety of relevant pretraining documents; thin-support subjects require orders-of-magnitude model scaling to answer competitively.
arXiv · · Controlled experimentCite authoritative sources and expert quotesEstablish your brand entityDemonstrate E-E-A-T signalsAdd statistics and quantitative data
Only 40–40.7% of all searches result in clicks to independent websites; 17–20% of searches click to Google-owned properties, indicating record low traffic to the open web.
Datos / SparkToro · · Observational studyOptimise for Google AI Overviews
User-generated content platforms cited 4.3x more frequently than publisher domains across 35,000 ChatGPT citations for SaaS prompts in December 2025.
G2 · · Controlled experimentCite authoritative sources and expert quotesEstablish your brand entityEarn digital PR and brand mentions
EU AI Act Article 50 mandates disclosure of AI-generated text for public interest content unless the content undergoes human editorial review and someone holds editorial responsibility.
European Commission · · Official documentation· ✓ OfficialCite authoritative sources and expert quotes
Across 1,094 U.S. product and service categories tested, 84.8% of categories have no single brand with clear dominance in ChatGPT results.
Growth Memo (Substack) · · Controlled experimentEstablish your brand entityUse domain-specific technical language
Google released OKF v0.2 with five new trust-related fields: provenance (sources), trust (generated/verified), freshness (stale_after), lifecycle (status), and attestation (Attested Computation type).
Google Cloud · · Official documentation· ✓ OfficialCite authoritative sources and expert quotesImplement JSON-LD schema markupKeep content current and updatedDemonstrate E-E-A-T signals
Google flags spammed internal search result pages as hacked content in Search Console.
Google · · Official documentation· ✓ OfficialEnsure AI crawlability and indexing
Google's crawl capacity limit is shared across all crawlers, meaning high demand from one crawler reduces capacity available to others.
Google · · Official documentation· ✓ OfficialEnsure AI crawlability and indexing
July 2026
Google's official documentation states that noindex directives are ineffective when pages are blocked via robots.txt, as crawlers cannot access the page to read the tag, potentially allowing the page to still appear in search results.
Google · · Official documentation· ✓ OfficialEnsure AI crawlability and indexingAvoid keyword stuffing
Google Ads impressions declined 11% year-over-year in Q1 2026 according to Optmyzr's benchmark data.
Optmyzr · · Observational studyAdd statistics and quantitative data
Trust in AI assistants in the U.S. is 28%, compared to 70% for search engines—a 42-point gap.
YouGov · · Observational studyCite authoritative sources and expert quotes
Google's John Mueller stated that 'are you a bot' interstitial pages can be indexed instead of actual content, causing real content to drop from the index or be replaced by the interstitial page.
Google · · Official documentation· ✓ OfficialEnsure AI crawlability and indexing
AI Overviews appear in 68% of local business-type searches, while local packs only appear in 39% of the same queries, creating a 29-point visibility gap.
Whitespark · · Controlled experimentOptimise for Google AI Overviews
Google official documentation specifies a two-week maximum window for content-based canonical fixes to take effect in search results, with faster resolution when new content is more distinct.
Google · · Official documentation· ✓ OfficialImplement JSON-LD schema markup
Google's patent describes a persistent autonomous search system where search sessions remain active in the background after the initial query, continuing to evaluate newly available information until satisfactory answers meet quality, authoritativeness, or completeness standards.
Google · · Official documentation· ✓ OfficialOptimise for Google AI Overviews
ChatGPT prompts leaked into Google Search Console, with AI searches creating inflated impressions while clicks remained flat, a phenomenon termed 'crocodile mouth' in Search Console.
Quantable · · Observational studyAdd statistics and quantitative data
A controlled study found that schema markup does not function as a citation cheat code for AI search results.
No Hacks · · Controlled experimentImplement JSON-LD schema markup
When AI assistants (Claude, ChatGPT, Gemini) received a follow-up question with buyer context (e.g., 'for a small team'), 62% of brand recommendations from the initial answer were removed, compared to 10% churn when the same question was simply repeated.
Clovion AI · · Controlled experimentEstablish your brand entity
Google warns that running A/B tests for unnecessarily long durations may be interpreted as an attempt to deceive search engines and could result in action taken against the site.
Google · · Official documentation· ✓ OfficialAvoid keyword stuffing
Text ads appeared on 29.45% of commercial queries analyzed, with ads appearing on 14,733 queries across U.S. commercial searches.
SE Ranking · · Controlled experimentOptimise for Google AI Overviews
Across 30 platform-topic tests on three AI engines, the sample size required for stable rankings ranged from 33 to 94 queries, with no single cutoff applicable universally.
arXiv · · Controlled experimentAdd statistics and quantitative data
20% of fintech homepages (55 sites) return less than 30% of content in raw HTML, with 47 sites returning zero readable content without JavaScript execution.
Web Performance Tools · · Controlled experimentEnsure AI crawlability and indexingUse server-side rendering for AI crawlers
Google officially disclosed that the new feature will indicate whether ads were created or modified with AI, rolling out globally across Google's main ad surfaces.
Google · · Official documentation· ✓ Official
Labrador retrieval pipeline accounted for 88.1% of primary search sources across 9,946 search runs, with Bright at 9.9%, Oxylabs at 1.7%, and SERP at 0.3%.
Chris Green Substack · · Controlled experimentAdd statistics and quantitative data
Google reported 19% revenue growth in search driven by AI experiences and all-time high queries.
The Verge · · Official documentationOptimise for Google AI Overviews
Google officially confirmed it does not use llms.txt or llms-author.txt files, and is unaware of other crawlers or LLMs using them.
Google · · Official documentation· ✓ OfficialAdd an llms.txt file
Every finding is scored.
Source credibility, recency, and independence decide how much each finding moves a tactic's confidence score.