State of GEO

1. Thesis

Generative Engine Optimization in mid-2026 confronts a paradox: 71% of all AI citation sources are single-engine exclusive — each platform reads a different web — yet peer-reviewed research shows self-bias causes AI to disproportionately cite AI-generated content, driving 79.6% of simulations toward diversity collapse. Against this centrifugal dynamic, brands with integrated SEO+GEO strategies report 81% measurable gains vs 36% for siloed approaches — because every platform converges on entity clarity, third-party proof, and extractable evidence. 94% of B2B buyers now use LLMs for vendor research; the question has shifted from ‘how do I rank?’ to ‘does every AI platform know who I am — and can they prove it from my content?‘

2. The State Now

AI Overviews trigger on 86.7% of US searches (commercial-intent 88.5%), up from 56.9% YoY, reaching 2.5B monthly users — outpacing ChatGPT’s 800–900M weekly. On an all-query basis AIOs appear on 48–50% of Google US queries — the 86.7% figure above measures commercial-intent prompts specifically, a different denominator — though 70% of triggering keywords show volatility; only 4–8% of AIO citations overlap with organic top-20. 37% of consumers now start search with AI; only 8% click through when an AI summary is present. Google shipped a Search Console AI visibility report — impressions only, no clicks — and confirmed no AI-specific schema is required: “GEO is still SEO at the core.” Content with author attribution is 4x more likely to be cited; the freshness window was cut from 365 to 180 days. Google I/O 2026 made AI Mode the default for 1B+ users, and the June 2026 Spam Update extended SpamBrain enforcement to AI Overviews and AI Mode. 50% of web articles are now AI-generated — the raw material of the self-reinforcing feedback loop underlying the diversity-collapse risk.

But aggregating platforms conceals more than it reveals. CiteLens confirms each engine follows its own citation rulebook: Google AI Mode cites top-10 organic at 93%, Perplexity at 89%, Claude follows brand search demand (53%), ChatGPT ignores both (30% overlap). 44% of cited pages appear in exactly one answer before vanishing (median lifespan 11–15 days).

ChatGPT has the narrowest aperture: it cites the vendor’s own first-party site 68% of the time, but only 15% of retrieved pages get cited — two-stage source selection. Wikipedia accounts for 29.7% of top cited pages, brand homepages 23.8%. Citation follows steep position decay — top retrieval result cited 58%, position 10 only 14%. ChatGPT and Google organic share only 4.2% of cited URLs — near-zero overlap. ChatGPT’s traffic share dropped 22 points to 64.5% as Gemini rose to 21.5%.

Perplexity is the freshness engine: content updated within 30 days earns 82% citation rate vs 37% for content over a year old — a 45-point advantage. It responds to new content in 5–7 days vs 3–5 weeks for ChatGPT and cites 13.8 sources per query (breadth) vs ChatGPT’s 7–8 (depth). 46.7% of its top-10 sources are Reddit. Perplexity cites at 17.7% vs ChatGPT’s 8.8% and converts B2B sign-ups at 11x traditional search. Perplexity abandoned advertising entirely in 2026.

Claude cites brand domains 64% of the time with zero Reddit citations yet is a UGC outlier at 2–4x the rate of other engines — its user-generated citations come from communities other than Reddit. It biases toward premium editorial — NYT, Atlantic, New Yorker at ~2x the rate of other platforms, with the longest freshness window. Claude and ChatGPT share only 8% of cited sources. Claude leads the enterprise LLM market at 32% share (vs OpenAI 27%) and drives 10.6% of signups but shows as 0.1% in GA4.

Gemini is the most schema-responsive engine, citing brand-owned sites at 52% — highest of any platform — and shares only 13.7% of citations with Google’s own AI Overviews on identical queries. The Gemini app reached 750M monthly users. The Gemini 3 update collapsed AI Overview–organic top-10 overlap from 76% to 38%; fan-out sub-query mapping is now dominant.

Grok exists in its own category: 45% of citations from X/Twitter (34% organic, 11% verified-only). Posts with 5,000+ likes get 4.6x citation probability.

Cross-platform landscape. Community platforms capture 52.5% of all AI citations vs 47.5% brand domains. YouTube generates 188,863 AI citations monthly (+56.4% MoM) — the most-cited single source. Wikipedia reached 83,191 citations (+55.2% MoM). Verified structured data accounts for 54.53% of distinct citation sources. 84% of AI citations come from earned media (Muck Rack third edition); journalism alone 27%. Conductor’s benchmark across 13,770 domains found AI referral = 1.08% of all web traffic, growing 1% MoM, ChatGPT driving 87.4%. Top-3 Google pages are 34x more likely to be AI-cited than pages ranked 31–100 — traditional SEO ranking still gates AI visibility.

Content architecture converges on extraction. Answer capsules immediately under H2 are the strongest single content signal — 72.4% citation predictor. Cited sentences average 10 words; nothing over 18 words gets cited. 44.2% of LLM citations come from the first 30% of text; claim-rich introductions get cited 2.1x more. Topical authority (0.76) predicts ChatGPT citations 4x better than domain authority (0.18). Source diversity compounds: one type = 18% coverage, five or more = 78%. Named authors are 25% more likely to be cited; opinion and analysis is cited more than neutral descriptions. Median cited content age collapsed to 298 days.

Economics and attribution. On queries with AI Overviews, organic CTR drops 61%; even queries without them saw 41% decline. AI traffic converts at 14.2% average vs Google organic’s 2.8% — a ~5x lift — with spreads by platform (Claude 16.8%, ChatGPT 14.2%, Perplexity 12.4%); a separate benchmark orders them differently (ChatGPT 15.9%, Perplexity 10.5%, Claude 5.0%) — platform conversion is measured inconsistently across studies. Brands cited in AI Overviews see +35% organic CTR and +91% paid CTR. Yet 27% of signups come from AI vs 0.5% in Google Analytics — a 54x undercount. 60% of AI-cited URLs never appear in Search Console or standard rank trackers; Citation Reach Rate averages 33%.

B2B is the most advanced frontier. B2B SaaS dominates at 34% of tracked citations. Forrester finds AI tools are the #1 cited information source in B2B buying. B2B AI traffic converts 9–23x vs organic. Healthcare publishers are losing 34–46% organic traffic as 63–85% of health queries answered directly in AI Overviews.

Technical barriers and enforcement. AI crawlers now account for 40–50% of bot-level activity; 65–70% are live user queries, not training. 73% of websites block AI crawlers; JS-rendered content fails 77% of parser attempts — static HTML with schema achieves 94% parse rate. AI systems fetch pages with ~2s hard timeout; TTFB under 1s is critical. From September 15, 2026, Cloudflare’s three-tier classification blocks Training and Agent crawlers by default on ad-supported pages. llms.txt has 784+ implementations but only Perplexity and Anthropic confirm use — 97% of published files get zero AI requests, and Google explicitly ignores it.

Local GEO is an open frontier. 88% of local businesses have no active GEO strategy. AI search shifts from proximity-first to context-first: intent and relevance override geographic distance. A 150+ review threshold applies for consistent AI visibility. Incomplete GBP data triggers a silence penalty.

The schema paradox persists. One study shows 2.3x citation lift from FAQPage, HowTo, Article, Product schema; another tracked 1,885 pages adding schema vs 4,000 controls and found zero independent effect. Both are valid: schema amplifies quality but does not create it. Attribute-rich Product/Review schema outperforms generic by 20pp (61.7% vs 41.6%). FAQPage triggers in 89% of Q&A queries with 3.2x lift.

3. What’s Winning

Entity remains the dominant signal class. The SEO-to-GEO Divergence Index shows brand entity mentions at NIS 0.918 vs domain rating at 0.397. Branded web mentions correlate r=0.664 vs backlinks r=0.218; top quartile averages 169 AI Overview mentions vs 14 for the next tier. The Brand Stature Ladder shows global brands at 73% citation rate, mid-market 44%, niche 11%. Wikidata QID with sameAs schema lifts citations 40% with zero ongoing cost. Brands with a Wikipedia presence earn 3x higher ChatGPT citation rates.

Review depth is the single highest correlated variable: 50+ reviews on G2/Capterra = 3.2x citation rate. Review velocity (r=0.72) outperforms star rating (r=0.31). A B2B SaaS case study went from 12% to 87% citation rate in 4 months by scaling G2 reviews from 43 to 287 plus 14 PR placements.

Answer-first content architecture. Answer capsules under H2 = 72.4% predictor. HubSpot’s AEO template — answer within 60 words after H2 — drove 3x lead conversion. Named-source quotations with credentials: +42.6% lift (strongest Princeton GEO tactic). 19+ data points per page: 5.4 citations vs 2.8. Original datasets remain the highest-ROI GEO tactic — AI cannot synthesize data it cannot access. Document structure drives 17.3% lift, headings accounting for 44.9% of the gain.

Third-party proof. Digital PR drives 25% of LLM citations but only 6% of practitioners use it — the widest evidence gap in GEO. An industrial lighting brand achieved 88.6% AI recommendation rate — own site contributed 4.5%, PR reprints drove 59.8%. Comparison tables +34%, llm.txt +32%, FAQ schema +28% coverage lift within 14–21 days. Source diversity compounds: one type = 18%, five or more = 78%.

Case study convergence. Compiled benchmarks show AI traffic gains up to 2,300% in the strongest documented case (a manufacturing vertical), clustering around a 90–180 day inflection point. Fulton: zero to $18,164/month in 12 months, Google organic doubled as side effect. Fintech B2B: +61% branded AI mentions, 189% comparison traffic growth. B2B SaaS outranked Salesforce and Zapier on 40+ prompts in 30 days. ChatGPT shopping citations surged from 8% to 87% in 5 months; AI-referred traffic converts at 30–40%. The GainFrame playbook — Quick Answer blocks, question H2s, FAQ schema, multi-schema JSON-LD, IndexNow — made ChatGPT the top acquisition channel at 31%.

4. Losing Signals

The KB currently flags no signals as formally deprecated or stale. Several contested findings and structural constraints are worth tracking.

Google self-conflict. Gemini and AI Overviews share only 13.7% of citations on identical queries. AI Overview–organic overlap collapsed from 76% to 38%. Optimizing for one Google surface does not optimize for another.

Only 36 global brands maintain top-100 visibility across ChatGPT, Gemini, AI Mode, and AI Overviews simultaneously.

Store-bought backlinks show near-zero transfer to AI citations. Google’s listicle crackdown (January 2026) caused 29–49% visibility losses; self-promotional listicles get cited but 43% of the time AI recommends a competitor.

Credibility paradox: being cited in AI answers does not guarantee being believed (Burson/Profound). AI Overviews reduced Wikipedia traffic by ~15% — causal evidence that AI answers cannibalize even authoritative sources.

Freshness is real but overclaimed. The controlled advantage is 25.7%, not the viral 4.3x figure. But the ~4.5-week half-life makes continuity discipline mandatory regardless.

5. What Changed This Cycle

This is the first biweekly update (the second edition) of the State of GEO living document. 109 new KB entries were incorporated across all sections. Significant additions include: the AI search diversity collapse finding (Graphite peer-reviewed), platform-specific citation rulebooks (CiteLens), the 44% one-hit wonder problem, ghost citations persisting 8+ months after deletion, content median cited age collapsing to 298 days, content velocity compounding (12+/month = 200x faster), topical authority as a 4x better predictor than domain authority, the llms.txt adoption reality check (784 implementations but near-zero usage), Perplexity’s 6-stage RAG pipeline and ChatGPT’s Labrador VIP tier, Google’s S-CTS/S-BERT AI detection systems, the June 2026 Spam Update extending to AI surfaces, local GEO as an open frontier (88% of businesses have no strategy), and diverging monetization paths (ChatGPT ad CPM collapse vs Perplexity abandoning ads). 69 existing entries were updated with new data points. No entries were removed.

6. What To Do

  1. Audit per platform, not in aggregate. Use GSC’s AI Mode and AI Overviews segmentation. Add dedicated tracking for ChatGPT, Perplexity, Claude, Gemini. The CiteLens finding that each engine follows distinct citation rules makes aggregate optimization impossible.

  2. Build the entity foundation first. Wikidata entry + sameAs Organization schema. Zero-cost, 40% citation lift. Wikipedia presence for 3x higher ChatGPT citation rates.

  3. Shift link budget to mention budget. Brand mentions are 3x stronger than backlinks. PR placements, G2/Trustpilot reviews, Reddit community. Review depth (50+ = 3.2x) and review velocity (r=0.72) are the highest-leverage activities. Source diversity compounds: one type = 18% coverage, five = 78%.

  4. Restructure for passage-level extraction. Every H2 needs a direct answer block in the first 40–60 words. Answer capsules under H2 = 72.4% predictor. Cited sentences average 10 words — short, front-loaded wins. HubSpot’s template drove 3x conversions. Named-source quotations (+42.6%), data-rich pages with 19+ stats, and original datasets compound.

  5. Adopt weekly-to-monthly publication cycles. Content <13 weeks is ~2x more likely to be cited. Content velocity 12+/month = 200x faster AI visibility. Perplexity rewards days-old content. Comparison tables +34%, llm.txt +32%, FAQ schema +28% coverage lift within 14–21 days.

  6. Audit AI crawler access — hard deadline. AI crawlers are 40–50% of bot activity — ensure 14+ user-agents are allowed in robots.txt. JS-rendered content fails 77% of parser attempts. TTFB under 1s is critical for the ~2s hard timeout. From September 15, 2026, Cloudflare’s three-tier classification blocks Training and Agent crawlers by default on ad-supported pages.

  7. Deploy AI Discovery Files, but temper expectations. llms.txt, llms-full.txt. Anthropic and Perplexity confirmed support; Google and Meta do not. 97% of llms.txt files get zero AI requests. File suite works as a competitive differentiator for long-tail agent-readiness, not broad discovery.

  8. Budget for continuity, not campaigns. $35–60K/year, 15–30% of marketing budget, 12–18 month payback. The 4.5-week half-life means GEO is a continuity discipline. Integrated SEO+GEO strategies report 81% gains vs 36% for siloed. The GEO market is $365.4M growing at 42.9% CAGR.

  9. Local businesses: act now. 88% have no GEO strategy. Target 150+ reviews, complete GBP data, and apply the FACTS framework.

7. Method Note

This is the second edition (first biweekly update) of the State of GEO, a living document synthesized from curated findings in the Citonyx knowledge base. Sources include peer-reviewed studies (arXiv, ACL 2026), independent audits (Ahrefs, Semrush, Averi, Profound, Yext, OtterlyAI, BrightEdge, SE Ranking, Ranqo, AirOps, Erlin, Foglift, Peec AI, Previsible, Graphite, CiteDash, CiteLens, QuickSEO, Zyppy, Meltwater, NeuraPulse, BeVisibleIQ, Discovered Labs, Passionfruit, Magna AI) and documented case studies. Each factual claim links to its source entry for independent verification. Underlying studies are predominantly English-language and US-centric. Given the ~4.5-week citation half-life, treat this as a July 2026 snapshot. Updated every two weeks.