The Generative Search Economy

Abstract

The digital customer acquisition landscape is undergoing a structural break. Traditional search interfaces, long dominated by index-based directory results, are being bypassed in favor of conversational, synthesized AI recommendation engines. Zero-click search, which stood at roughly 50% in 2019 and 60.45% in 2024, reached 68.01% of US Google searches during the first four months of 2026, while Google AI Overviews now trigger on more than 25% of all searches.[1][2] When AI Overviews appear, the click-through rate of the top organic result falls by approximately 58%.[3] AI search engines themselves post the highest zero-click rates in the history of the web.[4]

Yet the visitors who do click through from AI sources are highly qualified. This paper compares the acquisition efficiency of AI-recommended answers against organic, paid, social, and direct channels; explains why legacy SEO+ and thin programmatic content fail under retrieval-augmented generation; and presents Generative Engine Optimization as the research-grounded alternative. The central finding is that brands must optimize for inclusion, citation, and recommendation inside AI-generated answers, measured through Share of Model, rather than for high-volume clicks that the zero-click economy no longer delivers.[5][6][21]

1. The structural break in search

Randall Fishkin's baseline tracking of zero-click searches, which stood at approximately 50% in 2019, rose to 60.45% in 2024 and climbed to a peak of 68.01% of US Google searches during the first four months of 2026, meaning fewer than one in three Google searches now sends a click to the open web.[1] This phenomenon is driven by the rapid integration of Google AI Overviews, which now trigger on more than 25% of all searches, up from 13.14% in March 2025 and 6.49% in January 2025.[2]

When AI Overviews appear, they reduce clicks to the top-ranking organic page by approximately 58%, and traditional organic click-through rates contract sharply across positions.[3] AI search engines themselves have established the highest zero-click rates in the history of the web, with Perplexity near 93%, Google AI Mode at 88%, ChatGPT Search at 82%, and Microsoft Copilot at 78%.[4]

The commercial consequences are stark. Primary consumer research from Bain & Company indicates that about 80% of consumers now rely on AI-generated answers for at least 40% of their searches, producing an estimated 15% to 25% organic search-volume decline across multiple commercial sectors.[5] Gartner projects that traditional search engine volume will decline by 25% by 2026 as consumers embrace conversational search.[6] Despite this contraction, the visitors who arrive via AI recommendations are unusually qualified, which is why acquisition efficiency, not raw traffic, becomes the decisive metric.

2. Comparative acquisition efficiency: AI referrals versus traditional vectors

Traditional acquisition channels operate on a broad-funnel architecture in which users navigate awareness, research, and consideration across multiple open-web pages. Generative interfaces compress that journey by performing qualification, synthesis, and evaluation inside the conversational session. Consequently, visitors arriving via generative AI recommendations convert at rates that dwarf organic search, paid search, social media, and direct traffic.[7]

Independent analyses confirm that AI-referred traffic converts at 5 to 23 times the rate of standard organic search, even though generative AI traffic still represents roughly 1% of total web sessions across major industries.[7]

Cross-channel conversion and engagement. Figures compiled from AI-referral conversion studies and publisher analytics. [7][9][13]

Ahrefs' internal analytics illustrate the pattern: visitors referred by AI platforms represented only 0.5% of total incoming traffic but generated 12.1% of all new sign-ups, a 24.2-fold conversion premium over traditional organic traffic.[8] Similarly, Microsoft Clarity's tracking of more than 1,200 publisher and news sites found that LLM-referred visitors convert to sign-ups at 1.66%, roughly eleven times the 0.15% rate of traditional search and nearly four times the 0.46% rate of social media.[9] Conductor's benchmark data shows that 73% of visitors from AI referrals convert during their first session, compared with 23% of Google organic visitors.[9]

The financial value of these sessions has risen rapidly. Across retail sites tracked by Adobe Analytics, revenue per visit from AI referrals improved from just 3% of non-AI traffic value in July 2024 to 70% by May 2025, approaching parity by late 2025.[10]

3. Consumer psychology and the dark funnel

The conversion premium is driven by a fundamental shift in user intent. Traditional search sessions are brief and exploratory, with short, fragmented keywords. Generative queries are conversational and contextually dense, averaging around 23 words against the legacy search average of about four.[11] Users engage in multi-turn conversations, outlining detailed personal or business problems, and the model acts as an intermediary that filters options and compares alternatives inside the chat interface, a process often termed the dark funnel.

By the time a user clicks an outbound citation, the selection decision is highly developed. The web visit is no longer for early-stage discovery but to confirm and execute a transaction already decided during the session. This high intent is reflected in engagement: visitors referred by ChatGPT spend an average of roughly 15 minutes on-site versus eight minutes for traditional Google referrals, viewing nearly double the pages per session.[12]

4. Platform-specific acquisition and referral market share

Not all AI platforms generate identical referral volumes or buyer quality, because differences in platform design directly shape click-through behavior.

Platform conversion and referral characteristics, as an observed benchmark pattern from the cited Conductor/Digital Bloom data — not stable platform constants. Per-engine conversion ordering varies by vertical, query mix, and measurement window. [13]

Among AI referrers, ChatGPT is the dominant source, routing about 87.4% of all outbound AI search traffic across major industries; Perplexity behaves more like a research engine with high citation density.[13] In the benchmark data reported here, Claude shows the highest conversion rate at roughly 16.8% — an observed pattern in knowledge-heavy verticals within these panels rather than a fixed property of the platform; the per-engine ordering shifts with query mix, vertical, and measurement window.[13] While absolute referral volumes remain low, AI platforms collectively drove about 1.13 billion referrals globally in June 2025 against Google's 191 billion, and generative AI traffic is growing on the order of 165 times faster than traditional organic search.[14]

Growth signals reinforce the trajectory. Microsoft Copilot rose from 180 sessions in November 2024 to 4,534 by November 2025, a 2,419% year-over-year increase.[15] Early signal And on May 7, 2026, ChatGPT updated its interface to surface clickable brand links directly inside answers rather than hiding them in footnotes, a single change that lifted ChatGPT referral traffic 157.7% in one week and surged homepage referrals 354.7%.[16] Early signal

5. The strategic failure of legacy SEO+ and programmatic scale

In response to declining click-through rates, agencies have introduced SEO+ bundles that combine traditional on-page optimization with programmatic SEO, using databases, code, and automated generation to publish thousands of templated landing pages. These campaigns chase long-tail keyword real estate, but they are increasingly ineffective under modern retrieval-augmented generation.

Scaled content abuse and penalties

The failure mode is specific, and it is worth stating precisely: the problem is thin programmatic scaling, not programmatic scaling as such. Thin programmatic SEO scales content by swapping a single variable, such as a geography or niche, into a static template, producing near-duplicate pages that add no new information. Search engines now penalize these pages: under helpful-content and scaled-content-abuse policies, near-duplicate templated structures are treated as doorway pages designed to manipulate rankings.[17] The result has been severe losses for legacy directory models, with local directories and comparison sites experiencing traffic drops in the range of 70% to 90% alongside broad deindexing of templated subdomains.[17][19]

Data-rich programmatic content, by contrast, can still succeed. Pages generated at scale from genuinely distinct underlying data — each carrying unique facts, figures, and entity relationships not found elsewhere — deliver real information gain on every URL. The same automation that produces thin doorway pages can produce defensible, citable pages when the data layer underneath is unique rather than templated. The penalty attaches to the absence of information gain, not to the use of code to publish at scale; programmatic generation is a delivery mechanism, and its value is decided by what data it serves.[17][21]

Information gain and retrieval inefficiency

AI assistants and crawlers rely on RAG pipelines that are highly sensitive to information gain, a measure of how much new, unique data a document adds beyond a model's existing corpus. Static templates that are 95% identical across a domain offer virtually zero information gain, and RAG engines that recognize the repetition may exclude the entire domain from retrieval to save compute.[21] Old-style programmatic content is also artificially lengthened to hit keyword-density targets, but RAG systems operate in limited token windows, so high-volume, low-density content wastes processing tokens and is less likely to be retrieved and cited.

6. The sourcing disconnect: Google rankings versus AI citation

A major vulnerability of SEO+ packages built around traditional rankings is the gap between page-one search results and AI citation inclusion. Large-scale analyses indicate that fewer than 10% of the web sources cited in generative answers rank in Google's top ten for the same queries. Semrush research shows that about 90% of the pages cited by ChatGPT rank 21st or lower in standard organic results, and roughly 80% of ChatGPT-cited URLs do not appear in Google's top 100.[18]

Traditional SEO focuses on domain authority and backlink profiles, whereas generative engines prioritize factual correctness, structured definitions, and authoritative citations. Legacy B2B SEO+ also relies heavily on listings across major review platforms, yet those platforms have suffered catastrophic declines as AI Overviews answer comparison queries directly: TrustRadius down 92.2%, Capterra down 89%, Software Advice down 86.5%, G2 down 84.5%, and Gartner Peer Insights down 76.5%.[19]

Lily Ray's February 2026 analysis reinforces the link between traditional visibility and AI citation: across the tracked domains, 100% of sites that lost Google organic visibility during core algorithm updates also lost AI search citations, with citation drops of 27.8% on ChatGPT, 23.8% on Google AI Overviews, and 22.5% on average across LLMs.[20] Traditional visibility therefore remains a necessary baseline for discovery, but keyword scaling that lacks unique data or structured formats fails entirely to capture AI visibility.

7. Generative Engine Optimization: the research-grounded alternative

To capture valuable AI-referred traffic, strategy must shift from keyword optimization to Generative Engine Optimization. First defined in a foundational academic study by researchers at Princeton University, Georgia Tech, the Allen Institute for AI, and IIT Delhi, GEO focuses on making content readable, structured, and authoritative for LLM retrieval systems. The study tested optimization strategies across a 10,000-query benchmark and found that the strongest individual techniques improved visibility in AI-generated answers by roughly 30% to 41%, with a headline of up to about 40%.[21]

GEO strategy impact reported in the Princeton / Georgia Tech / Allen AI / IIT Delhi study (GEO-bench, 10,000 queries). Lifts reflect single-method interventions; the strongest land in the ~30–41% range, with the paper’s headline finding "up to ~40%." [21]

These findings show that keyword optimization alone is ineffective for GEO. Models show a strong citation bias toward factual assertions backed by precise numbers, clear source links, and direct quotes from verified experts.[21]

From component tactics to the complete translation layer

The GEO-bench techniques above are component-level interventions — individual edits applied to a human-facing page. They raise visibility, but they are bounded by the page they sit on: a single template still mixes navigation, marketing copy, scripts, and the few high-value signals a model actually wants, forcing the retrieval pipeline to do extra work to find and trust them. The most effective optimization is therefore not a tactic but an architecture: a complete translation layer that delivers every one of those signals at once, in the cleanest possible form, without touching the human site.

A full translation-layer implementation — the model behind platforms such as AnswerShare and Scrunch — reroutes AI bots and crawlers to a parallel, AI-optimized representation while preserving the human-facing experience unchanged. It serves clean-room HTML5, JSON-LD, and markdown payloads for documents from a cached CDN, grounds entities everywhere appropriate, and strips the noise that dilutes machine readability. The clean-room representation carries no client-side analytics overhead: AnswerShare does not add tracking scripts, beacons, tag managers, or analytics libraries to the AI-facing payload. Measurement is performed outside the payload — through edge logs, crawler classification, server-side aggregation, and controlled engine audits — preserving the low-noise surface the crawler receives. The architectural payoff is that it is designed to raise the retrieval-to-citation conversion rate: when a crawler retrieves a page, almost everything it receives is citable signal rather than chrome, so the share of retrievals that fail to convert into a citation collapses toward its floor. (This conversion concept is distinct from the Retrieval Token Cost metric — labeled RTCost in the metrics table — which measures the token-and-latency expense of extracting useful content; the two are unrelated.) Because the payloads are pre-rendered and cached, time-to-first-byte falls from roughly 2,500ms on an uncached dynamic page to about 20ms from the edge, which matters because retrieval pipelines penalize slow or timed-out fetches.

A complete translation layer delivers each isolated GEO-bench tactic simultaneously and at the architectural level, which is why AnswerShare frames full implementation as a Gold Standard Exemplar (GSE) of GEO engineering rather than a bundle of page edits. A GSE is a perfectly structured, high-authority content model that seamlessly blends unique data, explicit citations, and conversational formatting to serve as a structured blueprint for earning AI search citations — it is what AnswerShare builds to for every client.

Instrumenting the layer: the fifteen metrics that drive citation

An architecture is only as good as what it can measure. AnswerShare instruments every property against fifteen metrics that drive citation, organized into four families: a headline panel score, a reputation (NPS) family[27], a set of reasoning-quality keys, and an infrastructure family. Every published judgment score is produced by sending the same composed prompt and the same pre-fetched bot-path evidence to five independent inference engines and rolling their judgments up to a single number using a median with an outlier-drop pass — the median resists a single harsh or lenient grader, which is what makes a published figure defensible and reproducible by an outside party. The definitions and math below are disclosed; the methods used to optimize each metric to its target are proprietary.

At a glance, the fifteen metrics group into four families, each answering a different question an AI engine implicitly asks of a page:

The four metric families, for readers who want the shape of the framework before the detail. Each detailed metric below belongs to one of these families.

The full set, with the plain meaning and the math behind each metric, follows.

The fifteen metrics, continued.

The fifteen citation-driving metrics AnswerShare instruments, with plain meaning and the math behind each. Headline judgment metrics use a five-engine median-with-outlier-drop panel. The thresholds shown are AnswerShare internal engineering targets — the house values used for operational consistency and reproducibility across properties. They are not proposed as industry standards, and the correlation between hitting these thresholds and earning citations is still being validated against first-party measurement; it has not yet been independently published. Metric meanings and formulas are disclosed here; the optimization methods used to reach each target are proprietary.

Sourcing-lift case studies

A September 2025 case study by Go Fish Digital, which systematically restructured a brand’s content for information gain by embedding unique statistics and expert quotes, reported a 43% increase in AI-driven referral traffic, an 83% conversion lift, and a 25-fold higher conversion rate from AI leads than from traditional search leads.[22] Self-reported Separately, a B2B technology client working with Percepture reported a 212% increase in sales-qualified leads after optimizing for entity-rich conversational structures, with AI-referred traffic converting at rates about 12.8 times higher than standard search.[23] Self-reported When a brand is cited directly inside an AI response, it builds user trust before the visitor even reaches the site.

8. Sourcing differences and platform-specific citation patterns

AI platforms are not uniform in their sourcing models. ChatGPT and Perplexity display a pronounced citation chasm, relying on different data sources to construct answers.[24]

Platform citation frameworks. Citation counts per query reflect Qwairy Q3 2025; the ~10.4 figure circulates in secondary analyses while primary data reports ~7.9 for ChatGPT. [24]

ChatGPT effectively trusts what the broader web has aggregated about a brand, favoring third-party directories, Wikipedia, and professional networks. Perplexity functions as a high-authority research engine, favoring academic publishers, government portals, and video content. Because of these distinct models, only about 11% of cited domains overlap between ChatGPT and Perplexity, so a brand optimized for Perplexity’s academic and freshness model may remain nearly invisible inside ChatGPT’s directory-driven retrieval.[24]

9. Attribution modeling and operational implementation

Because generative platforms change user behavior and obscure direct referral data, traditional analytics frameworks systematically underreport AI-driven conversions. ChatGPT strips referrer data from paid-tier sessions via the noreferrer attribute, while clicks originating from Google AI Overviews are classified as standard organic search rather than AI-referred traffic. Many users also hold a detailed AI conversation, receive a brand recommendation, and then open a new tab to search that brand directly, so analytics attribute the conversion to direct or branded organic search and hide the true impact of the AI citation.

GA4 integration

To address the gap, organizations should build a custom channel group in GA4 dedicated to AI platforms, created under Admin, Data Display, Channel Groups, and placed above the standard Referral and Organic Search channels so it is evaluated first. The grouping rule uses a regular expression to isolate direct AI referrals:

.*chatgpt.*|.*openai.*|.*perplexity.*|.*claude.*|.*anthropic.*|.*gemini.*|.*copilot.*|.*poe\.com.*|.*you\.com.*

The legacy bard pattern has been dropped from the active rule because Bard is no longer a consumer-facing brand; retain it only in a separate historical-referrer segment if continuity with older data matters. Because AI referrer patterns shift as engines launch, rename, and change how they pass referrer data, this regular expression should be reviewed monthly and updated as new engines and domains appear.

Dynamic personalization

To scale programmatic content safely, organizations must move beyond single-variable templates and use content engines that dynamically assemble pages from context-specific logic blocks, swapping features, pricing, and integration data based on industry use case rather than swapping a single noun. To keep scaled builds high-value, publishers should monitor a uniqueness ratio across generated pages:

Unique fields are the data fields that vary across pages and total fields are the content fields in the template. Programmatic systems should require U ≥ 0.60 and pull from at least three independent databases, such as local market data, user reviews, and product-specific APIs, to satisfy the information-gain thresholds needed for indexation and retrieval.

10. Share of Model as the operational baseline metric

With ranking positions becoming less reliable, Share of Model (SoM) is the primary metric for visibility in the generative search economy:

To measure SoM, organizations should run regular automated tests using LLM APIs across a controlled prompt panel. A minimal reference implementation:

import openai
import pandas as pd
from datetime import datetime
import re

BRAND = "target_brand"
COMPETITORS = ["comp_a", "comp_b", "comp_c"]
PROMPTS = [
    "What is the best enterprise CRM for healthcare?",
    "Compare the top B2B SaaS marketing tools in 2026",
    "Recommend an automated supply chain tool",
]

def _mentioned(name, text):
    # Word-boundary match avoids substring over-counting
    # (e.g. "Ace" inside "space"). See production note below.
    return re.search(r"\b" + re.escape(name.lower()) + r"\b", text) is not None

def analyze_ai_share(prompt, brand_name):
    client = openai.OpenAI(api_key="sk-...")
    response = client.chat.completions.create(
        model="gpt-4o",
        messages=[{"role": "user", "content": prompt}],
        temperature=0.3,
    )
    text = response.choices[0].message.content.lower()
    return {
        "timestamp": datetime.now().isoformat(),
        "prompt": prompt,
        "brand_cited": _mentioned(brand_name, text),
        "competitor_citations": {c: _mentioned(c, text) for c in COMPETITORS},
    }

# Execute evaluations, compile data, and track SoM monthly

By querying APIs across 50 to 200 industry-specific prompt variations, organizations can establish a baseline AI visibility rate, track average citation positioning, and measure competitor share-of-voice over time.

11. Strategic conclusions

The empirical record shows that generative search has changed the dynamics of digital customer acquisition. While legacy SEO and paid campaigns chase high-volume clicks, the rise of zero-click search means brands must optimize for visibility inside AI-generated responses. Traditional SEO+ bundles and basic programmatic scaling fail to achieve that visibility because they lack information gain and do not provide the structured, factual data that modern RAG models require.[17][18][21]

Generative Engine Optimization offers a structured, research-grounded framework to improve a brand’s citation rate and authority across conversational engines. Component tactics such as structured data, verified statistics, and expert quotes each help, but the most effective implementation is architectural: a complete translation layer that serves clean-room, grounded payloads from a cached CDN and drives the share of failed retrieval-to-citation conversions to its floor. By adopting that architecture and tracking performance through Share of Model, organizations can capture highly pre-qualified, high-converting referral traffic and make their acquisition strategy more resilient to AI-mediated discovery in an AI-driven search market.[21][22]

Sources and Notes

All sources below are public resources used to ground this whitepaper. Accessed June 20, 2026 unless otherwise noted. Vendor case studies are labeled as self-reported where applicable. The in-text superscript numbers throughout this document are clickable and link directly to the corresponding source.

Source reliability tiers. To help readers weight the evidence, each source is assigned one of three reliability tiers. The tier reflects the nature and independence of the underlying data, not whether a given figure is correct.

Tier 1 — Independent evidence: academic and peer-reviewed research, platform-owned first-party data, and independent third-party measurement (search-engine documentation, and clickstream or analytics providers reporting their own measured data). This tier excludes any party with a commercial stake in the paper’s conclusion.

Tier 2 — Established industry research and large vendor datasets (large-scale studies and benchmark reports published by recognized SEO/AI-search vendors and independent analysts).

Tier 3 — Agency studies, case studies, self-reported vendor results, and AnswerShare first-party datasets (single-client outcomes, vendor-reported wins, and the author’s own measurements). AnswerShare, Top10Lists.us, LAVIDGE, the 100-site audit, and the λNPS companion paper all sit here by design: the paper does not grade its own evidence above independent research.

Identity — Identity-grounding records (Wikidata) are used only to disambiguate the author and the AnswerShare entity. They are not treated as evidence for the economic thesis and carry no evidence tier.

• [1] Tier 1 · SparkToro (Rand Fishkin), "In 2026, less than one-third of Google searches still send a click." Reports 68.01% zero-click rate for US Google searches (Jan–Apr 2026), up from 60.45% in 2024, using Similarweb clickstream data. URL: sparktoro.com →
• [2] Tier 2 · Semrush, "AI Overviews Study," and Conductor 2026 AEO/GEO Benchmarks Report. Confirm AI Overviews at 6.49% (Jan 2025), 13.14% (Mar 2025), and above 25% by late 2025/early 2026. URL: semrush.com →
• [3] Tier 2 · Ahrefs, "Update: AI Overviews Reduce Clicks by 58%." Study of 300,000 keywords (Dec 2025 data) finds position-1 CTR falls about 58% when an AI Overview is present. URL: ahrefs.com →
• [4] Tier 3 · Digital Applied, "Zero-Click Search Statistics 2026," compiling cross-engine analysis; Google AI Mode 93% separately corroborated by Semrush (Sep 2025). Perplexity, ChatGPT, and Copilot figures trace to a May 2026 cross-engine study. URL: digitalapplied.com →
• [5] Tier 1 · Bain & Company, "Goodbye Clicks, Hello AI: Zero-Click Search Redefines Marketing." Reports about 80% of consumers rely on AI summaries at least 40% of the time and an estimated 15–25% organic traffic decline. URL: bain.com →
• [6] Tier 1 · Gartner, "Gartner Predicts Search Engine Volume Will Drop 25% by 2026, Due to AI Chatbots and Other Virtual Agents" (Feb 19, 2024). URL: gartner.com →
• [7] Tier 2 · Ahrefs, "AI Search Traffic Converts ~23x Better" (conversion premium and ~1% traffic share), with corroborating Conductor and Semrush data on AI referral share. Vendor estimates of the premium range widely (~2x–23x) by methodology. URL: ahrefs.com →
• [8] Tier 2 · Ahrefs (Patrick Stox), first-party data: AI search accounted for 0.5% of traffic but 12.1% of sign-ups, a 24.2x conversion premium (June 2025). URL: ahrefs.com →
• [9] Tier 1 · Microsoft Clarity analysis of 1,277 publisher domains (LLM sign-up conversion 1.66% vs 0.15% organic) with Conductor 2026 AEO/GEO Benchmarks Report (73% first-session conversion). URL: conductor.com →
• [10] Tier 1 · Adobe (Adobe Analytics / Digital Insights), "AI-Driven Traffic Surges Ahead in Q2." Revenue per visit from AI referrals rose from 3% of non-AI value (Jul 2024) to 70% (May 2025), approaching parity by late 2025. URL: business.adobe.com →
• [11] Tier 3 · Conversational query length analysis (AI queries averaging ~23 words vs ~4 for traditional search); directional finding corroborated across multiple search-industry sources. URL: ranketai.com →
• [12] Tier 3 · The Digital Bloom, "Generative AI Website Traffic Share, February 2026," citing Similarweb: ChatGPT-referred users spend ~15 minutes on-site versus ~8 minutes for Google referrals. URL: thedigitalbloom.com →
• [13] Tier 2 · Conductor 2026 AEO/GEO Benchmarks Report (ChatGPT 87.4% of outbound AI referrals) and The Digital Bloom (Feb 2026) for platform conversion rates including Claude at 16.8%. URL: conductor.com →
• [14] Tier 1 · Similarweb, reported by TechCrunch: AI platforms drove 1.13 billion referral visits in June 2025, up 357% year over year; WebFX analysis cites ~165x faster AI referral growth. URL: techcrunch.com →
• [15] Tier 2 · Previsible, "2025 State of AI Discovery Report" (1,963,544 LLM sessions): Microsoft Copilot grew from 180 sessions (Nov 2024) to 4,534 (Nov 2025), +2,419% YoY. Small-base figure; read as an early signal. URL: previsible.io →
• [16] Tier 1 · Similarweb, "ChatGPT Referral Traffic Triples." Documents the May 7, 2026 ChatGPT interface update surfacing clickable links, with referral traffic +157.7% in one week and homepage referrals +354.7%. URL: similarweb.com →
• [17] Tier 1 · Google Search Central, "Spam policies for Google web search," defining scaled content abuse and doorway pages; consistent with observed local-directory traffic losses in the 70–90% range. URL: developers.google.com →
• [18] Tier 2 · Semrush AI search SEO study (about 90% of ChatGPT citations from pages ranking 21+ in Google), with Ahrefs AI-search overlap research finding roughly 80% of ChatGPT-cited URLs are not in Google’s top 100. URL: semrush.com →
• [19] Tier 2 · SE Ranking, "Review Platforms in AI Overviews" (Jan 2026): TrustRadius -92.2%, Capterra -89%, Software Advice -86.5%, G2 -84.5%, Gartner Peer Insights -76.5%. URL: seranking.com →
• [20] Tier 2 · Lily Ray, "Are Citations in AI Search Affected by Core Updates?" (Feb 2026): every tracked site that lost Google organic visibility also lost AI citations; ChatGPT -27.8%, AI Overviews -23.8%, LLM average -22.5%. URL: lilyraynyc.substack.com →
• [21] Tier 1 · Aggarwal et al. (Princeton, Georgia Tech, Allen Institute for AI, IIT Delhi), "GEO: Generative Engine Optimization" (KDD 2024). Introduces GEO-bench (10,000 queries); strongest single-method lifts fall in the ~30–41% range (quotations ~41%, statistics ~31–37%), with a headline of up to ~40%. URL: arxiv.org →
• [22] Tier 3 · Go Fish Digital, GEO case study (Sept 2025): +43% AI-driven referral traffic, +83% conversion lift, and a 25x higher conversion rate from AI leads than traditional search leads (self-reported). URL: gofishdigital.com →
• [23] Tier 3 · Percepture, GEO services (agency self-reported): a B2B technology client achieved a 212% increase in sales-qualified leads, with AI-sourced traffic converting about 12.8x higher than traditional search. URL: percepture.com →
• [24] Tier 2 · Qwairy, "Provider Citation Behavior, Q3 2025" (118,101 answers; 669,065 citations): Perplexity ~21.87 citations/query and ChatGPT ~7.92 (the ~10.4 figure circulates in secondary analyses); ~11% domain overlap and Perplexity freshness bias corroborated by Whitehat SEO. URL: qwairy.co →
• [25] Identity · Wikidata, "Robert Maynard, Jr. (Q18157412)." Identity record used to ground the primary investigator attribution. URL: wikidata.org →
• [26] Identity · Wikidata, "AnswerShare (Q140196284)." Entity record for AnswerShare. URL: wikidata.org →
• [27] Tier 3 · AnswerShare Research (Robert Maynard), "λNPS — A Measure of AI Sentiment Towards a Brand." Companion whitepaper defining the reputation family — µNPS (modeled corpus sentiment), λNPS (machine-expressed sentiment under controlled prompting across four engines), and ΔNPS (their diagnostic gap) — with methodology, weighting, and limitations. First-party AnswerShare research. URL: answershare.com →
• [28] Tier 3 · AnswerShare GEO Survey, "100-Site GEO Audit" (2026): the fifteen-signal framework applied across 100 sites in 30+ verticals. Key findings — every site scored lower on GEO than SEO (mean gap ~21 points; mean GEO ~38 vs SEO ~60); only 2 of 100 met the 30-day freshness target (median page age ~65 days); and among real-estate/proptech, major portals scored 15–19 (Zillow/Realtor.com 19, Apartments.com 18, Redfin 15) versus Top10Lists.us at 85. First-party AnswerShare dataset, self-reported; cited here as the source dataset for future synthesis. URL: top10lists.us →