# The No Nonsense Guide to Getting Cited by AI: Technical Edition *By Ethan Young | Created March 12, 2025 | Updated September 3, 2025* AI assistants aren't experimental features anymore. Google's AI Overview (AIO)—Google’s latest [SERP](https://en.wikipedia.org/wiki/Search_engine_results_page) feature that uses [generative AI](https://en.wikipedia.org/wiki/Generative_artificial_intelligence) to deliver answers—now appears for ≥12.8% of Google searches by volume, transforming ~2 billion interactions into AI-first experiences every day.[^1] Meanwhile, ChatGPT answers ~2.5 billion daily prompts, with 330 million coming from the U.S. alone.[^2] These billions of users likely include your most valuable audiences: Business-to-business (B2B) buyer adoption of AI assistants, from discovery and evaluation to decision, is approaching 100%.[^3] The effects are shocking. When Google deploys AIO—typically on informational queries already prone to [zero-click](https://en.wikipedia.org/wiki/Zero-click_result) behavior—both organic and paid [CTR](https://en.wikipedia.org/wiki/Click-through_rate) suffer as users increasingly rely on AI-generated answers without visiting websites.[^4][^5] While AI assistants are siphoning clicks, new traffic sources are exploding. AI-referred traffic in the U.S. grew 10–12× between July 2024 and February 2025, and the quality difference is noticeable: they're highly engaged users who bounce less often, view more pages, and stay longer.[^6] Most remarkably, visitors referred by non-Google AI assistants (e.g., ChatGPT, Perplexity, Claude) can be ~4.4–10× more likely to convert than the average visitor from traditional search (primarily Google).[^7][^8][^9] In short, brands getting cited by AI can counter CTR declines and boost revenue. Most companies today haven't yet grasped this shift in user behavior. They're still optimizing solely for Google rankings, publishing static content, chasing single high-volume keywords, or communicating the same generic message for all audiences across every stage of the buyer journey. That's why they're getting ignored by AI assistants. As a result, they’re missing opportunities to build their reputation and capture new sources of revenue. ## Principles that Shape AI Assistant Citation Behavior Popular frameworks and acronyms treat "AI" as a monolithic technology; proponents prescribe generic or misguided strategies that lead to confusion and wasted effort. But there are no shortcuts. AI assistants behave differently; they are proprietary, complex, and constantly improving—we're unlikely to fully reverse-engineer them. Still, many AI assistants appear to follow established [Information Retrieval (IR)](https://en.wikipedia.org/wiki/Information_retrieval) principles—the same foundational concepts that have guided search engines for decades. By combining published research with patterns in observable behavior, we can learn how AI assistants "think" about questions and answers. This insight lets us build strategies that work across different assistants, adapt as systems change, and align content with how assistants generally find, evaluate, and select sources. ### AI Assistants Have Model Memory and Real‑time Retrieval **Winning citations begins with understanding how AI decides what to believe.** Most assistants draw their knowledge from two pools: 1. **Parametric Knowledge:** what a [Large Language Model (LLM)](https://en.wikipedia.org/wiki/Large_language_model) "remembers" from [training data](https://en.wikipedia.org/wiki/Training,_validation,_and_test_data_sets)—web pages, Wikipedia, papers—frozen at a specific moment in time. It’s encoded in the model's [neural network](<https://en.wikipedia.org/wiki/Neural_network_(machine_learning)>) and forms its baseline view of the world, including you, your brand, and your industry. 2. **External Knowledge:** what an AI assistant [retrieves](https://en.wikipedia.org/wiki/Information_retrieval), often in real-time, to verify, update, or supplement its baseline parametric knowledge. This includes search engines, [knowledge graphs](https://en.wikipedia.org/wiki/Knowledge_graph) (e.g., [Google's Knowledge Graph](<https://en.wikipedia.org/wiki/Knowledge_Graph_(Google)>)), and databases that add current context beyond the model's knowledge cutoff date, improving answers and reducing [hallucinations](<https://en.wikipedia.org/wiki/Hallucination_(artificial_intelligence)>). AI assistants—often called RALMs ("retrieval-augmented language models")—blend these knowledge sources dynamically.[^10] In general, they search when parametric confidence is low or when freshness is required. > Does your content surface unsettled questions in your domain and resolve them? #### External Knowledge Today → Parametric Knowledge Tomorrow **Authority compounds.** Content that’s discoverable via retrieval of external sources *now* can become part of future training data *after* it's archived (even if AI assistants haven't cited you). That means brands can shift into parametric knowledge over time. This external knowledge → parametric knowledge relationship creates an upward spiral where each new model may “know” your brand better than the last. That matters because brands established in an assistant’s parametric knowledge may help set the starting point of answers while competing sources may struggle to break in.[^11][^12][^13][^14] > Are you building the authoritative presence that present and future AI assistants will trust by default? ### AI Assistants Have External Knowledge Source Preferences **Citation rates improve when your content lives where assistants actually retrieve information.** Different assistants show distinct preferences—ChatGPT favors Wikipedia while Gemini leans toward YouTube—reflecting differences in their underlying language models and proprietary knowledge ecosystems.[^15] It’s tempting to chase specific channels. For example, for product comparisons, AIO is ~3.1× more likely to cite a YouTube video than text presenting the same product details.[^16] But hyper-focusing on one platform could mean getting crushed by shifting algorithms or missing buyers discovering, evaluating, and deciding on other platforms: 1. **The landscape shifts:** new assistants emerge, partnerships reshape retrieval sources, and user preferences change. This volatility alone justifies diversification—yet the core channels remain remarkably stable: news and blogs account for 41–73% of citations across all major AI assistants.[^15] 2. **Assistants aren't (always) competing;** they may cover different steps in the buyer journey. While AIO commands ~98% of AI search by volume, ChatGPT quietly drives the most referral traffic of any AI platform, and non-Google assistants drive 4.4–10× higher conversion rates than traditional search.[^6][^7][^8][^17] [^18] This suggests Google Search is moving up-funnel (broad reach, early exploration) while ChatGPT skews down-funnel (deeper evaluation).[^19][^20] It's smarter to invest in formats that survive platform shifts first and then layer in opportunistic bets (e.g., Wikipedia, YouTube) based on observed lift. > Are you diversified enough to weather platform shifts while still showing up at every stage of the funnel? #### AI Assistants Prefer Recently Published Content **Because assistants fetch external knowledge to resolve uncertainty, they’re biased toward recently published content.**[^21] Newsrooms exemplify this, maintaining consistent 20–27% citation rates across assistants by optimizing for publication speed.[^15] Practically, AI accelerates content decay.[^22] On the upside, this dynamic indicates that, within AI assistants, external knowledge has influence over parametric knowledge during specific windows. This timing bias presents an opportunity. While teams face increased content demands, high-quality and up-to-date content can now compete more effectively against established players who have gotten lazy. Additionally, wire-distributed press releases can dominate early windows before comprehensive analyses exist. > Does your content strategy operationalize up-to-date industry commentary? #### AI Assistants Search Far and Wide **AI assistants don't just vary in where they look—they differ in how they search.** Google's "query fan-out" method exemplifies this: it generates multiple sub-queries to uncover relevant content that might never rank for the original search term.[^23] This architectural choice explains a puzzling finding: organic search rankings poorly predict AIO citations.[^24] This suggests that while Google uses its search index for discovery, it doesn't treat ranking position as the primary relevance signal. In this case, your visibility is determined by your rank across the set. So instead of fixating on single-query rankings, prompt your preferred AI assistant to reveal the long-tail sub-queries (across user intents/buyer journeys) for each of your cornerstone topics; sub-query rank may better predict citation likelihood than original query rank because assistants use derivative queries. These hidden questions might surface adjacent subtopics your competitors missed. > Does your content holistically cover the sub-questions around your main topics? #### AI Assistants Resist Shallow Content **Discovery alone won't earn citations.** Your content must deliver real value over presentation hacks.[^25][^26] AI assistants need your content when they can't find that specific perspective, unprecedented depth/accuracy, or data anywhere else. This doesn't require covering absolutely novel topics. It could be an expert analysis that synthesizes scattered public data or one that confirms established ideas with proprietary data. This quality premium will only intensify as inference costs go down and context windows grow. If AI assistants analyze hundreds or thousands of sources per query instead of dozens, they'll compare your content to a much larger pool. In that world, only genuine relevance and unique value will earn citations. > What is your unique information advantage? ### AI Assistants Generate Responses Differently **Response generation methods matter as much as retrieval.** For example, builders are trying to find better ways of handling knowledge conflicts—when AI assistant training data contradicts external sources, or when multiple sources disagree.[^27] In short, assistants don't synthesize conflicting viewpoints very well (on their own).[^28] RALMs show predictable biases in these situations.[^10] Even when presented with accurate corrections, they trust their parametric knowledge over conflicting external sources and favor widely-held viewpoints over minority positions.[^14][^29] Industry patents and academic papers point to a “generate-then-verify” method as a common mitigation strategy: the assistant drafts an answer from parametric knowledge first, then checks it against external sources, potentially revising or dropping segments as needed.[^11][^12] This behavior creates an opportunity. Content that provides definitive resolution to disputed topics may receive preferential treatment, as exhibited by the increased citation rate of comparison content in B2B contexts.[^30] > Does your content resolve contradictions that your industry struggles to reconcile? ### AI Assistants Weigh User and Query Contexts **AI assistants don't just answer questions—they remember who's asking.** For example, ChatGPT now references both "saved memories" (explicitly stored information) and "chat history" (inferred from past conversations) to personalize responses, while Google likely tracks buyer journeys across user profiles.[^31][^32] It's reasonable to assume this trend will continue, with AI assistants increasingly shifting from generating the best average answer to the best individual answer. This context-awareness reinforces why content must comprehensively answer questions across the buyer journey. If you cast a narrow net—covering only a few terms instead of the full fan-out of sub-questions—you’ll miss awareness and let competitors frame the narrative; skip technical rigor and you’ll vanish in evaluation; hide pricing or implementation details and you’ll slow or derail decisions. For example: - **Discovery:** Answer basic “what/why” questions with explainers and glossaries. - **Evaluation:** Help users weigh options with comparisons and pros/cons (e.g., "best X for Y situation"). - **Decision:** Provide pricing, timelines, and implementation details (e.g., “X migration plan to Y”). > Does your content anticipate every angle from which users might learn, evaluate, buy, and troubleshoot your solution? #### AI Assistants Don't Always Trust Your Content **Your business model influences where AI assistants source information about you.** In B2B queries ("top CRM software"), company sites and blogs account for ~17% of citations, while in B2C queries ("best smartphone brands") they’re cited in <4% of citations—meaning the vast majority come from third-party sources.[^30] This gap aligns with differences in user intent: B2B buyers seeking technical specifications or implementation details often look for vendor expertise; B2C shoppers comparing products tend to prefer third-party validation. AI assistants broadly reflect these sourcing patterns. That means companies should align their messages with buyer intent—technical rigor on-domain; consumer proof points off‑domain. For mixed-interest queries (e.g., “top pharmaceutical companies”), news and blogs supply ~70% of citations, highlighting the ongoing importance of third-party coverage for both B2B and B2C brands.[^30] Even so, B2B companies still gain more from comprehensive storytelling, technical documentation, and comparative content on their domain. > Are you investing in the right mix of owned vs earned content for your business model? ### AI Assistants Introduce New Technical SEO Requirements **AI assistants can't cite what they can't see or understand.** AI assistants improve answers by sending their own [web crawlers](https://en.wikipedia.org/wiki/Web_crawler)—OAI-SearchBot, ClaudeBot, PerplexityBot—to dive deeper into pages they already think are relevant. Yet each bot has distinct limitations: Google's crawler demands lightweight websites with fast [JavaScript](https://en.wikipedia.org/wiki/JavaScript) execution, while many AI crawlers don't execute JavaScript at all. That gap means humans may see one thing while assistants see another. The solution should be tailored to your technology stack (framework, CMS, rendering approach) as bots evolve, but generally includes: 1. **Make Content Fetchable:** Remove access barriers and use crawler-friendly rendering so search engines and AI bots can actually reach your content.[^33] 2. **Make Content Machine-readable:** Structure your content semantically and include metadata so machines understand context, not just text.[^34] > Does your company's site/blog make content both fetchable and machine-readable? ## AI Assistant Tracking Requires New Metrics The emergence of AI assistants as zero-click discovery and branding channels presents challenges for traditional attribution methods. However, significant revenue increases are possible if a brand is referenced prominently within these AI environments. For example, even a modest increase in incremental AI-referred sessions (e.g., 10%) combined with significantly higher conversion rates (e.g., 4.4×) can produce meaningful relative revenue lift (e.g., 44%)—even as traditional search traffic declines.[^35] The focus must shift from attributing conversions to understanding the broader influence of AI assistants throughout the entire customer journey. For example, we can monitor AI-referred traffic, bounce rates, page views, and session time vs other traffic sources.[^36] Then, we can observe lift in outcomes by analyzing trends and correlations in data, rather than direct attribution. Still, we need to look beyond traffic reports alone to explain shifts in competitive positioning. #### Benchmark AI Assistants to Quantify Change Measuring content, platform, and assistant performance against business metrics reveals precisely where to double down. The problem is, AI assistants are constantly developing. We need a reliable way to (i) know citation frequency (and extent); (ii) find industry blogs/news sites and topics driving rival citations; (iii) quantify changes in strategy. To pull this off, we need to develop a repeatable benchmark, which involves: 1. **Building** a representative query set.[^37] 2. **Sampling** responses and recording:[^38] - **AI Citations** – direct mentions or source links. - **Share of Answer** ($SOA$) – your slice of citations vs competitors.[^39] - **Organic Rank** – for original and sub-queries.[^40] 3. **Re-sampling** at regular intervals (e.g., quarterly) to monitor changes. Never imagine any fixed hierarchies of content types, platforms, assistants, or tactics. There are no silver bullets; focus on disciplined iteration. > Are you growing $SOA$ and rank across your query-set fast enough to turn declining $CTR$ into increasing revenue? ## Bottom-line: Build for Intent, Not Algorithms AI assistants aren't just another search feature. They're a completely new distribution channel with different rules. While many focus on ranking positions and click-through rates, AI assistant citations can still drive valuable outcomes through higher-converting traffic; the decline in traditional metrics (paid and organic CTR) doesn't necessarily mean failure if you're winning in this new channel. To win citations: (i) build presence in both training data and real-time retrieval; (ii) maintain regular updates to capture recency bias; (iii) answer comprehensive topic clusters and sub-queries; (iv) provide unique data or definitive expert synthesis; (v) tailor content to buyer stage and business model; (vi) optimize technical fetchability and machine readability; and finally, (vii) benchmark Share of Answer ($SOA$) and SERP rank to quantify change—until your brand becomes every assistant's default industry source. Zooming out, corporations have spent billions developing state-of-the-art search systems that understand user intent and context. Strip away the technical details, and you're left with a robust content framework that surfaces the right content at the right moment: (i) evolving with domains; (ii) adapting to platforms and assistants; (iii) aligning with business models; and (iv) serving users precisely—in B2B or B2C contexts, buyer journey stages, and customer avatars. This framework works because it aligns with the fundamental goal of all Information Retrieval (IR) systems: connect users with what they need. Create content that serves intent so thoroughly that any system designed to understand users will naturally surface it. The technical mechanics are just current implementation details. Still don't care? The external → parametric knowledge pipeline means every day you delay, your competitors encode themselves deeper into AI memory. This advantage is compounding daily. Are you looking for a dedicated team that understands how to produce the right content at the right time? Let's talk. **Have questions?** [[email protected]](mailto:[email protected]?subject=AI%20Visibility%20Guide%20Question) [^1]: [Ahrefs (May 19, 2025)](https://ahrefs.com/blog/insights-from-56-million-ai-overviews/) found that AIO now appears in 12.8% or more of all Google searches by volume (skewing toward longer, non-branded informational queries)—nearly 2 billion appearances daily, with frequency increasing monthly and occupying most screen real estate previously held by traditional search results. [^2]: [OpenAI (July 22, 2025)](https://openai.com/global-affairs/new-economic-analysis/) reported that ChatGPT handles ~2.5 billion prompts daily from global users (~330 million from U.S. users). [^3]: [Forrester Buyers’ Journey survey (Nov 2024)](https://www.forrester.com/report/b2b-buyer-adoption-of-generative-ai/RES181769) found ~89–90% of B2B buyers reported using generative AI tools for research in _every phase_ of their purchase process. [^4]: [Ahrefs (April 17, 2025)](https://ahrefs.com/blog/ai-overviews-reduce-clicks/) analyzed 300,000 keywords and found that AIO presence correlated with organic CTR drops of 34.5% for rank-1 pages compared to similar informational keywords without AIO. [^5]: [Seer Interactive (February 4, 2025)](https://www.seerinteractive.com/insights/ctr-aio) found that AIO reduces both organic CTR (dropping from 1.41% to 0.64% year-over-year) and paid CTR. [^6]: [Adobe Digital Insights Quarterly Report (June 2025)](https://business.adobe.com/content/dam/dx/us/en/resources/reports/adobe-digital-insights-quarterly-report/adobe-digital-insights-quarterly-report.pdf) analyzed trillions of visits and billions of transactions, finding that: AI-referred traffic (from ChatGPT, Perplexity, Copilot, Gemini) grew 10–12× from July 2024 to February 2025; engagement metrics improved with 23% lower bounce rates, 12% more page views, and 41% longer sessions versus other traffic; conversion gap narrowed from 43% lower in July 2024 to 9% lower by February 2025; revenue per visit reached parity during 2024 holidays with travel showing 80% higher RPV and banking showing 23% higher application starts from AI-referrals. [^7]: [Semrush (June 9, 2025)](https://www.semrush.com/blog/ai-search-seo-traffic-study/) studied 500+ high-value digital marketing and SEO topics, finding that: 50% of ChatGPT 4o response links point to business/service websites; the average AI-referred visitor (from non-Google sources like ChatGPT) converts ~4.4× more than the average Google Search visitor (via AIO or not); 90% of pages cited by ChatGPT typically rank 21+ in traditional search for related queries. [^8]: [Profound (August 6, 2025)](https://www.tryprofound.com/blog/agents-are-users-why-the-cloudflare-perplexity-fight-misses-the-point) analyzed its platform data and found that ChatGPT-referred users could convert at nearly 10× the rate of traditional organic search traffic. CVR: ChatGPT: 16.3%; Perplexity: 9.5%; Claude: 5%; Google Organic: 1.7% (baseline). [^9]: [Seer Interactive (June 3, 2025)](https://www.seerinteractive.com/insights/case-study-6-learnings-about-how-traffic-from-chatgpt-converts) analyzed GA4 AI traffic from a single Seer client from October 1, 2024 - April 30, 2025 and found ChatGPT had 9× higher conversion rates than traditional organic traffic. CVR: ChatGPT: 15.9%; Perplexity: 10.5%; Claude: 5%; Gemini: 3%; Google Organic: 1.76% (baseline). [^10]: RALM (Retrieval-Augmented Language Models) encompasses various implementations differing in timing (pre‑training, fine‑tuning, inference), supervision (learned vs. fixed retrievers), and conditioning (prompted, fused, or generative). [Ram et al. (August 1, 2023)](https://arxiv.org/abs/2302.00083) formalized the RALM acronym with In-Context RALM. [Hu & Lu (June 29, 2025)](https://arxiv.org/abs/2404.19543) established RALM as the umbrella term spanning [Retrieval Augmented Generation (RAG)](https://en.wikipedia.org/wiki/Retrieval-augmented_generation) and Retrieval Augmented Understanding (RAU). [^11]: [Zhang et al. (October 22, 2023)](https://arxiv.org/abs/2310.14393) demonstrated that pairing AI-generated passages with retrieved sources improves accuracy when knowledge conflicts exist by: generating multiple passages from parametric knowledge; retrieving external sources that may agree or conflict; matching generated and retrieved passages into compatible pairs; and processing matched pairs together using compatibility-maximizing algorithms. [^12]: [Google's U.S. Patent No. US11769017B1 (September 26, 2023)](https://patents.google.com/patent/US11769017B1/en) describes two approaches for AIO citation: Search-First retrieves content based on query relevance then generates responses, ensuring grounding but potentially missing parametric insights; Generate-First creates responses from parametric knowledge then searches for verification sources, leveraging model understanding but requiring post-hoc verification. [^13]: [Wu et al. (February 7, 2025)](https://arxiv.org/abs/2404.10198) found that models with lower confidence in initial responses (measured via token probabilities) more readily adopt retrieved content, while confident models resist contradictory information. [^14]: [Xie et al. (February 27, 2024)](https://arxiv.org/abs/2305.13300) found that RALMs encountering both supportive and contradictory external sources exhibit strong [confirmation bias](https://en.wikipedia.org/wiki/Confirmation_bias) toward parametric knowledge rather than synthesizing conflicting viewpoints. [^15]: [Search Engine Land (May 12, 2025)](https://searchengineland.com/how-to-get-cited-by-ai-seo-insights-from-8000-ai-citations-455284) analyzed ~8,000 citations across 57 queries, finding platform-specific patterns: news steady at 20–27%, blogs varying 21–46%, UGC typically under 4% (often <0.5%); Wikipedia leads ChatGPT (27%), YouTube dominates Gemini, expert review sites prominent for Perplexity (9%), Reddit most-cited for AIO; product/vendor blogs appear across engines (~7% for AIO/Gemini/Perplexity). [^16]: [GoDataFeed (February 14, 2025)](https://www.godatafeed.com/blog/google-ai-overview-prefers-video) analyzed video citations in AIO, finding YouTube citations in 78% of product comparison searches with significant industry variation, and video content 3.1× more likely to be cited than equivalent text content. [^17]: [SparkToro (March 10, 2025)](https://sparktoro.com/blog/new-research-google-search-grew-20-in-2024-receives-373x-more-searches-than-chatgpt/?utm_source=chatgpt.com) analyzed U.S. desktop behavior, finding Google processes ~14B searches daily versus ChatGPT's ~37.5M search-like prompts—a 373× gap. All AI tools combined represent <2% of search volume. [^18]: [Ahrefs (February 6, 2025)](https://ahrefs.com/blog/ai-traffic-study/#) studied 3,000 websites and found ChatGPT drives 50% of AI-referred traffic, potentially delivering disproportionate value. [^19]: [Search Engine Land (May 29, 2025)](https://searchengineland.com/mike-king-smx-advanced-2025-interview-456186) interviewed Michael King, who argued search functions as a branding channel despite industry focus on performance metrics. AI surfaces make this branding function undeniable by exposing brand information without requiring clicks, transforming non-branded searches into branded awareness within search results, creating more qualified traffic. [^20]: [Semrush (February 3, 2025)](https://www.semrush.com/blog/chatgpt-search-insights/) analyzed 80 million clickstream records, finding that: Google showed higher navigational intent while ChatGPT showed more informational intent; SearchGPT-enabled distribution mirrored Google with increased navigational, commercial, and transactional searches; SearchGPT-disabled prompts leaned heavily informational with many falling into "unknown" intent due to longer, detailed nature. [^21]: [Seer Interactive (June 15, 2025)](https://www.seerinteractive.com/insights/study-ai-brand-visibility-and-content-recency/) analyzed 5,000+ cited URLs across ChatGPT, Perplexity, and AIO, finding strong recency bias: 65% of citations from past year (2025), 79% from past two years, 89% from past three years. AIO showed strongest bias (85% from 2023–2025), followed by Perplexity (80%) and ChatGPT (71%). [^22]: Content decay refers to the gradual decline in performance and relevance of online content over time, leading to decreased traffic, lower rankings, and diminished engagement. It's a natural part of the content lifecycle. [^23]: [Google's Developer Search Documentation (June 19, 2025)](https://developers.google.com/search/docs/appearance/ai-features) confirms AIO and AI Mode use "query fan-out" technique (also called "query expansion")—issuing multiple related searches (e.g., intents, related or subtopics, specific named entities, and adjacent needs) to develop responses. Assistants retrieve sites ranking for synthetic sub-queries beyond the original query. [^24]: [Advanced Web Ranking (July 1, 2024)](https://www.advancedwebranking.com/blog/ai-overview-study) analyzed 8,000 keywords across 16 industries, finding that top rankings don't guarantee AIO citations: 33.4% of AIO links ranked top 10 organically while 46.5% ranked outside top 50. [^25]: [Aggarwal et al. (June 28, 2024)](https://arxiv.org/abs/2311.09735) found that adding substance (quotes, statistics, sources) and improving writing quality increases AI citation rates more than stylistic optimization like adding technical terms or unique keywords. [^26]: [Wan et al. (August 9, 2024)](https://arxiv.org/abs/2402.11782) showed that LLMs prioritize relevance over credibility indicators humans value (scientific references, neutral tone, authoritative sources). Substantive text additions may improve AI visibility by increasing information density rather than traditional credibility, providing more semantic relevance hooks. [^27]: Knowledge conflicts describe contradictions between parametric memory and contextual evidence. [Longpre et al. (January 12, 2022)](https://arxiv.org/abs/2109.05052) formalized conflicts as contextual contradictions to learned knowledge. [Xu et al. (June 22, 2024)](https://arxiv.org/abs/2403.08319) systematized taxonomies (context‑memory, inter‑context, intra‑memory) and mitigation guidance. [^28]: [Qian et al. (October 15, 2024)](https://arxiv.org/abs/2310.00935) found that LLMs can identify knowledge conflicts but struggle to pinpoint specific conflicting segments and provide appropriately nuanced responses. [^29]: [Jin et al. (February 22, 2024)](https://arxiv.org/abs/2402.14409) found RALMs follow [bandwagon effect](https://en.wikipedia.org/wiki/Bandwagon_effect)/[majority rule](https://en.wikipedia.org/wiki/Majority_rule) when facing conflicting evidence, trusting evidence appearing more frequently. [^30]: [Search Engine Land (May 12, 2025)](https://searchengineland.com/how-to-get-cited-by-ai-seo-insights-from-8000-ai-citations-455284) revealed AI assistants cite company content 4.25× more in B2B versus B2C queries (17% vs. <4% across platforms): B2C queries favor review sites, tech blogs, Wikipedia, Reddit/Quora with minimal company citations; B2B queries show company sites/blogs at ~17%, plus industry publications and analyst reports; mixed queries show ~70% news and blog citations. [^31]: [OpenAI (April 10, 2025)](https://x.com/OpenAI/status/1910378768172212636) announced that ChatGPT memory now references all past chats to provide personalized responses based on user preferences and interests for writing, advice, learning, and other applications. [^32]: [Google's U.S. Patent No. US11769017B1 (September 26, 2023)](https://patents.google.com/patent/US11769017B1/en) indicates AIO/AI Mode may rerank retrieved content by relevancy to recent queries and update overviews based on user interaction to reflect familiarity with certain sources/content; same patent cited for a different claim. [^33]: E.g., ensure [robots.txt](https://en.wikipedia.org/wiki/Robots.txt) and login or paywall gates don’t hide primary content from crawlers; prioritize [server-side rendering](https://en.wikipedia.org/wiki/Server-side_scripting) or static [HTML](https://en.wikipedia.org/wiki/HTML); keep pages lean; avoid heavy [DOM](https://en.wikipedia.org/wiki/Document_Object_Model) gymnastics; keep [interstitials](https://en.wikipedia.org/wiki/Interstitial_webpage) away from primary content; for complex apps (e.g., a [single-page application](https://en.wikipedia.org/wiki/Single-page_application)), provide fallbacks such as prerendered routes or static [JSON](https://en.wikipedia.org/wiki/JSON) [API endpoints](https://en.wikipedia.org/wiki/Web_API); Hybrid “serve crawlers differently” setups can work (not "[cloaking](https://en.wikipedia.org/wiki/Cloaking)")—just budget for maintenance. [^34]: E.g., write self-sufficient paragraphs; provide structured metadata using [JSON-LD](https://en.wikipedia.org/wiki/JSON-LD) with [schema.org](https://en.wikipedia.org/wiki/Schema.org) types (Organization, Person, Product) that loads without [JavaScript](https://en.wikipedia.org/wiki/JavaScript); include explicit bylines and credentials, publication and last-modified dates, and simple [changelogs](https://en.wikipedia.org/wiki/Changelog); keep schema consistent across pages (a shared `schema.json` helps); use [semantic HTML](https://en.wikipedia.org/wiki/Semantic_HTML) for headings, stable terminology, and consistent internal hyperlinks to canonical entity pages. [^35]: With baseline sessions $S_0$ and incremental net new AI-referred sessions $S_{\mathrm{ai}}$ (gross AI sessions minus cannibalized ones), relative revenue lift is $\Delta R/R_0 = \frac{S_{\mathrm{ai}}}{S_0} \cdot \frac{\mathrm{CVR}_{\mathrm{ai}}}{\mathrm{CVR}_{\mathrm{avg}}}$ where $\mathrm{CVR}_{\mathrm{ai}}/\mathrm{CVR}_{\mathrm{avg}}$ is the conversion rate multiplier; AOV held constant; If cannibalization is unmeasured, treat this as an upper bound. [^36]: This may require implementing UTMs, a referrer allowlist, or server-side tagging in conjunction with [Google Analytics 4 (GA4)](https://en.wikipedia.org/wiki/Google_Analytics). Additionally, as of June 17, 2025, AI Mode traffic is now included in [Google Search Console (GSC)](https://en.wikipedia.org/wiki/Google_Search_Console) performance reports, aggregated with regular search traffic; AIO data is also included, but it's not possible to isolate the performance of citations within GSC. Still, combine GSC and GA4 data for a more comprehensive view of traffic from AIO / AI Mode. [^37]: E.g., compile ~100–200 long-tail prompts across intents, buyer stages, user profiles; use high-volume, industry-specific keywords; use popular questions and FAQs; branded & unbranded; clustered by topic; mimic "query fan-out" by generating ~3-15 diverse sub-queries (e.g., adjacent needs). [^38]: Test different levels (per query, sub-queries, topical clusters, full set), platforms (Google Search, YouTube), and AI assistants (AIO, AI Mode, ChatGPT, Perplexity, Gemini, Claude); using adequate sample sizes to account for variance; measure sentiment in AI-generated answers. [^39]: $\mathrm{SOA}(q) = \frac{\text{your brand citations}}{\text{all brand citations}}$ for query $q$; aggregate (full set) as $\mathrm{SOA}(Q) = \frac{1}{|Q|}\sum_{q \in Q}\mathrm{SOA}(q)$. Weight by depth/placement (title, top answer, footnote) and link type (direct/indirect) and average across assistants (raw or impression-weighted). [^40]: With sub-queries of the original query as set $Q$, measure coverage as % of sub-queries where you rank at any position ($\mathrm{SQR}@\infty$, i.e., $k = \infty$), with finer granularity via $\mathrm{SQR}@k = \frac{|\{q \in Q: r_q \leq k\}|}{|Q|}$ where $r_q$ is your organic rank for sub-query $q$ (set $r_q = \infty$ if unranked). Choose $k$ for preferred ranking threshold (e.g., top 3, top 10).