AI Search Is the Whole Company's Problem, Not Just SEO's

AI Search Is the Whole Company's Problem, Not Just SEO's ![rw-book-cover](https://d34adp677peecb.cloudfront.net/static/images/article3.5c705a01b476.png) 🔗 **Source:** [AI Search Is the Whole Company's Problem, Not Just SEO's](https://www.youtube.com/watch?v=4qvdNixFvCc#transcript) > [!note]+ Document Notes > > [!info] Generated analysis — not the author's words. Sections below are machine-written > > from the captured source. Only text inside `> [!quote]` blocks is verbatim from the author. > > Do not quote or attribute any other line to the named author. > > tags: [ai-search, generative-engine-optimization, seo, content-strategy, brand-coherence] > author: Emma Young (Hostinger) > source: WordCamp US 2026 > reference: https://www.youtube.com/watch?v=4qvdNixFvCc > date created: 2026-08-19 > > # 🏷️ Title: AI Search Is the Whole Company's Problem, Not Just SEO's >  ## 🧭 Table of Contents - [📝 Summary](#-summary) - [📖 Main Content](#-main-content) - [🗒️ Notes](#-notes) - [📚 References](#-references-1) - [🗣️ Paraphrase](#-paraphrase) - [📰 Newsletter](#-newsletter) - [✅ Action Items](#-action-items) - [🧪 Hypotheses](#-hypotheses) - [📊 Methodology](#-methodology) - [❓ Questions](#-questions) - [🔮 Future Research](#-future-research) - [💡 Implications](#-implications) - [🧠 Learning Styles Assessment](#-learning-styles-assessment) - [📊 Accessibility Assessment](#-accessibility-assessment) - [🌍 DEIB Assessment](#-deib-assessment) - [✅ Logical Validity](#-logical-validity) - [🛡️ DARVO Analysis](#-darvo-analysis) - [🎯 Conclusion](#-conclusion) - [🔄 Refactor Appendix](#-refactor-appendix) - [🧰 Template Library](#-template-library) ### 🔗 References - SparkToro/Rand Fishkin zero-click search research (stat cited: 45% no-click in 2016 → 49% in 2019 → 60% in 2024 → 68% now) — URL not given in source - Adobe study on Gen Z/millennial TikTok-as-search-engine usage (65% Gen Z, 55% millennials) — URL not given in source - Lily Ray — AI-citation-share-vs-organic-ranking correlation case study; fake "Google leftover pizza" algorithm-update hoax that Google's AI Overviews repeated as fact within 24 hours — URL not given in source - Ahrefs study (referred to in talk as "foundations... with Aerops," likely garbled ASR for "Ahrefs") — 57-million-citation analysis finding 85% of AI citations originate outside the cited brand's own domain — URL not given in source - Mark Williams Cook — research on UK Members of Parliament using ChatGPT for Hansard speeches, showing convergent AI-generated word-choice spikes across 650 politicians — URL not given in source - WordCamp Europe 2026 — Emma Young's earlier version of this talk, referenced as the baseline her data has since updated against - Google Search Console — noted as now tracking Instagram/TikTok/X/YouTube performance directly ## 📝 Summary - **What happened.** Emma Young (head of organic marketing, Hostinger) argues that AI search (ChatGPT, Perplexity, Claude, Google AI Overviews) has restructured how customers find brands, and that this is now a whole-company problem, not an SEO-team problem. Hostinger kept its Google rankings after AI Overviews launched but watched clicks decline anyway, because Google increasingly answers the question before the click happens. - **Why it matters.** Zero-click search reached 68% (Sparktoro), meaning most searches never reach a website at all. AI answers are retrieval systems built on Google's index — being unranked in Google means being invisible in AI answers too — but even being ranked no longer guarantees traffic, since 85% of AI citations come from outside the brand's own domain (Reddit, press, forums, partners, socials). - **Key stakeholders.** SEO/content teams, brand/PR/comms teams, developers, paid media (PPC) teams, solopreneurs and agencies serving small/regional clients, and — implicitly — the customers whose journeys now span YouTube, Reddit, TikTok, and chatbots before ever reaching a company's own website. - **Risks and tradeoffs.** Brands that let each team (affiliates, landing pages, YouTube, social) tell a slightly different story become "incoherent" in AI outputs — worse, per Young, than being simply unknown, because incoherence is harder to fix than absence. A second risk: using generic AI tools to manufacture "consistent" content produces AI-generated sameness that becomes invisible (cites the Hansard MP speech-pattern convergence study). A third: paid search can mask an organic decline that never appears in weekly reports or board decks until revenue growth also softens. - **Bottom line.** Fix Google-indexability fundamentals first (fast, clean, crawlable pages) because AI cannot make an unindexed site visible. Then pursue brand coherence — the same core truth expressed natively across every platform where the customer already is — over sheer content volume on any single channel. ## 📖 Main Content - Young's core argument is structural, not tactical: AI search products are retrieval systems, not oracles. Every major AI product (ChatGPT, Perplexity, Claude, Gemini, Google AI Overviews) pulls from a search index — for most of them, Google's index — before generating an answer. This means classic SEO fundamentals (crawlability, page speed, clean site structure, actually being indexed) remain a hard prerequisite: if Google hasn't indexed and trusted a page, no AI tool can independently "discover" and cite it. She frames this via Lily Ray's case study showing sites that dropped in organic rankings saw AI citation share drop in the same timeframe, on the same sites, in sync — evidence the two systems are coupled, not independent. - Layered on top of indexability are three "authority signals" that determine whether AI actually picks a brand over a competitor once both are indexable: earned authority (press/media mentions from credible trade publications), borrowed authority (partners, affiliates, and trusted third parties who reference the brand), and community authority (authentic, unprompted conversations on Reddit/forums). None of these are owned by an SEO team alone — which is Young's pivot from "this is an SEO problem" to "this is a whole-company problem." - The talk's central metaphor is a cocktail: when affiliates, landing pages, YouTube, and stray old Reddit threads each tell a slightly different version of the brand story, AI blends them into an incoherent answer — not wrong exactly, but inconsistent. Young argues "incoherent" is a worse state than "unknown," because unknown is fixable through presence-building, while incoherence means a brand is everywhere but says nothing consistent. - She explicitly warns against the obvious fix (using AI to generate uniform content across channels), citing the Mark Williams Cook Hansard research: UK MPs who started using ChatGPT for parliamentary speeches began showing convergent, statistically detectable word-choice patterns across all 650 analyzed politicians — everyone using the same tool/prompts converges on the same "voice," which reads to AI systems (and audiences) as generic, not distinctive. Her proposed alternative is a coordinated-but-native content model: one core intent (e.g., "how to start a website") expressed differently per platform (Google/AI-citation article, YouTube video, Reddit thread) rather than a single style pushed everywhere. - Practical playbook by team: Content teams should front-load direct answers (answer in paragraph one, not four), ensure each sentence stands alone as a citable claim, and prioritize how-to/original-data content over opinion pieces (Hostinger reports a 52% increase in AI citations after applying these three changes to hundreds of pages over three months). She also flags a specific new risk: self-promotional "we are the best" listicles get cited ~70% of the time but the AI response excludes the listicle's author from its own recommendation — effectively free advertising for competitors. - Brand/PR teams should treat AI strategy and PR strategy as the same strategy: pursue earned media (HARO/Muckrack-style outreach), publish original data/research as "citation magnets," and manage what executives and spokespeople post on LinkedIn, since Perplexity specifically skews toward LinkedIn and high-authority publications. - Developers/technical teams should maintain fast, crawlable, semantic HTML and implement `llms.txt` files (unproven but low-cost); she also challenges Google's Core Web Vitals threshold (2.5s LCP) as too lax, citing emerging research that page loads under ~1 second (down toward 100ms) outperform the "passing" threshold, and recommends checking whether AI/vibe-coding tools auto-generate `llms.txt`, sitemap.xml, and robots.txt. - Paid media teams should pull non-branded organic search trends over 18 months; if organic is declining while paid spend holds revenue flat, that gap — often invisible in standard reporting — may be AI-search cannibalization rather than a healthy channel mix. - Solopreneurs/agencies are advised to own one topic completely (a single, exhaustively answered question as a citable asset) rather than spreading thin, and to audit client-side incoherence (copy, socials, Reddit presence) as both a service risk and a billable opportunity. - Young closes with a monthly audit ritual: open every major AI tool, ask 5 recurring customer questions plus a fixed prompt set ("what is [brand]," "[brand] vs [competitor]," "[brand] reviews and complaints," "is [brand] trustworthy"), log whether the brand appears, whether the description is accurate/consistent/confident (vs. hedged), and track it monthly in a spreadsheet — paired with a warning that many AI-visibility-scoring tools are "quietly rigged" because they inflate scores when the brand name is already seeded in the prompt. ### 🗒️ Notes // Source is a live conference-talk transcript without embedded timestamps (the transcript tool used returned plain text, not timecoded segments), so per the template's non-timestamp fallback, notes are organized by topic segment rather than timecode. - **Cocktail self-introduction / framing device.** Young introduces herself as a "cocktail" (SEO/content as base spirit, YouTube/localization/socials as chaos, strong AI opinions as garnish) — sets up the cocktail metaphor she reuses for "incoherent brand = bad cocktail" and "coherent brand = house cocktail (be your margarita)" later in the talk. - **Definitional framing of "AI search."** Explicitly scoped to ChatGPT, Perplexity, Claude, and Google AI Overviews — "places where you ask a question and get an answer instead of 10 blue links." She flags that her slide data has been updated six times in two months since a similar WordCamp Europe talk, and explicitly tells the audience hypotheses may be debunked or validated within 2–6 months — an unusually strong volatility disclaimer for a conference talk. - **Hostinger's baseline problem.** Years of "proper" SEO (thousands of pages, dozens of markets, millions of visits) kept rankings intact after AI Overviews launched, but clicks still declined — rankings staying flat while clicks fall is presented as the tell that the click-through funnel itself, not ranking quality, is the thing that broke. - **"AI telling our story without us in the room."** Personal/emotional framing: AI answers synthesize what's findable across the web about a brand, sometimes inaccurately or over-confidently, without the brand's own input — this reframes AI visibility as a reputation-management problem, not just a traffic problem. - **Customer-journey walk-through.** A five-step journey is laid out for a hypothetical Hostinger customer: (1) YouTube ("how do I make money with a website" — awareness), (2) Reddit ("best website builder," seeking unbranded opinion), (3) TikTok/socials (wants to see it done), (4) a chatbot (comparison/narrowing), (5) Google ("hosting prices 2026" — high purchase intent). Her point: optimizing only for step 5 makes a brand invisible for steps 1–4, where most of the actual decision-shaping happens. - **Zero-click statistics (Sparktoro).** 45% (2016) → 49% (2019) → 60% (2024) → 68% (now) of Google searches end without any click — presented as an accelerating trend, not a plateauing one. - **TikTok-as-search-engine statistics (Adobe).** 65% of Gen Z and 55% of millennials use TikTok as a primary/informational search tool — used to argue this isn't a "wait it out, it's just Gen Z" problem since millennials (the room's demographic, per Young) already show majority usage. - **ChatGPT scale statistic.** 2.5 billion prompts/day cited as evidence ChatGPT is "infrastructure," not a novelty; she separately notes that Google's AI Overviews alone reportedly reach more people than ChatGPT + Gemini + Claude + Perplexity combined. - **Retrieval, not magic.** Explicit mechanical explanation: AI products retrieve from a search index (mostly Google's) before generating an answer; a site absent from that index cannot be surfaced by AI regardless of AI-specific optimization — "there's no side door." - **Lily Ray citation-correlation study.** Sites that dropped in Google rankings saw AI citation share drop on the same timeline — used as the evidentiary anchor for "fix Google fundamentals first." - **Three authority signals.** Earned (press/trade publications), borrowed (partners/affiliates/experts), community (Reddit/forums) — explicitly framed as none being solely SEO's responsibility. - **85% external-citation statistic.** An Ahrefs-style study (name garbled in the ASR transcript as "foundations... with Aerops") analyzing 57 million citations found 85% of AI citations point outside the brand's own domain — meaning an SEO team optimizing only owned properties is addressing roughly 15% of the actual AI-visibility surface. - **Bad cocktail metaphor.** Different teams (affiliates, landing pages, YouTube, an old undiscovered Reddit thread) each contribute an uncoordinated "ingredient" that AI blends into an incoherent answer; incoherence is framed as worse than obscurity because it can't be fixed by more presence alone. - **Lily Ray's fake pizza hoax.** She fabricated a fictitious Google algorithm update (including a joke about Google approving it "between slices of leftover pizza"); Google's AI Overviews were repeating it as fact within 24 hours — used as proof that AI systems optimize for repetition/consensus signals, not truth-verification, and that this is trivially exploitable by anyone with sufficient reach. - **"Fixing incoherence with AI" is a trap.** If every brand uses the same generative tools/prompts/training data to manufacture cross-channel consistency, the result is homogenized, not distinctive, output. - **Mark Williams Cook / Hansard MP study.** UK MPs using ChatGPT for parliamentary speeches showed detectable convergent word-choice spikes (words like "underscores," "navigating") across 650 analyzed politicians — real evidence of AI-generated homogenization at a large, verifiable public-record scale. - **Before/after org model.** "Before": SEO/content/YouTube teams optimize independently for their own channel KPIs with no cross-team coordination. "After": one core customer intent expressed natively per platform (Google/AI-citation article, YouTube video with shorts, Reddit thread) — framed as an alignment strategy, not a new content strategy, and explicitly acknowledged as "simple in theory, hard in practice." - **Content-team tactical wins.** Answer-first paragraph structure; each sentence must stand alone as a citable unit; prioritize how-to/original-data content over opinion. Hostinger's internal test: 52% increase in AI citations after applying these three tactics to "hundreds of pages" over three months. - **Self-promotional listicle risk.** "We are the best" listicles get cited ~70% of the time by AI tools but the citing AI's actual recommendation excludes the listicle's own brand — i.e., brands are footnoting their competitors' visibility for free. - **Brand/PR-team tactics.** AI strategy = PR strategy now; earned media from one high-authority publication reportedly outweighs 100 mentions on low-authority sites; recommends HARO/Muckrack-style expert-quote outreach and original research/data studies as "citation magnets"; notes Perplexity specifically over-indexes on LinkedIn and high-authority publications, making exec/spokesperson LinkedIn presence a lever. - **Internal alignment exercise.** A quick internal test: ask people from different teams to independently finish the sentence "your brand is ___" without conferring, then compare — if internal teams can't converge on a description, AI won't either. - **Dev/technical tactics.** Structured data and fast, semantic, crawlable HTML still matter; `llms.txt` files are described as an unproven but low-cost bet ("we shall see if that works"); Core Web Vitals' 2.5s LCP threshold is characterized as too lenient given emerging data that sub-1-second (even ~100ms) load times outperform "passing" scores; vibe-coding/AI site builders should be checked for auto-generation of `llms.txt`, sitemap.xml, and robots.txt. - **Paid-media audit tactic.** Pull 18 months of non-branded organic trend data; a decline masked by flat/growing paid-driven revenue is framed as a hidden AI-search erosion risk that won't show up in standard weekly/board reporting until customer growth itself softens. - **AI-visibility-score caveat.** Many third-party AI-visibility scoring tools are described as "quietly rigged" — scores can rise simply because the brand name was seeded into the prompt, rather than reflecting genuine unprompted citation share; she recommends tracking who actually gets cited across a category instead of trusting a single up/down score. - **Solopreneur/agency tactics.** Own one topic completely as a citable asset rather than spreading broad; audit client brand coherence (copy, socials, Reddit) as both risk exposure and billable service; treat personal brand (LinkedIn, Reddit answers, podcast/YouTube appearances) as a genuine authority signal, not a side activity. - **Agave case study (coherent-brand example).** A margarita bar in Kaunas, Lithuania, that Young holds up as "be your own margarita": every channel (Google Maps, Instagram, word of mouth) reinforces the identical claim ("best margaritas in Kaunas") with no hedging — presented as the positive mirror of the "bad cocktail" incoherence metaphor. - **Monthly audit homework.** Ask 5 recurring customer questions plus 4 fixed prompts ("what is [brand]," "[brand] vs. biggest competitor," "[brand] reviews and complaints," "is [brand] trustworthy") across every major AI tool monthly; log accuracy, consistency, and confidence (vs. hedging) in a spreadsheet; notes Google Search Console now tracks Instagram/TikTok/X/YouTube performance directly, encouraging account verification there. - **Closing "ladder vs. map" metaphor.** 20 years of SEO optimized for a single ladder (one platform, one algorithm, climb-and-be-found); customers now move on a nonlinear, multi-platform "map" of questions asked in different voices, in different places, at different times — closing with three retained takeaways: (1) not indexed by Google = doesn't exist in AI answers, (2) coherence across platforms beats volume on one platform, (3) be your category's "house cocktail" — one consistent answer everyone agrees on. - **Q&A — Reddit reputation management.** Hostinger avoids "takeover" tactics on Reddit; relies on genuinely useful content organically getting shared, monitors brand mentions, runs its own subreddit for direct answers, and draws the line at "when you're turning more into sales than helpful/educational." - **Q&A — metrics/reporting philosophy.** Advises against tracking many vanity metrics; pick 1–2 metrics tied to an actual tested hypothesis/experiment rather than defaulting to broad traffic/ranking dashboards. - **Q&A — what changed since her WordCamp Europe talk (a few months earlier).** The zero-click rate moved from ~60% to 68% in just a few months; compare-pages are now outperforming listicles (reversing a listicle boom from ~6 months prior); the Lily Ray pizza hoax happened in the intervening months, following a similar earlier hoax by another creator (an eating-record claim) — both cited as evidence of how quickly AI Overviews can be manipulated with enough audience reach. - **Q&A — social media's role for small/regional businesses.** No universal allocation; effort should follow where a specific audience/customer base already spends time (e.g., a technical B2B audience may warrant Substack over TikTok) rather than chasing every platform uniformly. ## 📚 References - SparkToro zero-click search data (Rand Fishkin) — cited but not linked in source; searchable at sparktoro.com - Adobe TikTok-as-search-engine study — cited but not linked in source - Lily Ray — SEO/AI-citation researcher; case study and "fake Google pizza update" hoax — cited but not linked in source - Ahrefs 57-million-citation study on AI citation domain distribution — cited but not linked in source (name garbled in transcript audio) - Mark Williams Cook — Hansard/UK-MP ChatGPT speech-pattern research — cited but not linked in source - Google Search Console documentation on social-platform performance tracking — not linked in source ## 🗣️ Paraphrase - Young opens by reframing herself as a "cocktail" of SEO/content expertise, chaos (YouTube, localization, socials), and strong AI opinions, then states her thesis: AI search is a whole-company problem, not an SEO problem, and her data has shifted meaningfully even since a similar talk months earlier. - She recounts Hostinger's experience: strong SEO fundamentals kept rankings stable after AI Overviews launched, but clicks fell anyway, and AI tools began describing Hostinger inaccurately without any company input — a loss of narrative control layered on top of a traffic loss. - She walks through a five-stage customer journey (YouTube → Reddit → TikTok → chatbot → Google) to argue that Google is now the last, not the only, touchpoint, and that optimizing only for the final step leaves a brand invisible everywhere earlier in the journey. - She grounds the shift in data: 68% zero-click search, majority TikTok-as-search usage among Gen Z and millennials, and ChatGPT's 2.5-billion-prompt daily scale, alongside Google AI Overviews' outsized reach relative to all competing chat tools combined. - She demystifies AI search as retrieval, not generation from nothing: AI tools pull from Google's index before answering, so an unindexed page cannot be surfaced by AI either — Google fundamentals remain a hard floor, evidenced by Lily Ray's finding that ranking drops and AI-citation-share drops move together. - She introduces three non-SEO-owned authority signals (earned/press, borrowed/partners, community/Reddit) and the 85%-external-citation statistic to argue that most of a brand's AI-visibility surface sits outside the SEO team's control. - Using the "bad cocktail" metaphor, she describes how uncoordinated teams (affiliates, landing pages, YouTube, stray Reddit threads) produce an AI answer that's inconsistent rather than wrong — and argues incoherence is a worse, harder-to-fix state than simple obscurity. - She warns against the instinctive fix (using generic AI tools to manufacture consistency), citing the Hansard MP study as proof that identical tools/prompts produce convergent, generic-sounding output across totally unrelated speakers. - She proposes an alternative operating model: one customer intent, expressed natively (not identically) across each platform where that intent already surfaces, coordinated as an alignment strategy across teams rather than owned by one team. - She delivers a team-by-team tactical playbook (content, brand/PR, dev, paid media, solopreneurs/agencies), including Hostinger's measured 52% AI-citation increase from three specific content changes, and warns against self-promotional listicles that get cited but exclude their own author from the recommendation. - She closes with a monthly self-audit ritual (fixed prompts run across major AI tools, logged for accuracy/consistency/confidence), a warning that many AI-visibility scoring tools are "rigged" by prompt-seeding, and a final ladder-vs-map metaphor: 20 years of single-ladder SEO optimization is being replaced by a nonlinear, multi-platform map that the whole company — not just SEO — has to be present on. - In Q&A, she adds operational color: Hostinger avoids Reddit "takeover" tactics and lets useful content get organically shared; she recommends tracking 1–2 hypothesis-linked metrics over many vanity metrics; she notes the zero-click rate moved 8 points in just a few months since her prior talk and that compare-pages now beat listicles; and she frames social-platform investment as audience-dependent rather than universal. ## 📰 Newsletter Heads up, WordPress folks: your SEO checklist just got a whole-company subplot. At WordCamp US 2026, Hostinger's Emma Young laid out why AI Overviews, ChatGPT, and friends are eating your click-throughs even when your rankings look fine — 68% of searches now end with zero clicks, and 85% of what AI tools cite about your brand lives somewhere you don't control (Reddit, press, partners, socials). Her fix isn't "do more AI SEO" — it's cross-team coherence: same core truth, told natively on every channel your customer already trusts, instead of five departments quietly telling five slightly different stories. Practical starters for your next sprint: answer-first paragraphs (not paragraph four), one clear citable claim per sentence, `llms.txt` on your checklist next to sitemap.xml and robots.txt, and a monthly "ask ChatGPT/Perplexity/Gemini what they think of us" audit. Bonus WordPress angle: if you're shipping sites with AI page-builders, check whether they're auto-generating those `llms.txt`/sitemap/robots files — Young flagged that some tools quietly skip it. Full talk (and the margarita metaphor that makes it click) linked above. ## ✅ Action Items #### DevRel - Developer advocacy: build a short internal talk/deck translating "AI search is retrieval, not magic" for engineering teams who assume SEO is someone else's job. - Education: run a workshop on writing answer-first, single-claim-per-sentence documentation and blog content. - Developer experience: audit whether project docs/quickstarts read as citable, self-contained answers an AI could quote in isolation. - Developer marketing: track which docs/blog pages get cited by AI tools vs. simply ranked, as a new content-funnel metric. - Community management: monitor Reddit/forum mentions of the project without deploying "takeover" tactics — mirror Hostinger's stated approach. - Internal/external/cross-project: set up a recurring cross-team sync (content, DevRel, marketing, support) so brand/product descriptions stay consistent across channels. #### OSPO - Policy & Governance: none identified directly in source — talk does not address open source governance. - License Compliance: not applicable to this source — no licensing content discussed. - Security Management: not applicable to this source — no security-disclosure content discussed. - Software Inventory: not applicable to this source — no SBOM/inventory content discussed. - Developer Support: consider whether support docs are AI-citable (answer-first, atomic claims) as a support-deflection strategy. - Community Engagement: not applicable to this source in an OSPO-specific sense — general community-authority points are covered under DevRel above. - Strategic Alignment: treat AI-citation share as a new OKR category alongside traditional web traffic/adoption metrics. #### Community - Maintainers: audit whether project README/getting-started content front-loads the answer before context, per the talk's content-team tactic. - Contributors: not directly addressed in source — no contributor-onboarding content discussed. - Sponsors: not applicable to this source — no sponsorship content discussed. - Ecosystem: consider a shared, cross-project "AI visibility audit" ritual (monthly prompt set run across ChatGPT/Perplexity/Gemini) adapted from Young's homework exercise. #### WordPress - Core: not directly addressed in source. - Core Performance: revisit whether WordPress's own Core Web Vitals guidance should flag Young's claim that sub-1-second load times outperform the standard 2.5s LCP threshold — worth independent verification before adoption. - Core AI: evaluate whether Core's own AI-related features/guidance should reference `llms.txt` support, given her note that "some [tools] do and some don't" auto-generate it. - Core Program Team: consider whether this talk's cross-team-alignment framing is worth surfacing to Marketing, Training, and Documentation teams jointly, not siloed. - Mobile: not applicable to this source. - Playground: not applicable to this source. - Community: consider a WordCamp session/track specifically on AI-search coherence across community-run channels (WP.org, Learn, Meta, Make blogs). - Training: build a Learn WordPress module on "writing for AI citation" (answer-first structure, atomic sentences, how-to/data content). - Marketing: audit whether WordPress.org and Marketing-team messaging is consistent across channels (Make/Marketing blog, socials, WordCamp talks) per the talk's "brand is ___" internal-alignment exercise. - Support: evaluate forum/support content for AI-citability as a deflection strategy, mirroring the DevRel support item above. - Design: not directly addressed in source. - Documentation: prioritize answer-first restructuring for DevHub/HelpHub high-traffic pages, per the talk's specific content tactic that produced a 52% AI-citation increase for Hostinger. - Photos: not applicable to this source. - TV: consider whether WordCamp/WordPress.tv session pages are structured as citable answers (title, description, transcript availability) for AI indexing. - Polyglots: consider whether translated content maintains the same "one truth, native expression" coherence the talk recommends, rather than diverging per locale team. - Accessibility: not directly addressed in source beyond general content-clarity implications (see Accessibility Assessment section below). - Meta: not applicable to this source. - CLI: not applicable to this source. - Hosting: highly relevant to this source's audience — hosting providers (like Hostinger itself) should treat this talk as a customer-education opportunity (blog posts, docs) since hosting-adjacent content ranks heavily in the talk's own customer-journey example. - Tide: not applicable to this source. - Openverse: not applicable to this source. - Themes: not directly addressed in source. - Plugins: not directly addressed in source. #### The WPOCC - Programs: consider a microgrant or fellowship track for contributors researching AI-search/GEO impact on the WordPress ecosystem specifically. - Projects: a documentation sprint applying "answer-first, one-claim-per-sentence" structure to key WPOCC-adjacent resources could be a concrete, scoped project. - Funding experiments: not directly addressed in source. ### FAIR - Technical independence: not directly addressed in source — talk does not discuss federated/decentralized infrastructure. - Policies and covenants: not applicable to this source. - Productized governance artifacts: not applicable to this source. ## 🧪 Hypotheses - H1: Pages restructured with answer-first paragraphs and single-claim sentences will show a measurable increase in AI-tool citation rate within 90 days, replicating Hostinger's reported 52% figure. - H2: Brands with self-promotional "best X" listicles will be cited by AI tools at a materially higher rate than brands without such listicles, but will be excluded from the AI's own top recommendation at a similarly high rate (~70%, per the talk). - H3: A brand's AI-citation share correlates positively with its Google organic ranking position, and the two move together when either changes (per the Lily Ray case study referenced in-talk). - H4: Content generated with the same AI tools/prompts as competitors will show measurably lower linguistic distinctiveness (e.g., vocabulary diversity, sentence-structure variance) than human-authored or differentiated content, consistent with the Hansard MP convergence finding. - H5: Organizations that run a monthly cross-platform AI-citation audit (the talk's homework exercise) will detect organic-traffic-masking-paid-spend gaps earlier than organizations relying solely on standard weekly/board reporting. - H6: Third-party "AI visibility score" tools that seed the brand name into their own prompts will report systematically higher visibility scores than tools/methods that use unprompted, category-level queries. ### 📊 Methodology - H1 could be tested via a controlled before/after content audit: select a matched set of pages, restructure a treatment group per the three tactics, leave a control group unchanged, and measure AI-tool citation rate (via manual prompt sampling or an AI-visibility API) over a fixed window. - H2 could be tested by sampling a category of "best X" listicles across multiple brands, prompting AI tools with the underlying question, and coding whether the listicle's own brand appears in the citation vs. the final recommendation. - H3 requires a longitudinal dataset pairing Google Search Console ranking data with AI-citation tracking (via a visibility tool or manual sampling) across a shared timeline, looking for correlated inflection points as Lily Ray's case study reportedly found. - H4 could use standard computational-linguistics measures (type-token ratio, perplexity against a baseline corpus, stylometric distance) comparing AI-tool-generated brand content across multiple companies using the same underlying model/prompt template. - H5 would need paired case studies: organizations with and without a monthly AI-citation audit ritual, tracked for time-to-detection of an organic/paid divergence. - H6 could be tested by running identical category-level queries through several AI-visibility tools with and without brand-name seeding in the prompt, comparing reported scores. ### ❓ Questions - If AI tools increasingly retrieve from Google's index specifically, what happens to AI-search visibility for brands or content types Google has historically under-indexed or de-prioritized (e.g., certain non-English markets, older content)? - Is the 85%-external-citation statistic stable across industries, or does it vary significantly for categories with less third-party discussion (e.g., regulated industries, B2B niche software)? - How durable is the "compare pages now beat listicles" finding Young cited in Q&A — is this a lasting shift or a temporary reaction to a listicle glut, likely to reverse again? - What happens to the "coherence beats volume" strategy at brands large enough that true cross-team alignment (the "before/after" org model) is organizationally very costly to execute — is there a company-size threshold where this becomes impractical? - How should the WordPress project itself, as a large decentralized multi-stakeholder ecosystem (agencies, hosts, core contributors, plugin authors), pursue "one truth, many native expressions" when there is no single center of narrative control? - Given Lily Ray's fake-pizza-hoax finding, what governance or verification mechanisms (if any) are AI search providers building to prevent low-cost reputational manipulation of AI Overviews at scale? ### 🔮 Future Research - Independent replication of the sub-1-second-vs-2.5-second Core Web Vitals claim, ideally with a controlled dataset, since this directly contradicts Google's own published LCP guidance and warrants scrutiny before broad adoption. - A systematic study of "AI-visibility score" tool methodology and prompt-seeding bias, given Young's claim that many are "quietly rigged." - Sector-specific replication of the 85%-external-citation and three-authority-signal framework outside e-commerce/hosting, e.g., for open source projects, nonprofits, or B2B SaaS. - Longitudinal tracking of whether `llms.txt` adoption produces any measurable citation benefit, given Young's own "we shall see if that works" hedge. - Research into detection methods for AI-generated content homogenization at scale, extending the Hansard MP study's method to commercial marketing content. ### 💡 Implications - **Practice.** Marketing, content, DevRel, and support teams need a shared cross-functional review cadence for brand-description consistency — this is a process/org-design change, not just a content-style change. - **Policy.** As AI Overviews demonstrably repeat unverified claims as fact within 24 hours (the pizza hoax), this strengthens the case for platform-level provenance/verification requirements on AI-generated answers, though the talk does not propose specific policy mechanisms. - **Security.** The pizza-hoax finding is effectively a proof-of-concept for reputational manipulation via AI Overviews — organizations should treat AI-citation monitoring as part of brand/reputation risk management, not just marketing. - **Education.** Teams (technical and non-technical) need basic literacy in "AI search is retrieval, not generation-from-nothing" to avoid magical thinking about AI-specific optimization tactics that ignore indexability fundamentals. - **Funding models.** The talk implies AI-citation share should become a tracked, budgeted KPI alongside traditional SEO/paid metrics — currently, per Young, it's often invisible in weekly reports and board decks, meaning it's under-resourced relative to its actual business impact. ### 🧠 Learning Styles Assessment - Reading level estimate: this transcript, converted to prose, scores roughly at a Flesch-Kincaid grade 8–9 level (conversational spoken register, short-to-medium sentences, occasional filler/false starts typical of live speech, moderate use of marketing/SEO jargon defined in-line). #### 👀 Visual learners - Present: the talk clearly references slides, diagrams (before/after team-silo maps), and data charts (SparkToro trend line) that are not captured in this transcript-only source. - Improve: pair this transcript with the actual slide deck or screenshots of the SparkToro chart and before/after org diagrams for a complete visual reference. #### 🔊 Aural learners - Present: the source is itself audio/video — a live conference talk, well suited to aural learners as-is. - Improve: nothing needed; the original video (linked in the doc URL) directly serves this learning style. #### 🗣️ Verbal learners - Present: strong use of extended analogy (cocktail, ladder-vs-map, agave/margarita) and concrete examples (Hostinger's own metrics, the pizza hoax) throughout. - Improve: a glossary of the talk's specific terms (AI Overviews, zero-click search, GEO, llms.txt, authority signals) would help readers unfamiliar with SEO jargon. #### 🧩 Kinesthetic learners - Present: the talk includes a concrete, repeatable exercise (the monthly 5-question + 4-fixed-prompt AI audit) and a quick internal team exercise ("finish the sentence: your brand is ___"). - Improve: a downloadable checklist/template version of the monthly audit exercise would make it more immediately actionable. #### 🧠 Logical learners - Present: the talk is structured as a clear causal chain (indexability → authority signals → citation → coherence) with team-by-team tactical breakdowns. - Improve: a summary table mapping each named tactic to its owning team and expected metric would sharpen the logical structure further. #### 🧑‍🤝‍🧑 Social learners - Present: includes a live Q&A section with audience questions and Young's direct responses, modeling real-world application and discussion. - Improve: a discussion-guide version with the Q&A questions reframed as prompts for a team workshop. #### 🧘‍♀️ Solitary learners - Present: the closing "three things to remember" and the personal audit homework support individual, self-paced reflection. - Improve: a self-assessment worksheet walking through the three authority signals for one's own brand/project would extend this. #### 🌿 Naturalistic learners - Present: not directly addressed — the talk does not use sustainability/ecosystem framing beyond the loose "ecosystem of channels" metaphor. - Improve: not applicable to this source's subject matter; no natural extension without forcing an unrelated analogy. ### 📊 Accessibility Assessment #### Screen reader semantics - Present: this note applies clear heading hierarchy (H1/H2/H3) to the transcript content; the original video's own captions/accessibility are not verifiable from this transcript alone. - Improve: confirm the source YouTube video has accurate closed captions (not just auto-generated) and a text transcript published alongside it — this note's transcript itself doubles as one option. #### Keyboard navigation - Not applicable to this source — this is a static video/transcript document with no interactive UI to navigate. #### Contrast and readability - Not applicable to this source in Reader — visual contrast is controlled by the Readwise Reader app/theme, not this document's content. #### Text resizing and reflow - Not applicable to this source — handled by the Reader app's rendering, not the document content itself. #### Media and captions - Present: this document provides a full text transcript, which serves as an accessibility fallback for the source video. - Improve: verify the original YouTube video has manually reviewed (not purely auto-generated) captions, since the transcript pulled here shows signs of automatic-speech-recognition artifacts (e.g., "Chatubt" for ChatGPT, "hanser" for Hansard, "argrroni" for Negroni, "apperal spritz" for Aperol Spritz) that a manually corrected caption track would avoid. #### Forms and interactions - Not applicable to this source — no forms or interactive elements are present in a conference-talk transcript. #### Standards reference - Primary applicable WCAG 2.2 AA criterion: 1.2.2 Captions (Prerecorded) and 1.2.3/1.2.5 Audio Description or Media Alternative — satisfied at the note level by this transcript; the source video's own caption quality should be independently verified against the same criterion. ### 🌍 DEIB Assessment #### Cultural breadth - Present: the talk references a genuinely international example (a margarita bar in Kaunas, Lithuania) and Hostinger's stated multi-market SEO operation ("dozens of markets"), giving it more global grounding than a purely US-centric talk. - Improve: could extend with explicit discussion of how AI-search behavior (e.g., zero-click rates, dominant AI tools) varies by region/language, since the cited statistics (SparkToro, Adobe, ChatGPT prompt volume) are not broken out by market. #### Inclusive language - Present: the talk uses plain, informal, jargon-explained language ("AI search, I mean like ChatGPT, Perplexity, Claude...") accessible to a mixed-expertise conference audience. - Improve: a few terms (e.g., "GEO," "authority signals") are used without an explicit one-line definition on first use in the raw transcript; a glossary would help non-specialist readers. #### Neurodiversity support - Present: the talk is organized around a small number of memorable, repeated structural devices (the cocktail metaphor, the three-things-to-remember close) that aid retention. - Improve: the live transcript includes tangents, false starts, and self-corrections typical of unscripted speech; this note's Main Content and Notes sections already restructure the content into consistent, predictable chunks to reduce cognitive load versus the raw transcript. #### Representation - Present: the speaker centers her own team's (Hostinger's) direct experience and data rather than only citing third-party experts, and cites a named female researcher (Lily Ray) prominently as a primary evidentiary source. - Improve: the cited external researchers/sources named in the talk (Lily Ray, Mark Williams Cook) represent a narrow slice of the SEO/AI-research community; broader sourcing across regions and backgrounds would strengthen future versions of this talk. #### Psychological safety - Present: the talk explicitly reframes "incoherence" as an organizational/process failure rather than any single team's fault ("nobody really planned for the things to change"), which is a psychologically safer framing than assigning blame. - Improve: none identified as needed — the talk's framing already avoids blame-oriented language toward specific teams or individuals. ### ✅ Logical Validity #### Strong reasoning - Strong causal/correlational reasoning: the Lily Ray finding (ranking drops and AI-citation-share drops moving together on the same sites/timelines) is used appropriately as correlational evidence for a "Google indexability is a prerequisite for AI citation" claim, not overstated as proof of a specific causal mechanism. - Strong empirical grounding: the Hostinger 52%-AI-citation-increase claim is presented as an internal test result tied to three specific, named changes (answer-first structure, atomic sentences, how-to/data prioritization), which is more falsifiable and specific than a vague "content quality improved" claim. - Strong use of a natural experiment: the Hansard MP study (650 politicians, all independently adopting the same tool) is a genuinely strong quasi-experimental design for demonstrating AI-generated homogenization, since it removes company-specific confounds. #### Fallacies - **Anecdotal/small-N generalization** — location: the Agave margarita bar example. A single small business in one city is used to illustrate "coherence," which is a vivid but statistically ungeneralizable case; the underlying claim (coherent cross-channel presence correlates with being the default AI answer) is plausible but not established by one anecdote. Fix: pair the anecdote with the earlier quantitative findings (85% external citation, Lily Ray correlation) explicitly, rather than letting the anecdote stand alone as the closing proof point. - **Appeal to authority without full sourcing** — location: throughout, statistics are attributed to named sources (SparkToro, Adobe, Ahrefs/"foundations... with Aerops," Lily Ray, Mark Williams Cook) but no URLs, publication dates, or methodology details are given in the spoken talk. Fix: a companion slide deck or blog post with citations and methodology would let the audience independently verify each figure. - **False dilemma (mild)** — location: the "ladder vs. map" closing metaphor implies the old SEO model was a single ladder and the new model must be a full map; in practice, most brands operate somewhere between these poles, and the binary framing, while rhetorically effective, slightly overstates the discontinuity. Fix: acknowledge explicitly that fundamentals (the "ladder") remain necessary but no longer sufficient, rather than framing it as ladder-replaced-by-map. #### Tone of voice - Confident, self-deprecating, and conversational — the speaker uses humor (the cocktail bit, "secrets, don't tell anybody else," reacting to her own technical difficulties) to soften what is otherwise a data-dense, somewhat alarming message (declining organic traffic, AI misrepresentation risk). The tone stays consistently non-defensive about her own company's SEO track record even while describing its traffic decline, which supports credibility. ### 🛡️ DARVO Analysis Not applicable to this source — this is an informational conference talk about search/marketing strategy, not a response to an accusation or conflict narrative. No deny/attack/reverse-victim-offender pattern is present; the speaker does not frame herself, Hostinger, or any party as wrongly accused. The closest adjacent content is her description of AI tools "getting the story wrong" about Hostinger, but this is presented as a factual/technical limitation of retrieval systems, not as a denial of wrongdoing or an attack on an accuser. ## 🎯 Conclusion - **Bottom line.** AI search hasn't replaced SEO fundamentals — it has made them a floor, not a ceiling, while shifting real competitive advantage to cross-channel brand coherence that no single team (least of all SEO alone) can produce on its own. Being ranked in Google is necessary but no longer sufficient; the deciding factor is whether a brand tells one consistent, citable story across the ~85% of touchpoints (press, partners, Reddit, socials) that sit outside its own domain. - **Practical next steps.** (1) Confirm baseline indexability/technical SEO health before investing in AI-specific tactics. (2) Run the monthly cross-tool AI-citation audit (5 customer questions + 4 fixed brand prompts) to establish a visibility baseline. (3) Restructure highest-traffic content for answer-first, single-claim-per-sentence citability. (4) Convene a cross-team (content/PR/dev/paid) alignment session using the "finish the sentence: your brand is ___" exercise to surface existing incoherence before trying to fix it with more content volume. ## 🔄 Refactor Appendix — AI Search Is the Whole Company's Problem, Not Just SEO's > Density: 2–5 Literature atoms (H3), 0–3 Permanent candidates (H4). > Defaults: 3 Literature, 1 Permanent. Resonance filter: if you can't > articulate in one line why a claim matters, drop it. > > Konik claim-header rules: present tense, declarative, ≤10 words when > possible, searchable, unique, mirrors annotation focus if present. > > Atomicity: one declarative sentence per atom. "and"/"also" → split. > > Tag pool: #bmf · #addendum · #xref · #articleseed · #storystem · #share · #aphorisms ### 📚 Literature Atoms (H3) #### [H1] AI search retrieves from Google's index, not magic > [!quote] ID:rw_pending > "it's not magic... it's retrieval... for most of those AIs, that library is Google's." **Context & use**: - Mid-talk, explaining the mechanics of AI search products. - Why this matters: reframes AI-specific optimization as downstream of, not a replacement for, standard SEO indexability work. **Tags**: #xref #aphorisms **Highlight**: [view highlight](readwise://pending)  #### [H2] Incoherent brand presence beats unknown, but is harder to fix > [!quote] ID:rw_pending > "you're not unknown, but you are incoherent... unknown is fixable... but incoherent means you're everywhere and you can still say nothing." **Context & use**: - Bad-cocktail metaphor segment, describing uncoordinated team output. - Why this matters: directly actionable diagnostic for auditing a brand's own AI-citation coherence. **Tags**: #xref #articleseed **Highlight**: [view highlight](readwise://pending)  #### [H3] 85% of AI citations originate outside the brand's own domain > [!quote] ID:rw_pending > "85% of AI citations are coming from outside your brand's own domain." **Context & use**: - Cited from an Ahrefs-style 57-million-citation study. - Why this matters: quantifies why SEO teams alone cannot own AI-search visibility — most of the surface area sits with PR, partners, and community. **Tags**: #xref #addendum **Highlight**: [view highlight](readwise://pending)  ### 🧠 Permanent Candidates (H4) #### [P1] Cross-channel coherence outranks single-channel volume in AI search **Claim** (one sentence, Courtney's voice): When multiple channels repeat the same core claim about a brand or project, AI search tools converge on that claim as the default answer, regardless of how much content any single channel produced. **Evidence pointer**: [[#H2]] [[#H3]] **Scope & conditions**: Holds when a brand/project has presence across at least a few independent channel types (owned, earned, community); may not hold for brand-new entities with no third-party footprint yet, where the priority is first establishing any external presence at all. **Links** (≥1 with annotation): - [[AI search is retrieval, not magic]] — because extends (coherence only matters once the underlying pages are indexable) - [[Content strategy notes]] — because applies-in cross-team content planning  ### ✏️ Fleeting Queue - [F1] Independently verify Young's claim that sub-1-second (vs. Google's 2.5s) LCP outperforms the standard Core Web Vitals threshold before citing it as WordPress performance guidance — https://www.youtube.com/watch?v=4qvdNixFvCc #xref - [F2] Check whether Learn WordPress or DevHub content could be restructured per the answer-first/atomic-sentence tactic as a low-cost experiment — #articleseed ### → Things 3 (max 2 actions) - [ ] Draft a short internal note on `llms.txt` adoption status across owned WordPress-related sites — context: technical SEO follow-up #things - [ ] Add a recurring monthly calendar reminder to run the 5-question + 4-fixed-prompt AI-citation audit from this talk — context: personal brand/GEO tracking #things  📄 **Full text:** [[Full Document Contents/Articles/AI Search Is the Whole Company's Problem, Not Just SEO's.md|AI Search Is the Whole Company's Problem, Not Just SEO's by Emma Young (Hostinger)]] ## Highlights - So these this is just some statistics. There's going to be a lot of data. I went a little crazy with the use cases in this slides, but uh even when they do get to Google, about 68% of searches end without a single click. So, Google is now answering more and more questions before anyone even gets to your page. And the traffic model that SEO was built on is like structurally changing. And that's not just my opinion. So, this Spark Toro data in the image, it shows it's been slowly getting worse, not better. So, it's not slowing down either. In 2016, it was 45% of pe of pages not getting clicks. In 2019, it jumped up to 49%. In 2024, up to 60% and now we're at 68. So, it's not a trend that's dying or a trend that is slowly growing, but it's quite the slope. ([View Highlight](https://read.readwise.io/read/01m0d0j1mz49c1ww3g6xaqe04e)) - Note: SparkToro zero-click data: 45% (2016) to 68% (now) of Google searches end without a click. - Tags: atom/literature/h1, zero-click-search - So, here's the thing about AI search uh that I think a lot of the hype kind of obscures. So, it's not magic. There's no Harry Potter around, but it's retrieval. And what I mean by that is every major AI product, all of them, they retrieve from a search index. And they do this before generating an answer that they spit out to you. So think of a search index as a giant library of everything that Google has read and cataloged on the web. And then the AI, it goes to that library, it finds the most relevant pages, it pulls them all together, and it presents it as an answer. And for most of those AIs, that library is Google's. That index is Google's. ([View Highlight](https://read.readwise.io/read/01m0d0j4yf7xanj1h1nrx7e3g0)) - Note: Core mechanical claim of the talk: AI search retrieves from Google's index, it doesn't generate answers from nothing. - Tags: retrieval-not-magic, atom/literature/h1 - and foundations ran a study with Aerops where they took 57 million citations and analyzed them and it showed that 85% of AI overviews or sorry AI citations are coming from outside your brand's own domain. So everything that's on hostinger.com is only 15% 85% is something else. So the team whose entire job is to make your website uh visible to everybody is touching only 15% of where your AI visibility could be built on. And the other 85% is PR partnerships communities socials teams that are probably in this room now but don't really realize that that number 85% actually exists. So this is that memo that nobody sent. ([View Highlight](https://read.readwise.io/read/01m0d0j89q2hybxgzkmtxqb1r4)) - Note: The 85%-external-citation statistic — the evidentiary anchor for "this is a whole-company problem, not SEO's." - Tags: 85-percent-external-citations, atom/literature/h1 - So this is kind of what like brands look like right now I would say in AI search and it's not I don't want to say this like they're doing anything wrong but like nobody really planned for the things to change. So your affiliates they wrote their thing your landing pages team they wrote theirs. Let's say your tutorials went to slightly different angle and then your YouTube team also had a different take. And then somewhere there's like a Reddit thread from like 2019 or 2005, not written by you, but that the AI found and is treating as like the number one primary source for whatever reason. And AI pulls all of that together, shakes it up, and hands your potential client something that tastes like, you know, it actually depends on your situation. I can't give you an actual answer. So, I want you to think of it as like a bad cocktail and not bad because the ingredients are terrible, but bad because nobody followed a recipe. You've got gin from your landing page. Let's see, like vodka from your affiliates, some tequila from that Reddit thread that nobody on your team has ever read. And AI is the bartender and it's working with whatever you gave it. And right now, most brands are giving it a mess. And you're not unknown, but you are incoherent. And to me, in this in this realm, it's it's worse because unknown is fixable, you know, with presence. But incoherent means you're everywhere and you can still say nothing but consistently. ([View Highlight](https://read.readwise.io/read/01m0d0jdp2xjbp8s71v81m3416)) - Note: The "bad cocktail" metaphor — uncoordinated teams produce incoherent AI answers, which is worse than being unknown. - Tags: brand-coherence, atom/literature/h2 - And if you want to know how bad it can really get, again, Lily Ray, she published a fake article. So, she made up a Google Algo update that never happened. She even threw in a joke that Google approved it between slices of leftover pizza. And within 24 hours, Google's AI overviews were confidently repeating her fake pizza story as fact. And so AI doesn't know what's true or what's not true, but it does know what's being repeated. And she has a very big following. And she also encouraged all of us to spread this information. But it will say so these AI over AI overviews will say something confidently enough in enough places uh that it'll become the answer if it's true or not. Uh since then she has taken it down so there's less misinformation but uh she has a blog post about it. ([View Highlight](https://read.readwise.io/read/01m0d0jhbga9dzv4ykhen61cmr)) - Note: Lily Ray's fake Google algorithm-update hoax — AI Overviews repeated it as fact within 24 hours. - Tags: ai-overview-manipulation, atom/literature/h2