Seen, but not cited: What a billion sources say about your brand

Deck from Chris Donnelly's speech at BrightonSEO 2026 (October 2026).

Transcript

1. Seen, but not cited: What a billion sources say about your brand Chris Donnelly Searchable @trysearchable
2. But don't worry. We're OK.
3. A findings-led keynote by Chris Donnelly, CEO/Co-founder @ Searchable Seen, but not cited: what 1+ billion sources say about your brand 1.09B source and citation rows from real AI answers, Jan–Aug 2026 A research study by Searchable BrightonSEO United Kingdom · 2026
4. The question How do I get my brand to show up in AI?
5. Common Answers What people say about AI search. Three ideas the industry repeats most often. 01 02 03 Best gravel bike for long-distance racing? Specialized Diverge or Cannondale Topstone? AEO = SEO Best carbon gravel bike under $4,000? Just make more content Track the right prompts AEO is the same as SEO Publish more articles, expand landing pages and add schema. Choose the right questions, then monitor them over time. Apply the same rankings, traffic and authority playbook. Searchable · BrightonSEO United Kingdom · 2026 03 / 29
6. The reality What the evidence says instead. Each answer misses a different part of how AI search actually works. 01 Reality More content is not the same as more citations. Most of the answer is assembled off your website. Searchable · BrightonSEO United Kingdom · 2026 02 Reality One prompt is not the market. The engine rewrites, expands and brands the question itself. 03 Reality Rank is access, not selection. Each engine reads a different slice of the web. 04 / 29
7. Finding one Two-thirds of citations come from content you could never publish. On unbranded category questions, most of the evidence sits on surfaces no brand in the category controls. Everything the engine reads ~66% of what AI reads is earned: editorial, institutions, reviews, social and forums. 66% And that 2.9% splits by how the question is asked: 46% your site’s share of the reading when the buyer asks about you by name 31.1% <5% when they ask a category question without naming you 2.9% Earned Ground nobody owns Searchable warehouse · median of 7,307 brands · 14-day window to 2 Sep 2026 · name gate pooled across engines Competitor-owned Someone else's pages Your site Of the whole 05 / 29
8. Finding two One prompt is a point. Your prompt universe is the market. AI rewrites the question, expands it into its own searches and sometimes inserts brands before retrieval begins. HUMAN PROMPT ENGINE-WRITTEN SEARCHES best payroll software for 50 people One question What is the best payroll software for my business? Gusto vs ADP for growing teams payroll software with accounting integrations switching payroll providers checklist best payroll for contractors QuickBooks payroll alternatives ChatGPT capture corpus · July 2026 · 6,713 engine queries · 8,182 citations · brand injection judged on 1,000 of 854,792 responses WHAT THE CAPTURE SHOWED 99% use the engine's words, not the human's. 1 in 7 inject a brand before retrieval even begins. 06 / 29
12. Finding three SEO can get you a ticket. It cannot get you in the seat. SEO earns access to one index. AI systems still decide what to retrieve, read and cite. One brand, three retrieval systems 2 in 3 of a median brand's Google top-three pages were never read by ChatGPT. What got read instead Google AI Its own index, mostly end to end ChatGPT Building its own index + outside suppliers Claude Brave index + model memory A plain, unoptimised data page: cited 210 times while brand pages ranking above it went unread. Searchable study · 134 brands · ~88,500 Google top-three pages · June 2026 · rank-decay evidence is ChatGPT-scoped 10 / 29
13. Why now The opportunity will not look like this forever.
14. The window The Wild West is already starting to settle. We are moving from quick wins and tactics into systems, standards and durable category leadership. 2022 2024 Now Next 6 months Frontier Land grab Settlement Standards A new answer layer appears. Distribution is open. Brands test prompts, content and citation tactics. Indexes diverge. Leaders and repeatable systems emerge. Hacks get penalised. The benchmark gets harder to displace. ~$1.6tn invested since 2013 more than five times the Apollo programme, in today’s money AI investment comparison: Reuters, May 2026 · timeline framing synthesised from Searchable's tracked engine changes 12 / 29
15. The precedent Last time distribution reset, a few brands built the default. Social is the closest parallel: a new discovery surface, explosive organic reach, ads - and then algorithmic scarcity. AI search is here 01 02 03 04 Open distribution Organic growth Ads arrive Reach gets rationed Anyone can compound reach Category leaders emerge The platform monetises attention Paid access replaces free reach Amazon $2.7tn 28% of 2.4B monthly visits from organic search HubSpot $11bn ~8M organic visits a month to the blog Zillow $7.5bn organic is its largest traffic channel Company values dated to 2021–Sep 2026 · traffic estimates: Semrush, Similarweb and Ahrefs · see supplied datapack for vintages Zapier $5bn 1.6M organic visits a month to the blog 13 / 29
16. The channel shift Organic reach falls. The cost of access rises. The units differ, so every panel uses its own scale. The repeated pattern is the evidence: organic distribution falls while paid access and zero-click searches rise. Instagram −81% Engagement rate Facebook −88% Fans reached TikTok −58% Followers reached 2019 1.60% 2012 16% 2022 24% 2026 0.30% 2014 ~2% 2023 10% LinkedIn −50% Views per post Google Ads 2.3× Average search CPC Google Rising Zero-click searches Start 100 ~2015 $2.32 2019 49% One year 50 2026 $5.42 2026 68% Rival IQ · EdgeRank/Ogilvy · Fanpage Karma · van der Blom · WordStream/LocaliQ · SparkToro 14 / 29
17. Audience and money The audience arrived before the ad market. ChatGPT reached mass adoption before advertising became a material business. Other engines are still choosing how to monetize their audiences. Weekly active users 1B+ +800M OpenAI investor projection Ads launched · Feb 2026 ≈900M weekly users 1B+ 2030 ad revenue OpenAI's leaked projection, not reported revenue. 750M Other engine audiences 500M 250M 0 Aug 2024 $100B AI Overviews Gemini 2.5B monthly 1B monthly Ads appear on about 25% of results The standalone app remains ad-free Perplexity Dec 2024 Apr 2025 Aug 2025 Dec 2025 Apr 2026 Claude 780M queries/mo $65B run rate The company ended its advertising tests No ads; revenue is led by API and enterprise Aug 2026 Showing reported weekly-active-user milestones from Aug 2024 to Aug 2026 +800M OpenAI official weekly-active-user milestones, Aug 2024–Aug 2026 · Google, Perplexity and Anthropic official figures · projections are OpenAI's own, unconfirmed 15 / 29
18. The evidence What 1+ billion sources say about getting cited
19. The machine Six steps sit between the question and your brand. The funnel narrows at every stage. A brand can win - or disappear - before the answer is written. 01 02 03 04 05 06 Memory Query Retrieve Read Filter Answer What it already believes Writes its own searches Chooses an index ~10 sources ~1 gets credit Only then names a brand Searchable network-layer captures across ChatGPT, Claude and Copilot · Jun–Aug 2026 17 / 29
20. The memory layer Before an engine searches, it already has an opinion of you. Before retrieval begins, the model already carries a learned association between brands and categories. Model memory Brand shortlist Search - or no search ChatGPT Claude 15.2% ~50% 9.9% Memory shapes the shortlist Memory can replace retrieval Same pattern, third engine of buying questions have a brand written into the search before retrieval. of matched questions were answered with no web search at all. Our captures · ChatGPT injection study: 15.2% (CI 13.1–17.6) · Claude: 657/1,370 searched · Copilot: 373 prompts, 11 Aug 2026 Copilot of prompts were answered from memory on its own live connection. 18 / 29
21. The supply war AI engines are turning away from one Google-shaped web. Google is one input, not the internet. Each engine is building - or choosing - its own retrieval stack. ChatGPT Building its own index + outside suppliers Industry-documented, 2026 Claude Brave index + model memory Observed against Brave API AI Overviews Its own index, mostly end to end By construction Five stacks. Five different visibility problems. Non-Google crawlers still struggle with clientside JavaScript. A site Google can render may remain invisible elsewhere. Copilot Bing, server-side Documented in 373-prompt capture Perplexity Own crawler and index Company documented Searchable capture studies, July–Aug 2026 · official index/crawler documentation · ChatGPT own-index claim is industry-reported, not OpenAI-confirmed 19 / 29
22. Engine divergence Three engines. Three views of the world. Ask each engine the same buying question and it credits a different kind of source. Winning one index does not carry over to the next. ChatGPT AI Overviews Perplexity Editorial 12% Editorial 19.2% Editorial 25.8% Institutional 11.5% Social + forums 10.7% Reviews 2.9% Social + forums 0.7% Institutional 5.6% Social + forums 1.6% The same social and forum content counts 15× more in AI Overviews than in ChatGPT Searchable warehouse · 13–26 Aug 2026 · ~4M credited citations per engine · unbranded prompts · compare order, not absolute levels 20 / 29
23. The visibility equation AI visibility has an owned core inside an earned network.
24. The operating system One system. Two surfaces. One weekly measurement. Owned work creates the proof. Earned work distributes it. Weekly measurement keeps both surfaces connected. Owned and earned operating system · One weekly measurement cycle Owned Owned core Earned network Create the evidence AI can read 01 Let AI read the site 02 Find cited competitor pages 03 Create answer-shaped pages Earned Distribute proof across the source mix Measure weekly 01 Find the dominant channel 02 Win its leading platform Track the same prompt universe 03 Build the editorial pipeline 04 Land in the credited slice Owned proof and earned distribution move on the same weekly measurement cycle. Searchable · BrightonSEO United Kingdom · 2026 22 / 29
25. Owned visibility What actually moves owned visibility Start with access and answer-shaped pages. Treat ranking as a ticket. Leave tricks at the bottom. S A B C Let AI read it and make the page the answer This is the prerequisite. Full liftable extracts surfaced 91–100%. Rank somewhere, then restructure around questions Top 10 pages see about 4× the citation odds. Our site earned its first citations in 4 days (n=1). Find the page AI actually reads Domain read reached 95%. The top-ranked page reached only 42%. Schema and phrasing tricks We found no evidence of an effect in our data. Searchable studies · own site 2.9% + competitors 29.5% · rank effect is ChatGPT-scoped · restructure result is our site, n=1, no control 23 / 29
26. Owned, by industry How much of your fight is on owned ground? B2B tech fights about 40% of the battle on owned ground. Health, about 20%. Owned share by industry Median share of citations from owned domains 0% 10% 20% 30% 40% 50% Transport & logistics 43.5% B2B tech & media 41.4% Hospitality 33.7% Education 30.6% Finance 29.6% Retail & ecommerce 28.6% Professional services 26.9% Construction 26.4% Beauty & personal care 26.3% Health & medical 19.5% Industry median · 0–50% scale Searchable warehouse · median per brand · n≥30 brands per industry · 14-day window to 2 Sep 2026 · selected rungs from the supplied 14-industry dataset 24 / 29
27. Earned, by industry Where does your earned fight happen? Health is an institutions game. Retail is an editorial one. Read your row left to right - that is your channel budget. Earned source mix by industry Share of earned citations by source class 0% 20% Health & medical Institutions 40% 60% 74.1% 80% Editorial Other 100% 12.3% Education Professional services Finance Beauty & personal care Construction Transport & logistics B2B tech & media Hospitality Retail & ecommerce 19.6% 51.8% Each row totals 100% of earned citations. Searchable warehouse · pooled earned citations · 14-day window to 2 Sep 2026 · unbranded prompts · industry labels audited ~90% 25 / 29
28. Inside earned Earned is a universe of sources pieced into one answer. AI answers combine institutions, editorial, reviews, social and forums - then credit only a sliver of what they read. Share of credited source classes Editorial Long tail 0–100% 29% Institutions NIH 25% Reviews TripAdvisor · Trustpilot · Clutch · G2 4% Assembled into one AI answer Other / unclassified · 36% Social YouTube 53% · LinkedIn 19% 4% Forums Reddit 72% 2% Searchable warehouse · 95.2M earned citation rows · pooled · 14-day window to 2 Sep 2026 · channel shares exclude third-party and unclassified tail No single source wins the answer. The mix is the strategy. 26 / 29
29. Social formats Even inside a platform, AI has a format it prefers. Social is the channel. YouTube and LinkedIn are the platforms. Each platform still has a preferred format. Social source mix · Platform and format preferences Parent channel Social YouTube 52.6% LinkedIn 18.6% Other social · 28.8% LinkedIn 57% written formats Pulse articles and member posts carry the platform. YouTube Pulse articles 29% 89% main videos Main videos 89% Member posts 28% Shorts remain a small part of the cited format mix. Shorts 7% Other formats 43% Other formats 4% The channel sets the platform; the platform sets the format. Searchable warehouse · reads grain · 14-day window to 2 Sep 2026 · YouTube 2.12M reads · LinkedIn 757K reads 27 / 29
30. The runbook How to run the system. Five moves. One repeatable operating rhythm. 01 02 Measure the universe Make owned readable Map buying questions and split owned from earned. Fix access first. Build pages that are the answer. Searchable · BrightonSEO United Kingdom · 2026 03 Track engines separately Google, Bing, Brave and private indexes diverge. 04 Invest by battlefield Use your industry and channel mix - not a generic playbook. 05 Re-measure every week Turn every answer into the next build and distribution list. 28 / 29
31. Your first step is building your prompt universe. Map the questions your buyers ask AI, see which sources shape the answers and turn that evidence into your next move. Build your prompt universe Come see us at Booth #25. We'll help you set yours up. SCAN ME BrightonSEO United Kingdom · 2026 29 / 29