It's ALL AI search: Building a unified view for growth

Transcript

1. It’s ALL AI search: building a unified view for growth Ray Grieselhuber DemandSphere Speakerdeck.com/raygrieselhuber @raygrieselhuber
2. Hi, I’m @raygrieselhuber Founder & CEO of DemandSphere 18+ years experience in data systems, engineering, and enterprise SEO Speakerdeck.com/raygrieselhuber @raygrieselhuber Slides:
3. There is no traditional vs. AI search Speakerdeck.com/raygrieselhuber @raygrieselhuber
4. It’s all AI search - the question is the user experience Speakerdeck.com/raygrieselhuber @raygrieselhuber
5. 3 Search Experiences Speakerdeck.com/raygrieselhuber @raygrieselhuber
6. Let’s talk about search in LLMs Speakerdeck.com/raygrieselhuber @raygrieselhuber
7. The context window is not infinite Speakerdeck.com/raygrieselhuber @raygrieselhuber PROMPT INPUT COMPLETION PREVIOUS TOKENS NEXT TOKENS
8. “Infinite” meaning complete internet up to date in real time Speakerdeck.com/raygrieselhuber @raygrieselhuber
9. So, we need Live Retrieval Speakerdeck.com/raygrieselhuber @raygrieselhuber
10. The app wraps the LLM to determine tools Speakerdeck.com/raygrieselhuber @raygrieselhuber App (ChatGPT, etc) LLM (GPT-5, etc) USER INPUT MODEL QUERY LIVE RETRIEVAL
11. Live Retrieval “solves” the finite context window problem Speakerdeck.com/raygrieselhuber @raygrieselhuber LLMs Search Engines Retrieval Layer High quality & more relevant answers
12. Let’s talk about the Retrieval Layer Speakerdeck.com/raygrieselhuber @raygrieselhuber
13. The Retrieval Layer abstracts search engine queries Speakerdeck.com/raygrieselhuber @raygrieselhuber APP LLM QUERY RETRIEVAL LAYER External channels - web tool: - search(query) - open(url) - file connectors: - file_search etc - recording connectors Workspace channels Visuals & Files Policy & Guardrails Search operators
14. The search tool queries one or more search engines Speakerdeck.com/raygrieselhuber @raygrieselhuber External channels - web tool: - search(query) - open(url) - file connectors: - file_search etc - recording connectors Workspace channels Visuals & Files Policy & Guardrails Search operators
15. Google & Bing form the backbone of AI search Google AI Overviews Speakerdeck.com/raygrieselhuber @raygrieselhuber
16. Yes, ChatGPT is using Google Speakerdeck.com/raygrieselhuber @raygrieselhuber
17. Data Study (April 2025) • 10,000 prompts • 17 industries • Google • Bing • ChatGPT • Perplexity
18. Measure index overlap in citations Speakerdeck.com/raygrieselhuber @raygrieselhuber
19. Data in April 2025 was showing heavy Google usage Speakerdeck.com/raygrieselhuber @raygrieselhuber 51.72% Google Index ChatGPT 4o 14.26% Bing Index ChatGPT 4o 34.02% Custom / Remix ChatGPT 4o 37.12% Google Index Perplexity 45.63% Custom / Remix Perplexity 17.25% Bing Index Perplexity
20. ChatGPT is a Google wrapper (Gasp) Speakerdeck.com/raygrieselhuber @raygrieselhuber
21. What has changed? Speakerdeck.com/raygrieselhuber @raygrieselhuber
22. Explosion in query fan-outs Speakerdeck.com/raygrieselhuber @raygrieselhuber Commercial best winter tires Query fan-out Query fan-out Query fan-out Query fan-out Which brand is the best for winter? Why are Blizzak tires so good? Which is better for snow awd or snow tires? Do snow tires really make a difference?
23. Before: source query = retrieval layer query Speakerdeck.com/raygrieselhuber @raygrieselhuber Commercial best winter tires Query for live retrieval “best winter tires” RETRIEVAL LAYER
24. Now: source query = ~10-12 retrieval layer queries Speakerdeck.com/raygrieselhuber @raygrieselhuber Commercial best winter tires Queries for live retrieval “best winter tires” “which tire brand is best for winter” “why are blizzak tires so good” “which is better for snow awd or snow tires” “do snow tires really make a difference” etc. RETRIEVAL LAYER x 10
25. We know this because we capture the fan-outs Speakerdeck.com/raygrieselhuber @raygrieselhuber
26. ChatGPT (and others) are building their citations based on SERP data Speakerdeck.com/raygrieselhuber @raygrieselhuber
27. Let’s talk about Google SERPs Speakerdeck.com/raygrieselhuber @raygrieselhuber
28. It’s ALL AI search Speakerdeck.com/raygrieselhuber @raygrieselhuber AI Mode 2025 https://ai.google/our-ai-journey/ AI Overviews 2024 Gemini 1.0 2023 PaLM 2022 MUM 2021 Passage Ranking 2020 T5 2019 BERT 2018 Transformer Architecture 2017 TensorFlow 2015 Google Assistant 2016
29. The existence of Zero Click SERP Features does not mean that there are zero clicks 2024 Zero-Click Search Study Speakerdeck.com/raygrieselhuber @raygrieselhuber
30. Metric tree analysis Speakerdeck.com/raygrieselhuber @raygrieselhuber 70.5% Organic results 41.5% - Click 1,000,000 Searches 21.4% - New search 28.5% Google property 41.5% Click 58.5% “Zero Click” 37.1% Session ends
31. Metric tree analysis Speakerdeck.com/raygrieselhuber @raygrieselhuber 70.5% Organic results - Net CTR of 29% 41.5% - Click 1,000,000 Searches 21.4% - New search 28.5% Google property 41.5% Click 58.5% “Zero Click” 37.1% Session ends
32. Users are only on the first page Speakerdeck.com/raygrieselhuber @raygrieselhuber
33. But the index is the prize Speakerdeck.com/raygrieselhuber @raygrieselhuber
34. This is why Google removed num=100 Speakerdeck.com/raygrieselhuber @raygrieselhuber Commercial best winter tires Queries for live retrieval “best winter tires” “which tire brand is best for winter” “why are blizzak tires so good” “which is better for snow awd or snow tires” “do snow tires really make a difference” etc. RETRIEVAL LAYER x 10
35. Building a unified view Speakerdeck.com/raygrieselhuber @raygrieselhuber
36. The mindset Speakerdeck.com/raygrieselhuber @raygrieselhuber Understand model responses vs. live retrieval Connect relationships between available data Focus on human attention
37. The data stack Speakerdeck.com/raygrieselhuber @raygrieselhuber
38. Visibility monitoring Speakerdeck.com/raygrieselhuber @raygrieselhuber Data points: • Mentions • Citations • Share of Voice • Index Overlap Combine with: • SERP data • Log data • GA4 data
39. Referral data from GA4 Speakerdeck.com/raygrieselhuber @raygrieselhuber Analyze: • Sessions • Pages • Conversions • Revenue Use URL-level data to deliver additional insights
40. Log data is crucial Speakerdeck.com/raygrieselhuber @raygrieselhuber Analyze: • Bot behavior • Crawl issues • Content opportunities Combine with: • Visibility monitoring • Prompt research
41. SERP index overlap helps you predict visibility at scale Speakerdeck.com/raygrieselhuber @raygrieselhuber
42. Prompt research is a process, not a tool Speakerdeck.com/raygrieselhuber @raygrieselhuber Valuable data sources: - People Also Ask (PAAs) - Reddit - Keyword Planner - Synthetic query generation - Query fan-outs - Topic & intent mapping (content-focused)
43. Prompt volume should be seen as relational and directional Speakerdeck.com/raygrieselhuber @raygrieselhuber
44. Prompt volume challenge #1: there is no “Search Console for ChatGPT” Speakerdeck.com/raygrieselhuber @raygrieselhuber
45. Prompt volume challenge #2: Clickstream data is very messy Speakerdeck.com/raygrieselhuber @raygrieselhuber
46. Simplified metrics Speakerdeck.com/raygrieselhuber @raygrieselhuber
47. Simple data models to support relationships
48. Deceptively simple unified view of factors impacting behavior Speakerdeck.com/raygrieselhuber @raygrieselhuber
49. Takeaway #1: The lines are blurry between AI search experiences Speakerdeck.com/raygrieselhuber @raygrieselhuber
50. Takeaway #2: Question the origin of data sources Speakerdeck.com/raygrieselhuber @raygrieselhuber
51. Takeaway #3: Think in terms of unified views Speakerdeck.com/raygrieselhuber @raygrieselhuber
52. Come visit our booth! Speakerdeck.com/raygrieselhuber @raygrieselhuber x.com/demandsphere linkedin.com/in/raygrieselhuber Slides: