Day 1 of Tech SEO Connect brought together ten speakers who collectively mapped the new terrain of AI-powered search. The through-line was clear: the rules have changed, but the fundamentals haven’t disappeared—they’ve evolved.
What emerged wasn’t a single strategy but a framework for thinking about visibility in a world where answer engines arbitrate the relationship between brands and users. Here’s what I learned.
The Big Picture: How AI Search Actually Works
Krishna Madhavan (Microsoft) opened with the insider view. His core message: visibility now belongs to content that AI can trust, understand, and ground. The grounding pipeline has three phases—query understanding happens before retrieval, then retrieval, then evidence synthesis. Freshness affects all three. IndexNow drives 50% of clicked newly-indexed URLs on Bing. The tactical insight: use data-nosnippet for precision control rather than robots.txt as a blunt instrument.
Jamie Indigo (Cox Automotive) provided the skeptic’s counterpoint. “AI search is not a search engine,” she said. “Search engines are information retrieval systems. AI search is a model trained on a corpus plus retrieval augmentation, sometimes.” There is no AI index—your brand has a “tendency to exist.” User embeddings create personalized echo chambers. And critically: “Nothing AI says is ever true. It’s just probable. It’s spicy autocorrect.”
The tension between these perspectives is productive. Madhavan tells us how to work with the system; Indigo reminds us the system isn’t what it claims to be. Both are right.
Schema: The Data Layer for the Agentic Web
Martha van Berkel (Schema App) made the case that schema markup isn’t just for SEO anymore—it’s brand architecture for AI. Entity linking drives measurable results (she shared Brightview and Syncrometer case studies). Knowledge graphs solve hallucinations: Wells Fargo saw 91% accuracy with their knowledge graph versus 43% with GPT-4 alone.
The bigger point: NLWeb, a Microsoft project led by schema.org’s founder, uses structured data for agentic endpoints. MCP (Model Context Protocol) and Apps SDK are coming for agent access. Schema is becoming the API layer between your content and AI systems.
Alex Haliday (AirOps) reinforced this with his “yes” on schema: go deeper, prioritize high-value pages, and don’t forget ImageObject for Google Images traffic. His framework of five tactical decisions (markdown-only pages: no; llms.txt: no; schema: yes; server logs: yes; answer blocks: yes) gave clear guidance for immediate implementation.
Testing: What the Experiments Reveal
Jori Frost (FoodBoss) built test sites (JoriCon, JoriFest) specifically to study RAG bot behavior. Key findings: routing failed because bots wanted efficiency and grabbed embedded JSON-LD directly. Structure enables reasoning—flat text couldn’t answer complex questions, but structured data could. Context bias is real (first touch matters). And “appetite vs. crawl budget”—bots stop after about 3 pages deep. Her conclusion: schema is the API for logic.
Brie Anderson (Beast Analytics) brought methodology discipline with her Beast Cycle framework: Benchmark → Explore → Analyze → Strategize → Test → repeat. Her emphasis on getting KPI agreement before testing, visualizing data properly, and using a Start/Stop/Scale decision framework applies whether you’re testing traditional SEO or AI optimization. The SEO Jobs case study (6.5% application increase from a single page optimization) showed how iteration compounds.
Video: The Shortcut to AI Visibility
Cindy Krum (MobileMoxie) made a compelling case for video as the fastest path to AI citations. YouTube is the top cited source in AI platforms, and videos appear directly in AI Overviews—they’re not summarized like text content. Google has added a short-form video tab. Engagement signals matter more than metadata. The first 3 seconds are critical. Audio is processed for trending words.
Her tool stack: Opus Clips for auto-clipping long videos, ManyChat for DM automation, VidIQ/MetroCool for analytics, Waff.com for clip farming. The warning: platform risk is real—YouTube is deleting channels. Diversify across platforms.
Technical Infrastructure: Links, Logs, and LLMs
Serge Bezborodov (JetOctopus) tackled internal linking at scale. Text embeddings fail with large sites (10K pages = 100M permutations) and content-thin product pages. His solution: category-based matching—compare within category, then parent categories. The donor/acceptor model identifies crawl budget pages versus underlinked business pages. Don’t touch existing links; build a layer on top. He released JetOctopus Internal Linker as a free, open-source tool that works with Screaming Frog exports.
Baruch Toledano (SimilarWeb) pulled back the curtain on how SEO platforms actually work. His triangulation principle: no single data source is trustworthy alone. Even GA and GSC don’t align perfectly. For LLM cost optimization, his three-step approach (test feasibility with best model → try embeddings → distill with teacher-student) achieved 12x cost reduction. The reminder: behind every metric in our tools are engineering decisions about cost, accuracy, and validation.
Jamie Indigo (Cox Automotive) emphasized log file analysis as ground truth. The bot agreement is broken—they’re not crawling politely, not always declaring themselves, and not necessarily sending traffic. Her advice: befriend your SREs, watch for undocumented user agents, track 499 errors (when ChatGPT gave up waiting for your server), and repeat disallows for CC bot since Common Crawl is the foundation many models are built on.
Decision Intelligence: Making Better Choices with Data
Tyler Gargula brought the DECIDE framework: Define, Extract, Clean/Transform, Integrate, Distribute, Execute. His emphasis on avoiding bias (confirmation, selection, outcome, recency) and integrating multiple data sources addressed a universal challenge: “You might be solving the wrong problem perfectly.”
Key insight: use metric slopes rather than period-over-period comparisons. Month-over-month or year-over-year misses the context between those points. Slopes show true trajectory. And every analysis needs next steps—if it doesn’t lead to a decision, what was the point?
Emerging Themes Across Day 1
- Structure beats prose. Multiple speakers (Madhavan, van Berkel, Frost, Haliday) emphasized that AI systems prefer structured data they can parse and reason with. Schema isn’t optional anymore—it’s the interface layer.
- Freshness is a first-class signal. Madhavan showed freshness affects all three phases of the grounding pipeline. Publishing cadence is now a competitive advantage.
- Meta descriptions are back. Both Indigo and Haliday noted that ChatGPT’s snippet often is the meta description. Time is a flat circle.
- Performance matters for real-time retrieval. Time-to-first-byte affects whether you get cited during RAG. Core Web Vitals influence Google AI systems.
- Video is underrated. YouTube citations appear directly in AI Overviews without summarization. It’s a shortcut to brand visibility.
- Triangulate everything. No single data source tells the truth. Cross-reference logs, analytics, and rank tracking. Be skeptical of AI’s self-reporting.
- The bot agreement is broken. AI crawlers aren’t playing by the old rules. Plan for aggressive crawling, undocumented user agents, and uncertain traffic reciprocity.
What’s Actionable Now
From Day 1, these are the immediate priorities:
- Audit your schema. Go deeper than basic implementation. Prioritize high-value pages. Add ImageObject. Think of schema as your API to AI systems.
- Rewrite your meta descriptions. They’re being used as snippets by AI crawlers. Spoil your content—give the answer upfront.
- Implement IndexNow. If you’re not using it, you’re leaving freshness signals on the table.
- Start log file analysis. Identify AI crawlers (documented and undocumented). Track what they’re hitting and whether it’s useful. Monitor 499 errors.
- Block non-HTML from AI crawlers. Use X-Robots directives. If they cite your API JSON, users land on unusable pages.
- Disallow CC bot where appropriate. Common Crawl is the foundation for many models. Your robots.txt for AI crawlers should include it.
- Test video content. YouTube is the top cited source. Videos appear without summarization. Engagement beats metadata.
- Fix your site speed. Time-to-first-byte affects real-time RAG retrieval. Core Web Vitals affect Google AI systems.
The Day 1 Takeaway
Day 1 established that AI search isn’t replacing traditional SEO—it’s adding a new layer on top. The fundamentals (structure, performance, freshness, quality content) still matter. But the delivery mechanism has changed. AI systems are intermediaries now, and optimizing for them requires understanding how they actually work, not how they’re marketed.
As Jamie Indigo put it: “We are still tech SEOs, and this is still a bot.” The tools and tactics are evolving, but the mindset—curiosity, skepticism, experimentation—remains the same.
Be a feral trash cat. Share what you learn.
Day 1 Speakers
- Krishna Madhavan — Microsoft — Grounding pipeline, freshness, IndexNow
- Martha van Berkel — Schema App — Schema markup, knowledge graphs, agentic web
- Jori Frost — Experimentation framework, testing AI crawlers
- Alex Haliday — AirOps — 5 tactical decisions for AI SEO
- Brie Anderson — Beast Analytics — Testing methodology, Beast Cycle framework
- Cindy Krum — MobileMoxie — Video SEO as shortcut to AI visibility
- Serge Bezborodov — JetOctopus — Internal linking at scale
- Tyler Gargula — Decision intelligence, DECIDE framework
- Baruch Toledano — SimilarWeb — Data validation, LLM optimization at scale
- Jamie Indigo — Cox Automotive — AI assumptions, trash cat methodology
