Steve Toth has spent 15 years in SEO and now runs one of the sharpest AEO/GEO consultancies in B2B. His core argument: stop tracking where your brand ranks in LLM responses. Start measuring whether LLMs represent you accurately.

His Trust Alignment Framework scores how well ChatGPT, Perplexity, and Gemini answer questions about your product across six pillars - vertical, company size, comparisons, pricing, integrations, and features. The gap between your "sales-grade answer" and the LLM's answer is your visibility problem.

Steve walks through live demos showing how ChatGPT Deep Research and Perplexity surface follow-up refinements - and how collecting those refinements across 5-8 runs reveals which deal-breaker topics matter most in your category. He also shares a Claude project that clusters Google Search Console keywords by intent, giving B2B teams a proxy for LLM search demand when no reliable prompt volume data exists.

The conversation covers how each model cites differently - ChatGPT prefers general pages, Google AI Mode pulls specific passages from case studies and UGC - and why passage-level optimization matters more than page-level. Steve closes with his Spellbook case study: 90% non-branded organic traffic growth by targeting emerging keywords in the legal AI space and capitalizing on competitor sentiment gaps.

---

Key Takeaways

1. LLM leads convert 4-5x higher than Google traffic - ChatGPT referral visitors spend 4-5x more time on site and convert at 4-5x the rate. These buyers arrive pre-educated with specific deal-breakers already defined. Your sales team closes them faster.

2. Stop tracking brand mentions in LLMs - measure representation accuracy instead - The Trust Alignment Framework compares your ideal sales answer against what the LLM actually says across six pillars (vertical, company size, comparisons, pricing, integrations, features). The delta is your real visibility gap.

3. LLM prompt volume tools are unreliable - use intent clustering as a proxy - Every word added to a prompt makes it less likely to be searched twice. Steve built a Claude project that clusters Google keyword data by intent and aggregates volume across the entire cluster, giving directional demand signals for AEO prioritization.

4. Each AI model cites sources differently - ChatGPT favors first-party "ultimate guide" pages. Google AI Mode pulls specific passages from case studies and UGC. Claude uses the Brave search index. Optimizing for one model does not guarantee visibility in others.

5. Passage-level optimization beats page-level for AI Mode - Google AI Mode uses a passage ranking index, not a page ranking index. It looks for 100-300 token excerpts that support its reasoning chain. You can pepper relevant content across case studies, homepages, and comparison pages rather than building one monolithic page per topic.

Learn More

- SEO Notebook - https://seonotebook.com - Steve's weekly SEO newsletter, running since 2019

- AI Notebook - https://ainotebook.com - Weekly newsletter focused on AEO/GEO strategies

Connect with Gen:

- www.supermarketers.ai

- www.linkedin.com/in/genfurukawa

Why Keyword Volume Is Useless for AI Search (And What to Track Instead) | Steve Toth, Founder @ Notebook Agency