Steve Toth delivered a presentation on strategies for optimizing content for AI and Large Language Models (LLMs), positioning LLM optimization as “the opportunity” in modern B2B marketing. Toth, the founder of AINotebook.com (22,000 subscribers) and CEO of Notebook Agency, presented a technical framework for transitioning from traditional SEO to ensuring brands are accurately cited and, critically, recommended by AI engines. His methodology focuses on preempting the buyer journey by optimizing for LLM refinements, eliminating deal-breaking criteria, and implementing a data-driven Truth Alignment Framework.
Key LLM Optimization Systems
The framework relies on a shift in focus from discovery to establishing the brand as a definitive, compliant solution within an LLM’s reasoning chain.
- AI Refinement Synthesis: Toth explained that LLMs often use agent-like tools, such as Deep Research, and initiate a process called refinement synthesis by asking follow-up questions after an initial query.
- Process: Aggregate these refinement questions across different models to identify common themes (e.g., comparisons, pricing, ICP fit).
- Goal: Optimize content for these themes to ensure the brand appears in the subsequent, high-intent AI search results.
- Deal Breaker Detector: LLMs function as fit engines, finding the right solution based on user constraints. This tool identifies critical friction points—such as missing features, integration gaps, or compliance requirements—that lead to silent disqualification during an AI-driven conversation.
- AI Info Pages: These are structured, bot-facing markdown pages designed specifically for LLMs. They contain comprehensive, authoritative brand facts (founding, services, clients, integrations, compliance).
- Action: Link the page sitewide (e.g., “Hey AI, learn about us“) to maximize LLM discovery and citation.
- Comparison Page Strategy: Ultra-relevant comparison pages materially aid LLM reasoning by centralizing evaluative data.
- Optimization: Create pages tailored to specific ICPs (e.g., “ClickUp versus Asana for Marketing Agencies“) and include dynamic dates in title tags (e.g., “November 2025“) to leverage LLMs’ explicit recency bias.
Crucial Content Pillars for AI Selection
Toth identified the top criteria that LLMs use to refine queries and the common deal-breakers that cause disqualification, necessitating explicit, passage-level coverage on-site.
| Pillar | LLM Function | B2B Criteria (SaaS Focus) |
| Refinement Synthesis | To narrow initial query relevance (LLM “lanes”). | Comparisons, ICP Mentions, Reviews, Pricing/Budget Information, Integration Capabilities. |
| Deal Breaker Coverage | To eliminate non-fitting solutions (fit engine). | Country-Specific Compliance (e.g., SOC 2, GDPR, HIPAA), 24/7 Support, Free Plan Limits, Integration Gaps (e.g., QuickBooks, Slack). |
| LLM Signal Strategy | To ensure accurate pickup by models like ChatGPT/Perplexity. | Abundance of accurate, consistent information; maintain content across ~250 sources that can skew facts. |
The Truth Alignment Framework (TAF)
The TAF is a system designed to ensure brands are not just mentioned, but are recommended at the final stage of the buyer journey, making the LLM as knowledgeable as a top salesperson.
The framework consists of a continuous optimization loop:
- Truth Notebook Creation:
- System: Centralize a validated ontology/taxonomy of product truths and sales-grade answers, seeded from help docs, battle cards, and sales transcripts.
- Interrogation and Scoring:
- Process: Use buyer-style prompts (testing non-branded fit scenarios) to interrogate LLMs.
- Key Metric: Measure four factors to produce a Truth Score: Accuracy, Source Clarity/Consolidation, Coverage of key truths, and Recommendation Presence.
- Remediation and Optimization:
- Action: Address root causes, such as on-site consolidation gaps or off-site misrepresentations. Rebuild content for better retrieval and saturate credible third-party domains with accurate truths.
- Evidence: Clients like Maptin, Spellbook, and Ownr used this framework to become top recommended solutions in their respective categories.
Actionable Takeaways
To immediately begin optimizing for the age of Against Search, Steve Toth recommends the following technical steps:
- Build an AI Info Page: Create a structured, bot-facing markdown page with authoritative brand facts (compliance, integrations, ICP).
- Optimize Help Center for Retrieval: Structure FAQs and help articles to explicitly answer multi-criteria buyer queries in easily retrievable passages.
- Focus on Refinement Themes: Use the Deep Research Synthesizer to classify themes (pricing, comparison, ICP) and align passage-level content to those exact themes.
- Launch ICP-Specific Comparisons: Create comparison/alternatives pages tailored to specific user segments and include dynamic, current dates in title tags.
- Address Deal-Breakers: Run the Deal Breaker Detector to surface friction points; produce FAQ rebuttals that explicitly cover compliance, support, and integration gaps.
- Implement TAF: Establish a Truth Notebook (validated facts) and baseline your brand’s Truth Score via LLM interrogation prompts.
- Monitor and Adapt: Pursue abundance of accurate, consistent information across credible sources; monitor LLM bot activity and serve specialized, clear content to improve citation quality.
My Take
Toth’s framework is clearly built for B2B SaaS companies with sales teams and marketing budgets. But strip away the enterprise layer, and there’s a core insight here that matters for solo publishers and affiliate marketers too: LLMs are becoming the new gatekeepers, and they don’t care about your Domain Authority.
The Truth Alignment Framework is essentially what good affiliate content has always done—answer real buyer questions honestly—except now you’re optimizing for a machine that reads your entire page, not a human who skims headings. The “AI Info Page” concept is interesting but irrelevant for most publishers. What is relevant: structuring your comparison and review content so LLMs can extract clean, passage-level answers to specific buyer queries. If ChatGPT can’t pull a clear recommendation from your “Best X” article, you’re invisible in the AI layer.
The Deal Breaker Detector concept is the most immediately actionable idea here. Run buyer prompts through multiple LLMs and see where your recommended products get disqualified. Then cover those objections explicitly. It’s basically AI chatbot optimization applied to your existing content rather than building new systems from scratch.
What’s missing from this talk: any acknowledgment that LLM recommendations are still wildly inconsistent. I’ve tested the same buyer prompt across ChatGPT, Perplexity, and Gemini and gotten completely different product recommendations. The source poisoning risks are real too—competitors can flood LLM training data with misleading information. Don’t bet your entire strategy on LLM optimization. It’s an additional channel, not a replacement for adapting to AI search more broadly.
Bottom line: The refinement synthesis approach—mapping follow-up questions LLMs ask—is genuinely useful for structuring content. The rest is enterprise packaging around principles that good topical coverage already addresses. Start with the refinement mapping, skip the agency fees.