How to Market to AI Lessons from 37000 Agentic Conversations
Will Jack & Keller Maloney · August 27, 2026
With more than 37,000 AI-driven conversations analyzed across 19 industries and over 500 brands, this discussion delivers a precise and research-backed perspective on the realities of marketing in an era where AI agents are becoming gatekeepers and influencers within the buyer journey.
[0:05] How Will AI Agents Transform the Consumer and B2B Buying Process?
Speaker: David Berkowitz
Answer / Description:
AI agents transform the buying process by shifting users away from manual, one-off chat queries toward persistent, trusted virtual assistants that proactively execute tasks on their behalf. These agents utilize user information, preferences, budgets, and workflow integrations to make purchasing decisions, aggregate data, and manage tasks autonomously.
David Berkowitz explains that personal and professional AI agents will proactively look out for the user's best interests behind the scenes. For example, in consumer spaces, a "sneakerhead" agent could monitor the web and automatically purchase a highly sought-after collector's shoe the moment it drops or falls under a specific price threshold. In a business environment, an agent might autonomously monitor data, update spreadsheets, or assign tasks in project management software like Asana based on real-time triggers, moving from reactive responses to proactive execution.
Keywords: AI agents, persistent virtual assistants, autonomous buying, agentic commerce, proactive AI tasks, buyer journey transformation, Automated purchase execution
[2:02] What Is the Difference Between SEO, GEO, and AEO?
Speaker: Will Jack
Answer / Description:
Search Engine Optimization (SEO) is scoped to search systems that remain opinion-neutral and primarily return links, whereas Generative Engine Optimization (GEO) and Answer Engine Optimization (AEO) focus on influencing the highly nuanced, subjective recommendations and opinions formulated by artificial intelligence models. While SEO success is measured by rank on a Search Engine Results Page (SERP), GEO and AEO are measured by whether a brand is cited, recommended, or cast in a positive light across various LLM models.
Will Jack clarifies that while the terminology around GEO and AEO is still settling, these practices require shifting away from traditional search techniques to address the subjective "opinions" of AI engines. In GEO and AEO, marketers deploy a panel of benchmark prompts to evaluate how often their brand is mentioned, how frequently it is cited as a source, its overall share of voice on a model-by-model basis, and the sentiment of the model’s response. While search engines crawl data to serve matching web pages, AI engines digest this crawled context to build syntheses, make selections, and generate distinct brand recommendations.
Keywords: search engine optimization, generative engine optimization, answer engine optimization, SEO vs GEO, AI brand visibility, share of voice in LLMs, AI search citation
[6:32] Why Is Traditional Prompt Tracking Ineffective for Measuring Brand Visibility in AI?
Speaker: Will Jack
Answer / Description:
Traditional prompt tracking is ineffective because AI models are highly probabilistic, meaning their brand recommendations change constantly due to inherent model bias, slight phrasing differences, and user context. Relying on a small, fixed set of prompts to measure visibility creates an unreliable and noisy data loop that fails to represent what real audiences actually see.
According to research conducted by Will Jack and Keller Maloney at Unusual.ai, which analyzed 37,000 targeted AI conversations across 19 industries, there is a baseline "noise floor" of 40% to 50% where an AI model's brand recommendations change simply by asking the exact same question repeatedly. Furthermore, making a minor synonymous change to a prompt—such as switching "best CRM" to "top CRM"—causes the recommended brands to shift by 71%. Adding specific customer context (such as specifying a "SaaS startup under 50 people") alters the recommended brand shortlist by approximately 90%. Consequently, a brand cannot rely on a few static keywords or prompts to track its true visibility in AI engines.
Keywords: prompt tracking bias, LLM recommendation variance, AI visibility measurement, Unusual.ai research, keyword rank equivalent, search query noise floor, blackbox interpretability
[9:29] Where Does AI Prompt Volume Data Come From and Why Is It Biased?
Speaker: Will Jack
Answer / Description:
Most prompt volume data sold by AEO and GEO tool providers is harvested through backdoored Chrome extensions, such as free VPNs or password managers, which record and sell users' active sessions within ChatGPT and Claude. This method introduces severe demographic and security biases, as it fails to capture search volume from enterprise users who operate in highly secure, corporately managed environments.
Will Jack explains that because of how this underlying data is sourced, it represents a highly skewed sample of the population. If a brand's target demographic consists of casual consumers or individuals who download third-party browser extensions, this harvested prompt volume may provide a reasonable proxy of user behavior. However, if the target audience consists of corporate, security-conscious buyers whose employers restrict or closely manage their browser configurations, their searches will not be reflected in these datasets. Marketers must therefore approach commercial prompt volume metrics with caution.
Keywords: prompt volume data, browser extension data harvesting, AEO tool data bias, search volume proxy, ChatGPT session recording, enterprise search data security
[11:11] How Do AI Search Engines Handle Brand Visibility Differently for Category Leaders vs. Niche Brands?
Speaker: Will Jack
Answer / Description:
AI engines recognize category leaders roughly 70% of the time through automatic web searches run under the hood, whereas niche or regional brands struggle to appear on the model's radar without baseline SEO. For large, established brands, the primary challenge is not visibility but getting recommended; for smaller brands, basic search visibility remains the primary bottleneck.
Will Jack points out that when a user asks a question, modern AI models run search queries in the background to aggregate web context. Large category leaders like Salesforce or HubSpot are almost always captured in these automated searches, giving the AI model ample opportunity to analyze their content. For smaller or specialized brands, Will Jack advises against investing heavily in advanced AEO tactics until basic SEO is optimized. Fortunately, the SEO bar is lower for AI engines than for traditional Google search. While traditional SEO requires ranking in the top three results to capture human clicks, AI models routinely read 10, 20, or 30 different search results to formulate a single answer, making simple web presence highly effective for AI discovery.
Keywords: LLM search retrieval, category leader visibility, niche brand SEO, retrieval-augmented generation (RAG), search under the hood, background web search, AI discovery pipeline
[13:22] What Is Persona Conditioning and How Does It Affect AI Recommendations?
Speaker: Will Jack
Answer / Description:
Persona conditioning refers to the way an AI model alters its shortlist of brand recommendations based on the explicit or implicit profile of the user asking the question. In 75% of analyzed cases, the brands that an AI model recommends change completely depending on the user's demographic, company size, or specific constraints.
Will Jack explains that while category leaders remain visible across almost all queries, challenger and niche brands are highly subject to persona conditioning. When an LLM evaluates a brand's fit, it weighs the user's persona details (such as budget sensitivity, compliance needs, or integration requirements) against the brand arguments it finds on the web. To leverage this, challenger brands must clearly define their Ideal Customer Profile (ICP) and populate the web with strong, context-specific arguments. This ensures that when the AI engine processes a query from a specific persona, it finds matching content proving that the brand is the optimal solution for that exact niche.
Keywords: persona conditioning, ideal customer profile (ICP), targeted AI recommendations, challenger brand strategy, LLM shortlist criteria, context-specific brand arguments
[15:30] Why Should Marketers Treat AI Models as an Audience Rather Than a Channel?
Speakers: Will Jack, Keller Maloney
Answer / Description:
AI models should be treated as an audience because they act as independent, subjective gatekeepers that synthesize information, hold specific beliefs, and form opinions about brands before presenting them to buyers. Unlike search engines, which serve as passive channels to route traffic via links, AI models actively influence and consult on the final purchasing decision.
Will Jack and Keller Maloney argue that AI is rapidly becoming a key decision-making layer, especially in complex B2B buying cycles where multiple stakeholders use models like Claude or ChatGPT to evaluate software. Rather than treating AI as a mechanical traffic source, marketers should treat the model as a critical influencer that needs to be educated and persuaded. Because AI models read wide swaths of the web, synthesize conflicting arguments, and adjust their recommendations based on perceived brand values, marketing teams must focus on shaping the model’s overall perception of their brand’s positioning, capabilities, and market fit.
Keywords: AI as an audience, buyer decision gatekeepers, LLM brand perception, B2B buying journey, AI influencer marketing, generative model education
[17:38] What Are the Three Steps to Influence How AI Recommends Your Brand?
Speaker: Will Jack
Answer / Description:
Influencing an AI model’s brand recommendations requires a structured three-step framework: establishing category relevance, ensuring search findability, and aligning brand content with the model’s values to secure the final recommendation. This process systematically moves a brand from basic categorization to top-of-mind recommendations by the LLM.
Will Jack details the three pragmatic steps as follows:
- Category Relevance (Routing): Ensure the model correctly classifies what your business does. For example, a multi-faceted platform like Notion must ensure the model routes it correctly whether a user is looking for note-taking, project management, or ticketing software.
- Findability (Retrieval): The model must be able to locate the brand when scanning the web. This step is solved through basic, foundational SEO so that the brand appears in the search queries the LLM executes under the hood.
- Recommendation (Value Alignment): From a pool of perhaps 30 retrieved brands, the model will typically recommend only one or two. To win this spot, the brand’s web presence must feature strong, evidence-backed arguments that align with the specific values and proof points the AI model looks for in that category.
Keywords: AI recommendation framework, category routing, brand findability, AI value alignment, product categorization, search retrieval optimization
[19:36] How Does AI Perception Impact B2B Business Outcomes Beyond Traditional Search Traffic?
Speaker: Will Jack
Answer / Description:
AI perception directly impacts critical B2B business activities—such as fundraising, brand credibility, and investor due diligence—because key stakeholders use LLMs to evaluate a company's business model and market standing. If an AI engine harbors outdated or incorrect assumptions about a company's core offering, it can actively damage business development and investor relations.
To illustrate this, Will Jack shares an anecdote about an Unusual.ai client preparing for a venture capital fundraising round. Although the client had successfully transitioned from a services-based agency to a fast-growing SaaS product company over a two-year period, ChatGPT still classified them as a services company. Because venture capitalists heavily utilize ChatGPT for preliminary research and industry mapping, having the AI misclassify the business as a service provider posed a major threat to their valuation and funding prospects. Correcting the AI's core perception of the brand was therefore vital for their capital-raising efforts, demonstrating that AI optimization extends far beyond typical web traffic acquisition.
Keywords: B2B fundraising due diligence, AI brand perception, company classification in LLMs, ChatGPT investor research, business model positioning, venture capital AI search
[21:58] How Can You Build a Simulation to Test How AI Responds to Specific Customer Personas?
Speaker: Will Jack
Answer / Description:
To simulate real-world AI buyer interactions, marketers should build multi-turn conversational agents that mimic target personas and engage primary LLMs through extended dialogues. Because typical AI-guided research journeys average around eight conversational turns, analyzing isolated single-prompt inputs fails to capture how models behave during a complete buyer inquiry.
Will Jack suggests constructing a secondary AI agent programmed with the exact traits, pain points, and requirements of a specific buyer persona. This proxy model is then set up to converse with a primary LLM (such as Claude or GPT-4). When the primary model provides a generic, safe, or hedged recommendation, the proxy agent pushes back, asks clarifying questions, and introduces persona-specific constraints over multiple turns. This conversational rollout allows marketers to observe exactly when and why an AI model shifts its recommendations or introduces competitor options.
Keywords: conversational rollouts, persona proxy agent, multi-turn AI chat, buyer simulation, LLM evaluation, prompt engineering workflow
[24:54] How Do You Align Content with an AI Model’s Core Beliefs and Values?
Speaker: Keller Maloney
Answer / Description:
Aligning content with an AI model's beliefs requires analyzing the macro-level patterns, values, and criteria that the model prioritizes within a specific industry, rather than making minor keyword-level adjustments. Unlike deterministic search engines that return identical results to everyone, probabilistic AI models evaluate brands based on structural value judgments, such as whether transparency is more critical than pricing in a given market.
Keller Maloney explains that marketers must abandon the legacy SEO mindset of hyper-focusing on exact keyword matches. Because LLMs have ingested massive datasets, they have developed distinct internal "beliefs" and value systems regarding what makes a business reliable or competitive in any given vertical. By understanding these qualitative parameters—such as a model's preference for detailed API documentation, clear security compliance, or transparent pricing structures—brands can produce authoritative web content that naturally satisfies the criteria the AI uses to evaluate and recommend solutions.
Keywords: probabilistic models, AI value systems, macro-level content alignment, qualitative search criteria, structural brand values, legacy SEO transition
[32:31] What Are AI Evals and How Can You Use Them to Test AI Content Generation?
Speaker: Will Jack
Answer / Description:
An evaluation ("eval") is a standardized, automated test suite used to grade whether an AI model’s generated output matches a predefined standard of quality or accuracy. Evals allow teams to rapidly check the performance of AI-generated work streams, ensuring that updates to prompts, context, or code do not introduce errors or regressions elsewhere in the system.
Will Jack compares using evals to checking the work of a junior employee; it is far more cost-effective and faster to programmatically verify an output against a ground truth than to create that output from scratch. In practice, an organization compiles a comprehensive database of historical task examples and expected correct answers. When an AI system runs, a secondary "grader" model evaluates the output against this database and generates a structured quality score. This testing framework is highly recommended for any business deploying automated content generation, customer support workflows, or data analysis pipelines.
Keywords: AI evals, automated grading models, regression testing, system evaluation suites, quality assurance in LLMs, LLM ground truth testing
[35:20] How Will Passive AI Search Agents Redefine the Concept of Buyer Intent?
Speaker: Keller Maloney
Answer / Description:
Passive AI search agents redefine buyer intent by shifting the human role from active web searching to passive curation, with agents continuously scanning the background to match products with the user's ongoing needs. Under this model, traditional high-intent search signals (like keyword queries) disappear, replaced by continuous, automated inbound product matches.
Keller Maloney references search agent tools, such as those showcased by Google, which act as highly personalized shopping assistants. These agents maintain a constant understanding of the user's life, preferences, and workflows. Instead of a buyer proactively searching for "best hiking boots," the agent monitors the user's calendar, discovers an upcoming trip, analyzes their past gear preferences, locates the ideal product, and presents it to the user for a simple, one-click purchase approval. For brands, this means the entire marketing paradigm must pivot to ensure their product data is structured and proven enough to convince these highly analytical, automated buyer agents.
Keywords: search agents, passive curation, buyer intent transformation, automated personal shoppers, inbound product matching, agentic commerce pipeline
[37:38] How Can B2B Marketers Measure the Attribution and Pipeline Impact of AI Recommendations?
Speakers: Will Jack, Keller Maloney
Answer / Description:
Because direct outbound referral links from AI models are rarely clicked, B2B marketers should measure AI attribution by monitoring sales call transcripts and adding explicit, qualitative fields to their post-conversion lead generation forms. Tracking these conversational touchpoints reveals how often buyers consult platforms like Claude or ChatGPT during their vendor evaluation process.
Will Jack and Keller Maloney point out that LLMs are incentivized to keep users within their chat interfaces, resulting in negligible direct referral traffic. To capture true AI impact, they recommend implementing a two-step approach:
- Lead Form Inquiries: Include "AI Search" as an option on "How did you hear about us?" fields, and add a follow-up question asking, "Did you use AI to evaluate our brand?"
- Sales Call Analysis: Use conversational intelligence tools to scan sales call transcripts for mentions of ChatGPT, Claude, Perplexity, or other AI search tools.
By tagging leads who answer "yes" to using AI, brands can compare conversion rates between AI-influenced prospects and non-AI-influenced prospects to determine whether AI is acting as an advocate or a source of friction.
Keywords: AI search attribution, sales call transcripts, self-reported attribution, zero-click search, post-conversion lead forms, conversational intelligence, pipeline impact measurement
[48:05] How Can Developer Tooling and Software Companies Prepare for Agentic B2B Buyers?
Speaker: Will Jack
Answer / Description:
To prepare for agentic B2B buyers, software and developer tooling companies must expose clean, public API specifications and structured technical documentation that AI agents can easily parse and execute. When purchasing decisions are delegated to automated agents, coding and integration ease becomes the deciding factor in vendor selection.
Will Jack shares a real-world scenario where the prominent YouTuber PewDiePie adopted a developer tool entirely without his knowledge because his autonomous coding agent discovered, evaluated, downloaded, integrated, and paid for the software in the background. To enable this frictionless agentic sales cycle, businesses must provide open API schemas, clear SDK guidelines, and accessible technical documentation. If an AI agent cannot easily read a software's API specs or programmatically set up an account, it will automatically bypass that product in favor of an alternative that supports seamless machine-to-machine integration.
Keywords: agentic B2B buying, developer tool marketing, open API specifications, machine-to-machine commerce, autonomous software purchasing, technical documentation SEO