Lesson 2: Why Buyer Personas Matter for AI Visibility

Why AI Search Recommends Different Brands to Different People (Lesson 2) AI search personalizes brand recommendations by persona, which means a single generic prompt tells you almost nothing about your real visibility. In Lesson 2, Nick Clark, Research Engineer at Gumshoe and PhD student at the University of Washington, walks through a live demo of Claude's stored memory, explains the four layers of personalization, and shows how persona-level visibility data doubles as brand positioning research. Why does AI search give different people different brand recommendations? Because users delegate information seeking to these systems rather than just retrieving links. To produce a good recommendation, the model needs a picture of your context and preferences. In the demo, a personal Claude account with stored memory returns niche, design-led cycling jersey brands matched to the user's taste. A brand new account with no memory, given the identical prompt, returns generic picks. Same question, different answer. What are the layers of personalization in AI search? Location. Geographic personalization applies even when logged out, similar to Google. Prompt wording. Real prompts don't look like SEO keywords. They look like "best trail running shoes for wet conditions on technical trails near Seattle," and that phrasing shapes which brands surface. Persistent memory and conversation history. Logged-in models save discrete facts across sessions and retrieve context from past chats. Mention a race near Portland once, ask about hotels weeks later, and the model connects them. Custom instructions. Explicit user preferences injected into every conversation, for example excluding UK brands to avoid international shipping. What is a persona in GEO? A semi-fictitious individual representing a target customer, with demographic and behavioral detail attached. Familiar territory for brand strategists, less so for technical SEOs. Frameworks like STP (segmentation, targeting, positioning) and VALS (values and lifestyles) apply, and Gumshoe builds personas during onboarding or imports existing ones a client already has. How much does visibility actually vary by persona? Gumshoe sees a standard deviation of roughly six percentage points across personas as a baseline. Lululemon shows a 64 percentage point spread between its highest and lowest visibility personas, with sustainability-oriented customers at the bottom. Brands that appear invisible on platforms using only generic prompts are often highly visible, just to a specific segment. Why is persona data also positioning research? AI is performing compression over the public internet, aggregating blog posts, Reddit threads, and third party reviews in your category. Ted Chiang's New Yorker piece describing ChatGPT as a blurry JPEG of the web captures the idea. Probing the model reveals which aspects of your brand are legible, which are out of date, and where it hallucinates. Lululemon's sustainability gap is a strategic finding, not a technical one. That level of market perception insight traditionally required focus groups and a dedicated research budget. Chapters 00:00 What this lesson covers 00:33 Why delegation drives personalization 01:38 Live demo: stored memory inside Claude 02:31 Personalized vs blank account, same prompt 03:45 Why generic prompts hide real visibility 04:05 The six point standard deviation across personas 04:18 Lululemon's 64 point persona spread 05:18 What a persona is and how to build one 06:00 STP, VALS, and building personas in Gumshoe 06:50 Layer 1 and 2: location and prompt wording 08:09 Layer 3: persistent memory and prior conversations 09:41 Layer 4: custom instructions 10:54 Why models sample from several distributions, not one 11:40 AI as compression over the internet 12:29 Ted Chiang and the blurry JPEG of the web 13:30 Replacing focus groups with persona analysis 14:39 Summary Instructor Nick Clark, Research Engineer at Gumshoe and PhD student at the University of Washington. His research includes a study of Reddit communities pushing the custom instructions interface to its limits, which informs how this course thinks about user-directed personalization. About this course A practical course on generative engine optimization from the research team at Gumshoe, where we measure AI visibility across models, personas, and market segments. Lesson 1 covers how AI search systems work and why GEO differs from SEO. Upcoming lessons cover technical site optimization for AI crawlers, content formats models favor, and citation measurement. Subscribe for Lesson 3. #GEO #GenerativeEngineOptimization #AEO #AISearch #AnswerEngineOptimization #BuyerPersonas #BrandPositioning #ChatGPT #AIVisibility #DigitalMarketing