What the CARL Lab Data and a University of Richmond Classroom Taught Me About AI Search Optimization Strategy

We have been tracking AI search visibility since 2023. Here is what 156,000 data points showed — and what working with real brands in a classroom revealed about why most strategies stall before they succeed.

The research grew out of a practical problem. I was spending close to $2,000 a month on third-party tools to track AI search visibility for clients, and none of them gave me what I actually needed. So I built my own. That became CARL Intelligence — the prompt tracking layer inside the CARL suite — and once daily data started flowing across hundreds of queries and multiple large language models, the patterns that emerged demanded a more rigorous response.

That is how the CARL Lab was born. A formalization of something that was already happening: a disciplined effort to understand the mechanics behind outcomes that most practitioners were still treating as guesswork.

Courtney Turrin — who leads the research and brings a background in conservation biology at William & Mary and neuroscience at Yale — designs our studies, manages the data at scale, and keeps us honest about the difference between findings and opinions. This is a research program with a methodology, and its findings are heading toward an SSRN preprint.

Then I took the body of work into a classroom at the University of Richmond.

What 156,000 Data Points Confirmed About AI Search Inclusion

The CARL Lab dataset spans seven clients, three AI platforms, and 505 unique tracked prompts over 60 days. The most significant finding to date:

• 0.976 — Spearman correlation between semantic centroid distance and AI inclusion rates

• 156K+ — prompt rows analyzed across ChatGPT, Claude, and Perplexity

• 23 — additional hypotheses currently in active analysis

A Spearman correlation of 0.976 is exceptionally strong. Content that drifts from its primary topic — what we call high semantic centroid distance — is systematically less likely to be cited by AI systems. The correlation held across clients, platforms, and industries. One finding. It changed how we write everything.

The concept belongs to a broader framework we call Synthetic Epistemology — a working vocabulary of 24 terms that describe how large language models evaluate, weight, and surface content. These are not theoretical constructs. They emerged from the data and are being tested against it.

The CARL Model — our living ruleset for AI search content strategy — updates as new findings come in.

What Teaching Practitioners Revealed About Strategy Execution

Teaching Building Brand Authority in an AI-Driven Search World at the University of Richmond was where the research met reality.

The course is built around a peer learning model — no fictional case studies, no sanitized examples. Participants work with their own brands, their own AI search results, pulled live in the room. When you put marketers from completely different industries through the same exercise simultaneously, you stop seeing isolated problems and start seeing structural ones.

The pattern that kept surfacing had nothing to do with technical execution. It was behavioral.

Most brands are losing in AI search because they abandon a correct strategy before it has time to work.

One participant discovered the AI models could not distinguish his commercial services from residential competitors in his category. His strategy was correct. His content was consistent. The AI simply had not separated the signal yet — and his first instinct was to scrap everything and start over.

Another ran her brand search in incognito for the first time. Logged in, she was seeing a favorable, personalized result. Cold, incognito — the picture was different. She had been measuring the wrong baseline and drawing false confidence from it.

Both confirmed what the CARL Lab data had already indicated: the six-month window between first publish and meaningful AI traction is real. Brands that hold their strategy through it hold their positions on the other side. Brands that pivot at month three or four reset the clock.

Jenna Pace, Senior Strategist at Xponent21, described it well: AI is a mirror. It reflects back the signal you have built — focused or scattered, attributed or anonymous, maintained or neglected. The mirror does not lie. And it does not rush.

The Content Fundamentals That AI Search Data Keeps Confirming

Across both the lab data and the classroom, the pattern is consistent. The brands winning in AI search are doing disciplined marketing — precisely, over a sustained period. Specifically:

1. The winners define a niche and stay in it. High semantic centroid distance — content that wanders across too many subjects — correlates directly with lower inclusion rates. Depth on a focused topic outperforms breadth across many topics every time.

2. They attribute content to real, credentialed people. Anonymous content and generic bylines are invisible to AI authority signals. Named authors with verifiable expertise — linked profiles, published credentials, consistent positioning — register differently.

3. They answer the questions buyers actually ask. Our Most Valuable Query (MVQ) framework tracks these across LLMs specifically because they differ from traditional keyword lists.

4. They publish on a cadence and hold it. Consistency is a trust signal. So are gaps. Content freshness decay begins around 90 days — a skyscraper article that has not been reviewed becomes a liability faster than most brands realize.

5. Declining content is a refresh target, not a retirement candidate. A formerly strong article that has lost impression share is a priority optimization target. Declining visibility is a signal to refresh, not abandon.

6. They show up across multiple touchpoints. Google, LinkedIn, Maps, directories — frequency across platforms compounds the trust signal faster than depth in any single channel. LinkedIn, YouTube, and Reddit are among the top AI training and citation sources. That is a research-backed visibility strategy, not a social media recommendation.

AI raised the stakes for quality and removed the tolerance for thin content and manufactured authority signals. It is harder for bad marketers. For precise, consistent ones, the fundamentals still win — they require more discipline and more patience.

How Long Does AI Search Optimization Actually Take?

CARL Lab data suggests approximately six months from first publish to meaningful AI traction for a focused content strategy. Six months of consistent publishing, cluster-building, and content refreshing while the models index, evaluate, and begin treating a brand as a reliable source.

One additional finding: AI answers are personalized. The brand showing up consistently for a target topic is winning because of accumulated signal depth — many pieces, many platforms, many months. The incognito result — not the logged-in result — is the only honest baseline for strategy purposes.

Frequently Asked Questions About AI Search Visibility Strategy

How is AI search visibility different from traditional SEO rankings?

Traditional SEO produces a ranked list where position one is the most valuable. AI search produces a personalized, synthesized answer — there is no position one. The goal shifts from ranking to inclusion: being cited or referenced as a trusted source within the generated response. Inclusion frequency across tracked queries is the metric that replaces rank position.

How do you accurately measure whether your brand appears in AI search results?

Prompt tracking — running a consistent set of target questions through multiple large language models daily, in incognito mode with no logged-in accounts — is the only reliable method. Logged-in results are personalized and will overstate your visibility to cold prospects. CARL Intelligence automates this process. The incognito result is your true baseline.

Does publishing more content improve AI search visibility?

Volume alone does not produce inclusion. The CARL Lab''s semantic centroid distance finding (Spearman 0.976) confirms that content staying tightly focused on its primary topic outperforms high-volume content covering too many subjects. Quality, topical depth, and expert attribution matter more than output quantity.

Why does AI show different results for my brand depending on who searches?

AI models personalize answers based on search history, location, account data, and behavioral signals. A logged-in user who has previously engaged with your content sees a more favorable result than a cold prospect running the same search in incognito mode. Strategy built on the logged-in result is strategy built on a distorted baseline. Always measure incognito.

What is the CARL Model and how does it relate to AI search strategy?

The CARL Model is a living set of evidence-backed rules for content writing, optimization, and brand knowledge structure — maintained through ongoing research at the CARL Lab. It covers how to write for AI inclusion, how to optimize existing content, and how to build a brand knowledge base that produces coherent authority signals across all published content. As new research findings come in, the model updates.

What This Means for Discover AIO Members, and Where the Research Goes Next

Discover AIO exists as the publication layer for this research program. When findings emerge from the CARL Lab, they publish here first. When the Synthetic Epistemology framework is formally submitted to SSRN, the full context will be here. When the CARL Model updates, the reasoning will be documented here.

Every recommendation in this community — on content structure, publishing cadence, topic focus, authority signaling — is drawn from data we have run, tested, and interrogated. That is the architecture we have built.

The next cohort of Building Brand Authority in an AI-Driven Search World at the University of Richmond begins September 22, 2026. If you are in the Richmond, Virginia area and want to work through this material with your own brand as the live case study, registration is open now.

Stay in it. The mirror will catch up.

Will Melton is the Founder and CEO of Xponent21, a Richmond, Virginia digital marketing and AI optimization agency founded in 2015. He has been tracking AI citation strategy since August 2024 and has maintained consistent citations across Google AI Overviews, Claude, ChatGPT, and Perplexity throughout 2025. He teaches Building Brand Authority in an AI-Driven Search World as a continuing education course at the University of Richmond — next cohort begins September 22, 2026. He is the creator of the CARL suite and the originator of the Synthetic Epistemology framework, developed through ongoing large language model behavior research beginning in 2023. Discover AIO is his community platform for practitioners building expertise at the intersection of AI and search.