Wil Reynolds, founder of Seer Interactive, shares how losing 80% of organic traffic actually revealed that his team had been tracking the wrong metrics for years. We get into why AI visibility is a vanity metric, how 44% of LLM users include brand names in their prompts, the real security risks of wrong phone numbers in AI answers, and why trust (not visibility) is the only moat that matters.Chapters00:00 - 80% traffic drop, pipeline went up06:43 - The pressure of experimental channels with real budgets08:28 - AI visibility is a vanity metric13:49 - People are tracking the wrong prompts17:15 - AI is getting your phone number wrong (and scammers know it)22:30 - Trust vs visibility: the real optimization target30:26 - Trust is a moat, hacks mortgage your brand32:20 - Be seen, be believed, be chosen44:15 - What digital marketers should do right now49:33 - Where to find WilKey TakeawaysTraffic is no longer the leading indicator - Seer lost 80% of their organic traffic over two years. Pipeline went up. Social converts at 5x organic. The correlation between search traffic and revenue has broken for many businessesAI visibility is a vanity metric - When ChatGPT doubles the length of an answer, your "visibility" goes up without any more humans seeing you. If visibility grows but leads don't, you're the sucker. Track visibility against your pipeline, not in isolation44% of LLM users put brand names in their prompts - Wil's team watched real humans use LLMs and found nearly half include specific brands. Head-to-head brand comparisons are the prompts worth tracking, not generic category queriesTrust is the only real moat - Michelin built a restaurant guide in 1900 that still drives foot traffic and pricing power. RAMP launched 52 AI-generated restaurant pages in a day. One is rotisserie chicken made by a chef. The other is a chicken nugget. Both are chicken, but only one builds trustBe seen, be believed, be chosen - Visibility gets you in the room. But if people Google your team and nobody shares their content, if you're not speaking at conferences, if your newsletter has no engaged readers, belief falls apart. Trust transfers from people, not listiclesConcepts DiscussedThree Types of AI Search | Search-led (web index + AI), answer-led (hybrid with tool use), and fully generative (training data only). Each requires different optimization.Training Data Lag | Gemini 3.1 launched with training data from January 2025, a 16-month gap. Work done since then has zero provable impact on training-data-based answers.Brand-in-Prompt Behavior | 44% of observed LLM users include brand names in their prompts. Changes the optimization target from "show up for generic queries" to "win head-to-head comparisons".Phone Number Hallucination | LLMs serve wrong phone numbers for businesses, creating fraud exposure. Especially dangerous for financial services targeting elderly customers.Recommended Stack Visibility | AI coding tools recommend tech stacks from training data. Being the default recommendation in Claude Code or Codex creates durable, hard-to-displace visibility.Connect with Wil ReynoldsLinkedIn: linkedin.com/in/wilreynoldsLinkedIn Newsletter: Thinking Out LoudSeer Interactive: seerinteractive.comNo Hacks is a podcast about web performance, technical SEO, and the agentic web. Hosted by Slobodan "Sani" Manic.