Ask an LLM about a well-known company and it answers fast, and usually gets it right. Ask something obscure and it can be just as confident and completely wrong. This episode traces that behavior back 40 years, to two rival ideas about how to make a machine understand the world, and shows how their recent fusion is what actually answers questions about your company in AI Search today.

What you’ll hear in this episode:

• Why hallucinations run on the exact same machinery that produces correct answers
• The hand-built approach: Cyc, triples, the semantic web, and how Freebase became Google’s Knowledge Graph
• The pattern-learning approach, and how Word2Vec turned a word’s meaning into math you can do arithmetic on
• How the 2017 Transformer and “attention” let a model read a whole sentence in context, and how next-word prediction trained GPT-3
• Why RAG bolts the two systems together, and what it means that you can edit the “knower” but only earn the “talker”
  

Chapter Headings

[00:02:09] Why an LLM can be confidently wrong

[00:05:45] Teaching machines by hand: Cyc and the 1980s

[00:07:18] Triples, the semantic web, and Freebase

[00:08:51] Google's Knowledge Graph: things not strings

[00:14:08] The other approach: patterns, not facts

[00:16:00] Word2Vec turns meaning into math

[00:18:05] Attention and the Transformer

[00:23:07] BERT, GPT-3, and next-word prediction

[00:26:25] Why hallucinations happen

[00:28:44] Bolting the two brains together with RAG

[00:35:42] The takeaway: the knower you edit, the talker you earn

Resources

Cyc Project (1984) — https://www.cycorp.com/ | https://en.wikipedia.org/wiki/Cyc

Freebase (2007–2016) — https://en.wikipedia.org/wiki/Freebas...)

Wikidata — https://www.wikidata.org/ | https://www.wikidata.org/wiki/Wikidat...

Google Knowledge Graph (2012) — https://blog.google/products/search/i... | https://support.google.com/knowledgep...

Word2Vec (2013) — https://arxiv.org/abs/1301.3781

"Attention Is All You Need" (2017) — https://arxiv.org/abs/1706.03762

BERT (2018) — https://arxiv.org/abs/1810.04805

Retrieval Augmented Generation (2020) — https://arxiv.org/abs/2005.11401

GPT-3 (2020) — https://openai.com/api/

Find the transcript for this episode and additional resources on our blog: https://victorious.com/podcast/history-of-ai-search/

Every marketing team has a search dashboard. But that dashboard has a structural blind spot: it can’t see what’s happening in AI search, and it won’t flag that anything is missing. When a buyer finds your brand through ChatGPT, Gemini, Claude, or Perplexity, that visit typically lands as direct traffic with no source attribution. There’s no indication AI was involved and nothing in your reporting to suggest a gap is there.

What you’ll hear in this episode:

• Why most AI-driven sessions land in your direct traffic bucket, including why Google’s May 2026 GA4 patch only covers part of the problem
• How AI Overviews are compressing click-through rates by 58% on top-ranking queries, so session counts decline while rankings hold steady and your report gives no explanation
• Why a brand appearing consistently in AI responses can see almost nothing in its referral data
• The case for building measurement infrastructure now, when AI traffic is 0.2% of total visits but growing at 623% year over year, and you can’t retroactively create a baseline
• Four signals that together give you a credible picture of AI visibility: AI brand share of voice per platform, branded search volume trend, a GA4 AI referral segment, and self-reported attribution on every lead form

UK Competition and Markets Authority, “Further CMA action to secure a fairer deal for businesses and improve Google Search services in UK,” June 2026, https://www.gov.uk/government/news/further-cma-action-to-secure-a-fairer-deal-for-businesses-and-improve-google-search-services-in-uk

Similarweb, “The Downstream Impact of AI Visibility,” June 2026, https://www.similarweb.com/corp/the-downstream-impact-of-ai-visibility/

Ahrefs, “What Is a Good Organic CTR? Real Website Benchmarks (June 2026),” June 2026, https://ahrefs.com/blog/what-is-a-good-ctr/

Ahrefs, “Update: AI Overviews Reduce Clicks by 58%,” February 2026, https://ahrefs.com/blog/ai-overviews-reduce-clicks-update/

Semrush, “Why Category Entry Points Belong in Every AI Search Strategy,” June 2026, https://www.semrush.com/blog/category-entry-points-ai-search/

Loamly, “The AI Traffic Attribution Crisis: Why Your Analytics Are Wrong,” February 2026, https://www.loamly.ai/blog/ai-traffic-attribution-crisis

Contentsquare, “2026 Digital Experience Benchmark: AI-Referred Traffic,” April 2026, https://contentsquare.com/blog/ai-referred-traffic/

Find a complete list of resources at: https://victorious.com/podcast/how-to-tell-if-ai-search-is-working/

As AI platforms reshape the search landscape, many marketers are falling back on an old, unreliable playbook: chasing individual keyword ranks. In this second episode of The Search Signal, Michael Transon explains why a single snapshot of an AI search response is just noise—and why you need to update your tracking strategy. He breaks down how platforms like ChatGPT, Copilot, and Perplexity source their information differently, gives a hands-on blueprint for auditing your brand visibility, and shares the tactical levers required to earn a compounding share of voice.

What you'll hear:

• Why tracking AI like traditional keyword ranking fails, and why a sustainable approach means measuring a collective share of voice percentage instead.
• The distinct ways Perplexity, Copilot, and ChatGPT index and source their answers, and why performance on one engine doesn’t translate to another.
• How to conduct a rigorous manual audit in incognito mode to establish an accurate baseline and map out your competitors' visibility.
• The differences between citation rate and mention rate, and the unique, high-intent roles they play in the conversion process.
• The three pillars for moving the needle on your presence: digital PR outreach, structured entity signals, and answer-first commercial page optimizations.

All sources referenced in this episode are linked below.

Sources

• The Search Signal, Episode 1: https://victorious.com/podcast/ai-search-right-now/
  Ahrefs, "AI Overviews vs. AI Mode" https://ahrefs.com/blog/ai-overviews-study/
• SparkToro, "Search Happens Everywhere," https://sparktoro.com/blog/new-research-search-happens-everywhere-an-analysis-of-41-websites-with-significant-search-activity/
• Google, "Optimizing Your Website for Generative AI Features on Google Search," https://developers.google.com/search/docs/appearance/ai-overviews
• iPullRank, "Google Algorithm Leak Analysis," https://ipullrank.com/google-algo-leak
• The Growth Memo, "Users Behave Differently in AI Overviews," https://www.growth-memo.com/p/users-behave-differently-in-ai-overviews
• Adobe, "AI-Referred Traffic Conversion Analysis," https://business.adobe.com/blog/the-latest/adobe-digital-insights-ai-traffic-grows-significantly-during-2024-holiday-shopping-season
• SISTRIX, "Position One Click-Through Rate Data, March 2026," https://www.sistrix.com/blog/

Complete list of resources at: https://victorious.com/podcast/ai-search-right-now/