Welcome, humans.

Google just admitted they handle 8 trillion searches a year. But here's the thing—even a 1% AI error rate means 80 billion wrong answers.

And that's exactly what's happening right now.

SEO expert Mark Williams Cook just revealed to us how AI is fundamentally breaking search. When we looked up one of our own portfolio brands, for example, Google's AI overview was actually taking bad reviews from a completely different company with a similar name and applying them to us.

Now multiply that problem by the scale mentioned above.

We knew we had to get a handle on how AI is changing SEO, so we teamed up with Mark, who has been doing SEO for 22 years (he's survived 8 “deaths of SEO”), and he says what's happening now is different. The link graph that powers Google is starting to rot. And OpenAI is losing $10-15 billion this year trying to figure out how to make AI search profitable. 

In our latest podcast episode, Mark breaks down the reality of SEO in the AI age, including why he calls AI a “leaky bucket”, how how to approach SEO for your company in the age of AI:

Here’s some of our favorite moments:

• (3:00) The “leaky bucket“ problem: Why AI's nondeterministic outputs create systemic issues at Google's 8 trillion query scale “

• (8:55) AI conflating companies with similar names: Mark's example of how AI confused his agency “Candour“ with “Candour Talent“ recruitment company

• (17:24) The “BSing“ paper: Why hallucinations should be called what they really are

• (37:00) Why the link graph is rotting: As AI agents replace human browsing, 20+ years of Google's ranking signals deteriorate

• (54:54) Reddit: The new SEO battleground: “There's a party going on there. I'm going to climb in through the window“ isn’t the right strategy.

• (56:50) Google's true system goal: “Make ungodly amounts of money“ - not reward good content

• (63:30) “No robots in the cantina“: Why Mark draws terrible MS Paint images instead of using AI

• (28:41) The vegetative electron microscopy problem: How AI errors spread through academic papers

Watch / Listen now on YouTube | Spotify | Apple Podcasts

ICYMI: Kwasi Ankomah from SambaNova Systems just revealed the brutal math behind agentic inference (using agents in production): AI agents use 10-20x more tokens than regular chatbots. And if you're running that on traditional infrastructure, you're looking at $3M monthly bills that could be $100k instead.

The companies that figure out efficient inference will own the next generation of AI applications. And the energy math alone (10kW vs 100kW+) is about to reshape the entire industry.

Listen / Watch now on YouTube | Spotify | Apple Podcasts

One more thing… Mark predicts things will get “significantly worse before they get better.” Once spam gets into LLM training data, it's like “ink in water”… nearly impossible to remove.

There’s actually a great video on that you can watch here. The TL;DR:

• Kurzgesagt, a German animation channel with 24M subscribers making beautiful explainer videos about science, fed AI research tools a fake video topic about brown dwarfs.

• The AI returned solid info... plus 20% completely made-up facts delivered with the confidence of a kid who definitely didn't study for the test.

• One “credible source?” An AI-written article.

• Weeks later, they found a YouTube channel repeating those exact fake facts.

• Now there's a “real source: online for the next AI to cite, and the misinformation loop is complete, and honestly, we're all living in it now.

• While companies scramble through bureaucracy to fix brand impersonation, scammers can generate fake content faster than ever.

Resources from this episode:

• AlsoAsked - Mark's tool for understanding search queries.

• Candour - Mark’s agency.

• Search with Candour podcast (Spotify / Apple) - Mark's SEO podcast.

• OpenAI on why language models hallucinate (full paper).

• “Vegetative electron microscopy”, i.e. how AI errors spread.

More resources to check out:

• James Brockbank’s investigation in Search Engine Journal

• Seer Interactive’s test — “Footers are back like 2003” 

• Netcraft — how LLMs end up recommending phishing sites

P.S: Mark's advice for website owners drowning in conflicting guidance? Focus on three things: structured data, brand signals on every page, and monitoring for impersonation.

Stay curious,

The Neuron Team

P.P.S: Love the newsletter, but don’t want these new podcast announcement emails? Don’t unsubscribe — adjust your preferences to opt out of them here instead.