The thing I keep coming back to after this week's episode is how much of the LLM optimization conversation is really just a second chance to do entity SEO right. We had this conversation in 2013 when Hummingbird launched. We had it again when BERT arrived. Now we're having it again because AI Overviews exist. The tools are different. The stakes are higher. The underlying problem — you haven't told machines clearly who you are and what you do — hasn't changed.

This week I sat down with Grant Simmons — SEO practitioner since the mid-90s, former VP of Search at Homes.com (where he led recovery from a 60% traffic loss), and currently Head of SEO and GEO at Fiat Growth in San Francisco. Grant is one of the sharpest thinkers I've had on the show about what entity SEO actually means in practice. Here's what stood out.

LLM hallucinations are a content architecture problem

Grant's framing is clean: LLMs make stuff up about your brand because you haven't filled the gaps on your site. You haven't explained who you are, what you do, who you serve, where you serve them — clearly and unambiguously. The EntityMap.org standard (just launched by Dixon Jones and Fred Loral at Waikay) is an attempt to formalize a 'meaning layer' for key pages: associations, connections, disambiguation. It's not a description of the page. It's a map of what the page means and how it connects.

Schema is still worth doing — here's why

The mid-surface schema case holds: Google uses it for ontology categorization. LLMs that ground on Google inherit that. RAG in LLMs is largely a wrapper for Google results today. So what helps Google helps the LLM layer. There's a limit to the ROI of deeply nested schema — but at mid-surface depth, the lift is low enough that the 'why not' wins.

We're beta testing the beta testers

Grant named something the industry needs to hear directly: most new LLM SEO tools being shipped are shite, and the builders know it. We're shipping for feedback. The right posture is to not outsource uncertainty to LLMs — treat their output as the answer they gave you, not the answer you wanted. First-party data (log files, first-party pixels) is where high-confidence signals actually live.

Reporting is the biggest unresolved agency problem

This was the sharpest practical point in the episode. The gap between insights (SEOgets, GA4) and KPI-connected narratives that justify SEO investment — that's what no one has solved yet. GA4 is Google's punishment for the industry. Looker Studio is the second punishment. Grant's tools at appslicer.com (including an AI Overview impact estimator and SEO attribution work) are trying to close this gap. The conversation about the 'degapinator' — triangulating Ahrefs + GA4 + GSC to reconstruct a query click map — is one every SEO should be having.

“We will learn more by seeing outcomes than we can by listening to Google.”

— Grant Simmons