Happy Tuesday!
This week, I sat down with Bernard Huang (founder of Clearscope) to talk about something that's been keeping a lot of us up at night: How do you actually know if your AEO efforts are working?
Spoiler: It's messy. But Bernard gave us a clear-eyed framework for navigating it.
1/ Attribution in AI Search is Only Getting Worse
Here's the problem: When ChatGPT mentions your brand and a consumer Googles you afterward, that visit gets credited to search, not AI. The referral path is invisible.
Bernard's take: Between AI-driven dark traffic, cookie deprecation, and data privacy laws tightening everywhere - we should expect attribution to get murkier, not clearer.
For now, he recommends tracking three directional signals: 1. Branded search impressions going up 2. Brand mentions in AI responses 3. AI citations (when models source your pages)
2/ The 3-Layer Model for AI Brand Reputation
Bernard broke down how LLMs form opinions about your brand across three layers:
1. Training data: You were either in the snapshot or you weren't. 2. Grounding / Validation: This is where the model does live web searches to augment its answers. This is the layer you can influence right now. 3. Memory & Context: What the model knows about that specific user.
Layer two is where all of us should be focused.
3/ Bernard Called Our ORM Playbook "The Most Solid He's Ever Heard"
Here's what we're doing at Jolly right now for LLM reputation management:
1. Query ChatGPT (and other models) on our most important brand terms and commercial keywords. 2. Pull the cited "Sources" from the side panel. 3. Prioritize outreach to the sources that appear most frequently.
Bernard added a key upgrade: Run your prompts at temperature zero via the API. That strips out variance and shows you which sources the model reliably, consistently grounds its answers in.
4/ Content Strategy in an AI-First World
Both topic clusters and comprehensive hub pages still matter, but for different reasons:
1. Temperature-zero web searches = head terms. Build comprehensive, high-quality pieces targeting these. 2. Long-tail, low-frequency web searches = still worth targeting individually.
The goal is domain surface area.
5/ Reddit's Future Depends on One Thing
Reddit's value to LLMs lives or dies based on whether Reddit can hold its ground as a human-first platform.
My take: Reddit remains a meaningful signal in both Google and AI Search through 2026, and likely 2027 barring a major licensing shakeup.
The Big Unlock: Stop Fighting for Traffic, Start Fighting for Mentions
If you're an in-house SEO, and leadership is demanding you "get the traffic back" - that's the wrong conversation. The traffic isn't coming back.
What replaces it? Brand mentions, AI citations, and branded search queries.
Until next week, Greg