Everyone's talking about "optimizing for AI search" like it's some mysterious black box. It's not. AI models evaluate authority the same way humans do — they just do it faster and at scale.
Here's what I've learned from 14 months of tracking which of my pages get cited by LLMs and which don't.
The Authority Signal Stack
Think of LLM authority like a credit score. No single factor determines it, but all the factors reinforce each other:
Level 1: On-site signals. Schema markup (Article, Organization, FAQ), clean HTML structure, proper heading hierarchy, and clear author attribution with credentials. These are table stakes. Without them, nothing else matters.
Level 2: Citation network. How many other authoritative domains reference your content? Not backlinks in the SEO sense — I mean actual citations where your brand name or content URL appears in credible articles, research papers, and Wikipedia references.
Level 3: Entity consistency. Do different AI models agree on what your brand is? If ChatGPT thinks you're a "SEO tool" but Perplexity thinks you're a "content agency," you've got an entity resolution problem. Consistent entity descriptions across platforms matter more than most people realize.
Level 4: Temporal authority. Are you consistently publishing fresh content, or is your last update from 2023? LLMs heavily weight recency. A brand that publishes weekly will outrank a brand with better backlinks but stale content.
Level 5: Behavioral signals. Do users who land on your page stay there? Do they bounce immediately? LLMs don't directly measure this, but the search engines that feed them data absolutely do. Dwell time and engagement are the hidden signals.
The Hard Part Nobody Talks About
Building authority for humans and building authority for LLMs used to be two different strategies. They're converging.
The same things that make content persuasive to a human reader — original data, clear structure, named sources, logical argument flow — are exactly what LLMs weight when deciding whether to cite you.
If you're writing for LLMs at the expense of human readers, you're doing it wrong. The overlap is nearly 100% now.
Write something genuinely useful. Structure it well. Back it up with evidence. Publish consistently.
That's the whole strategy. Everything else is optimization theater.