From SEO to HEO: What a 158% Jump in LLM Traffic Taught Me About Optimizing for Humans

A few months ago, I started noticing something in our website analytics more often than before: sessions arriving from chatgpt.com. Then gemini.google.com. Then claude.ai and perplexity.ai. There was a pattern.

By the numbers: in the year since, sessions referred from LLMs to CTL's site increased 158%.

That's not a vanity metric. Every one of those sessions is someone who asked an AI a question about tech transfer, licensing, or startup formation, and the AI decided our site was the answer worth serving up. That's a new front door to the organization, and what led me to prioritize it in my future planning.

Here's the framework I used to think about it, and what I actually did.

The framework: SEO → AEO → GEO → HEO*

I think about this as four overlapping layers, not four separate disciplines:

• SEO (Search Engine Optimization): the foundation. Structure, crawlability, keyword relevance. Still necessary. Not sufficient anymore.

• AEO (Answer Engine Optimization): writing so the answer is extractable. Featured snippets, voice search, "position zero." This is where content stops being a page to browse and starts being a passage to lift.

• GEO (Generative Engine Optimization): optimizing specifically for LLMs. ChatGPT, Gemini, Claude, Perplexity. These systems synthesize pages. If your content isn't structured for extraction and citation, it doesn't exist to them.

• HEO (Human Experience Optimization): the layer that ties it together. None of these matters if the answer doesn't actually serve the person on the other end, whether that person arrived via a search bar or a chat window.

That last layer is the one I keep coming back to. It's tempting to treat AEO and GEO as a technical arms race against algorithms. But the sites that win are the ones that make it genuinely easy to understand what something does and why it matters. AI systems are, in that sense, just very literal readers. Write for the human, structure for the machine, and the citations follow.

* Credits to Stephanie Swanson, who provided the framework on LinkedIn.

What I actually did

Two changes drove most of this:

Structured data and schema markup. I created new pages with the framework in mind, audited CTL's core program pages and implemented schema markup so the underlying content — what a program does, who it's for, how to engage — is legible to machines, not just to a human scanning the page. This is the unglamorous, back-end work that almost never gets credit, but it's what lets an LLM parse a page confidently enough to cite it.

Answer-first content blocks. I rewrote key pages so the direct answer — the definition, the process, the "what is this" — sits at the top, in plain language, before the narrative and nuance. This is a real shift from how most institutional content gets written, where context comes first and the point comes last. LLMs (and, frankly, busy humans) don't want to excavate for the answer. They want it up front, with the depth available for anyone who wants to keep reading.

Neither of these was a full site relaunch. They were deliberate edits to existing pages, layered onto work I'd already been doing on accessibility and organic SEO. That's part of the point: this isn't a separate initiative bolted onto your content strategy. It's the same content discipline, extended one layer further.

A few examples of where it's showing up:

Our annual report content, which I built out in January, is now one of our strongest LLM landing points — right behind the homepage. It's dense, factual, and structured for scanning, which turns out to be exactly what these systems want to lift from.

Our program pages built around comparison tables and step-by-step how-to information have also performed well. That's not a coincidence — tables and sequential steps are some of the most extractable content formats there are. If a human can scan it in five seconds, an LLM can usually parse it in less.

Not every page has caught up at the same pace, and that's an honest, useful data point too. One of our flagship innovation programs has only basic markup so far and hasn't seen the same LLM traction, even as the program itself has gained real momentum. We published a four-year impact report on it in January, which likely drove a good amount of the renewed attention it's getting. Recognition and technical readiness don't always move on the same timeline, and this is a reminder that the structural work still has to catch up to where the program already is.

What I'm still working through

I'll be honest about what this data can and can't tell me. A 158% increase off a small base is a meaningful signal, not a mature channel yet. I don't have a clean way to isolate how much of the lift is the schema and answer-block changes versus the simple fact that more people are using LLMs to research vendors and partners than they were a year ago. Both are almost certainly true at once.

But there's one number that reassures me the effort is pointed in the right direction: it's not just that LLMs are citing us more, it's what happens after. Visitors who arrive via an LLM stay engaged an average of 42 seconds per session, compared to 28 seconds sitewide — 52% longer — and their engagement rate runs 56%, against 38% site-wide. A citation only means the AI thought the page was worth surfacing. Those engagement numbers tell me the page was actually worth reading once someone got there, which is the harder thing to earn and the one that matters more.

The broader lesson, for me, isn't really about chasing another acronym. It's that every layer of this — SEO, AEO, GEO — eventually resolves back to the same question: are you making it easy for someone (or something reading on someone's behalf) to understand what you do? Get that right, and the visibility takes care of itself, in whichever interface people happen to be asking from next.

I expect the tools and terminology here will keep shifting. The discipline underneath — clarity, structure, answering the actual question — won't.