// FINDING_1

Finding 1: The Winners Cite Their Sources, Aggressively

The cleanest separator we found is how much a page links out to other websites.

The median page in the Both group carries 56 external links. The median AI-only page carries 29. The median organic-only page, 28. The pages that win both systems link out at roughly twice the rate of everyone else, and it isn't a couple of freak pages dragging the average. The whole distribution shifts.

Classic SEO doctrine treats an outbound link as leakage. Every link out is authority you're handing to a stranger. The pages winning both systems behave the opposite way. They read like research documents. A Mayo Clinic condition page cites its medical sources. A NerdWallet comparison links to every provider it reviews. An IRS guidance page links across the rest of IRS.gov and out to the regulations it rests on.

There's a plain reading of this. To a machine deciding what to quote, a page that shows its sources looks like evidence. To Google, that same page earns links and satisfies the reader who lands on it. Citing generously correlates with winning in both places at once.

// HONEST_PART

The Honest Part: Where the Both Group Actually Lives

Before this turns into a playbook, two facts keep it grounded.

First, the Both group isn't spread evenly across query types. A third of it sits in YMYL queries (health, finance, safety), and another 30% in informational ones. In local queries ("dentist near me," "coffee shop near me") it barely exists, at 5%. If your business runs on local intent, the win-both pattern is close to unavailable to you. Google answers local with the map pack, the AI engines cite directories, and the two systems almost never land on the same URL.

Second, we checked whether the outbound-link gap is just a category trick. Maybe Both pages link out more only because health and finance pages always do. So we held the category constant and re-measured. Inside informational queries alone, the median Both page has 52 external links against 21 for AI-only. Inside YMYL alone, 58 against 33. The gap survives. It belongs to the group, not to the topics the group happens to sit in.

There's one caveat we can't scrub out. The 985 Both URLs come from only 531 domains, and the biggest of them are exactly who you'd guess: Cleveland Clinic, Mayo Clinic, Wikipedia, IRS.gov, Fidelity. Institutionally credible, well-engineered sites. Some part of "the Both group has better structure" is really "the Both group is made of better-built websites." Copying their HTML won't transplant their reputation. What the data gives you is the profile of a winner, not a causal recipe for becoming one.

// METHODOLOGY

Methodology

Dataset. 500 queries, 100 in each of five intent categories (informational, commercial, transactional, YMYL, local), US market, English. For each query we captured Google's organic top 10, Google AI Overview citations, ChatGPT citations, and Perplexity citations, deduplicated to 8,397 unique URLs, then crawled all of them and saved raw HTML.

Analysis sample. 6,576 URLs with HTTP status 200 and parseable HTML. Structure signals (headings, lists, links, semantic tags) come from our own parser over the saved HTML; schema signals come from the crawl's structured-data extraction.

Metric choices. We report medians for count metrics because the distributions are heavy-tailed (one page carries 1,693 external links; means overstate the typical page). Percentages are shares of each group.