How AI Search Engines Actually Decide What to Cite

Most “AI SEO” content right now is speculation dressed up as certainty. I wanted actual signal, so I tracked the same 40 long-tail queries across Google, Perplexity, and ChatGPT search every few days for three weeks and logged which pages got cited.

What I actually measured

For each query I recorded:

• Whether the page ranking #1 in Google also got cited by the AI answer
• How old the cited page was
• Whether the cited page had a direct, quotable answer in the first 150 words

What correlated with getting cited

Pages with a direct answer near the top outperformed pages that opened with scene-setting or a story. This is a fairly boring, obvious-in-hindsight finding, but the count made it obvious in the data.

// crude version of the check I ran against each result
function hasDirectAnswer(html, query) {
  const firstParagraph = extractFirstParagraph(html);
  return firstParagraph.length < 400 && mentionsAnswerShape(firstParagraph, query);
}

What didn’t correlate the way I expected

Backlink count, in this sample, mattered less than I assumed going in. That doesn’t mean links don’t matter for classic Google ranking — it means the AI citation layer seems to be doing something closer to “which page answers this most directly and recently,” not “which page has the most authority.”

Practical takeaway

Write the direct answer first. Save the story and the caveats for after. That’s a small thing that’s easy to test on your own posts.