“We need to do something about AI search” — that order has started landing on a lot of marketing desks lately. Ask ChatGPT or Perplexity about your category, and you’d like your name to come up. Fair enough; I’d want the same.
So the question becomes: what do you actually do? And the reflex answer is “treat it like SEO.” But that reflex is where people go wrong, because AI search has different taste in content than Google does. A lot of the old tricks just don’t land.
There’s a study that took that question head-on, early and seriously. Let’s start there.
What GEO Even Means
GEO is Generative Engine Optimization (ie getting your content cited or referenced inside an AI’s answer, rather than ranked in a list of links). The acronym is clunky, but think of it as a cousin of SEO.
SEO chases the top of Google’s results page. GEO chases being the source the AI reaches for when it writes its reply — on ChatGPT, Perplexity, and the rest. The goal slid from rank to citation.
The Study That Put 40% On It
A Princeton-led team coined the term and presented it at KDD 2024 (one of the top data-mining conferences). So this isn’t a vibe somebody had on LinkedIn.
They built “GEO-bench” — a big pile of real user questions across a lot of different domains — and then ran the tedious comparison nobody wants to run by hand: tweak the content this way, that way, and see which tweaks made the AI more likely to cite it.
The result: certain tweaks lifted a piece of content’s visibility in AI answers by up to 40%. And the tweaks that worked were these:
- Back your claims with statistics and concrete numbers.
- Cite your sources clearly.
- Quote credible third-party voices.
Line those up and there’s nothing exotic about them. The winning move was just “make the writing more trustworthy and better-sourced.” It’s a deflating little conclusion, honestly. I’d half-expected some special technique, so the boring answer was a bit of a letdown — and also kind of a relief.
The Tactic That Flopped Is The Interesting Part
The more telling result is the thing that didn’t work. Keyword stuffing — the classic SEO move of cramming in every related term you can think of — was near-useless for getting cited.
Think of it like a résumé with ten certifications dangling off the top and nothing underneath. The AI looks past the keyword pile and asks “okay, but what’s actually here?”
And a second study points the same direction. A 2025 preprint from Chen and colleagues at Toronto found that AI search (ChatGPT, Perplexity) systematically favors third-party, authoritative, earned media over a brand’s own self-promotion. An outside expert saying it carries more weight than your own page saying the same thing.
So there are really two complementary moves here, not one. Tighten your own content with numbers and sources — that’s the first study. And get talked about in other people’s credible places — that’s what the second study rewards. You want both running, not one or the other.
The Catch: There’s No Silver Bullet
Before you rebuild every page, one honest caveat from the GEO research itself: effectiveness varies a lot by domain. What lifts citations in one vertical does much less in another, and there’s no universal trick that works everywhere.
So the right posture isn’t “apply the playbook and trust it.” It’s “measure in my own category and verify.” Treat the 40% as proof the levers exist, not as a number you’ll personally hit.
What To Actually Do Monday Morning
So here’s the takeaway. If you want to get cited by AI, don’t burn time on keyword tricks. Raise the trustworthiness of the writing — numbers, sources, third-party voices — and the citations tend to follow. Princeton’s experiments moved visibility by up to 40% on that alone.
In practice, don’t try to fix everything at once. Pick one existing page that matters — a key product or category page — and go through it asking “what’s the evidence for this claim?”, then shore it up with a real number and a real source. Prove the lift on one page before you scale the effort across the site.
And then watch whether it’s working. How much AI actually cites or mentions you isn’t something you can eyeball; it drifts, and a single check tells you nothing. You have to track it over time, which is exactly the gap HexScope fills — it’s an AI Brand Tracker that monitors how AI cites and mentions your brand month over month. Pair that with the own-content and third-party moves above, and you’ll start to see which lever is paying off and which one isn’t.
Maybe the boring answer is the whole answer. Write things worth citing, then go check whether you got cited.
Sources
