If you do marketing, you’ve poured time and budget into SEO for years. Rank high on Google — that was the whole premise of getting found online. So it’s a little unsettling when the data starts saying the premise is moving under your feet.
People who pick up ChatGPT as a daily habit are running fewer Google searches. Not a hunch — a measured drop. London Business School and UCLA tracked the browsing behavior of 2,014 households across roughly 1.2 million URLs to pin down how much, and which kind.
A 20% Drop — But Not Across The Board
Households that used ChatGPT three weeks running cut their traditional online search volume by more than 20% within 20 weeks. That’s the headline. But the part that tells you what to actually do is which searches fell.
- Question-style searches (“what is X”, “how to X”) dropped a lot.
- Navigational searches (“Amazon”, “Rakuten” — the ones whose whole point is to land on a site you already know) barely moved.
- Brand-name searches held up the same way.
So ChatGPT is siphoning off the “I want to know something” queries, the exploratory ones. The “I want to go to that site” queries don’t care that ChatGPT exists.
And it doesn’t land evenly. Small sites lost a big chunk of traffic while large, established ones held up. Educational content took the worst of it — Stack Overflow visits fell, while Wikipedia and Reddit shrugged it off. If you’re a small site, it’s a bit like a giant department store moving in next door.
GEO Isn’t A Replacement For SEO. It’s A Different Game.
So if that traffic is draining, where does it go — and how do you play the new board? This is where “GEO” shows up.
GEO is Generative Engine Optimization (ie getting your content referenced inside the AI’s answer, rather than ranked in a list of links). Old SEO chased the top of Google’s results page. GEO chases being the source the AI reaches for when it writes its reply — on ChatGPT, Perplexity, Gemini.
A Princeton-led team coined the term and presented it at KDD 2024 (one of the top data-mining conferences). On their benchmark, applying GEO methods lifted a piece of content’s visibility in generative-engine answers by up to 40%. So this isn’t vibes; somebody put a number on it.
AI Search Has A Bias In Where It Looks
But “just do GEO” is too easy, because AI search has a quirk worth knowing. A 2025 study by Chen and colleagues at Toronto dug into how these engines actually pick their sources.
AI search systematically favors earned media — third-party, authoritative stuff — over your own site and your social posts. An industry write-up or an expert review gets picked up more readily than your own page saying the same thing. Google, by comparison, spreads its attention across owned, social, and earned fairly evenly.
And the engines don’t agree with each other. ChatGPT, Perplexity, and Gemini differ sharply in how much domain diversity they want, how fresh they like their sources, and how sensitive they are to the exact phrasing of a query. With SEO you were courting one engine. With GEO you’re courting several, each with its own taste.
Add the big-brand bias on top — AI leans toward names it already knows — and the game gets lumpy fast.
Even Downstream, It’s Still Early
One more piece of real data, because the funnel doesn’t end at the click. Kaiser and Schulze (Goethe University Frankfurt and Mannheim Business School) tracked 973 e-commerce sites — about $20 billion in combined revenue — for 12 months, weighing 50,000-plus ChatGPT referrals against 164 million transactions from the traditional channels.
ChatGPT’s conversion rate beat paid social, but still trailed organic search, email, and the other established channels. The fun wrinkle: the more complex the product category, the better ChatGPT referrals performed. So for the stuff people genuinely need to deliberate over, “ask the AI, then buy” is already working.
Conversion rates climbed steadily across the year — but a falling average order value kept eating into the overall gain. So it’s not “ChatGPT replaced Google” yet. It’s a clear growth trend that just hasn’t arrived.
The Move: Keep SEO, Split Defend From Migrate
As Wharton’s Stefano Puntoni argues in HBR, this was never “SEO is dead.” Search is splitting in two — the part where people still click through and read, and the part the AI answers on the spot.
So don’t torch your SEO. Sort it. Work out which of your content is the defendable kind (navigational, brand, the things people come to you for by name) and which is the exploratory kind that’s migrating into AI answers. For that second pile, rework it for the new game at once: cleaner FAQ structure, clear first-party sourcing, and the third-party mentions AI search rewards. And use the LBS/UCLA behavior data and the Princeton GEO work as your gauge — every quarter, look at which query types are actually slipping.
The teams that lose here are the ones treating this as one switch to flip. It isn’t. It’s a split to manage — and the split is the whole strategy.
Sources
- [R1] Lambrecht, A., Padilla, N. (London Business School) & Lam, H.T., Hollenbeck, B. (UCLA), “The Impact of LLM Adoption on Online User Behavior”, 2026. LBS
- [R2] Aggarwal, P., Murahari, V., Rajpurohit, T., Kalyan, A., Narasimhan, K. & Deshpande, A., “GEO: Generative Engine Optimization”, KDD 2024. arXiv
- [R3] Chen, M., Wang, X., Chen, K. & Koudas, N., “Generative Engine Optimization: How to Dominate AI Search”, 2025. arXiv
- [R4] Kaiser, M. & Schulze, C., “ChatGPT Referrals to E-Commerce Websites”, SSRN Working Paper, 2025. SSRN
- [R5] Puntoni, S., “AI Is Upending Marketing on Two Fronts”, Harvard Business Review, 2026-02-23. HBR
