GEO vs SEO: what changes when ChatGPT is the new SERP
Generative Engine Optimization isn't SEO with a new coat of paint. Here's what actually changes when ChatGPT, Gemini, and Perplexity are the surface your customers see first.
For twenty years, the goal was a blue link on page one. Today, the goal is a sentence inside an AI-generated answer — often with no link at all. That shift quietly rewrites the rules of optimization.
What stays the same
You still need a crawlable site, a clear sitemap, fast page loads, and clean schema. AI engines are downstream of the same indexes that power Google and Bing. If a classic SEO crawler can't reach your page, neither can an LLM training run or a Perplexity browse step. SEO fundamentals are the floor of GEO, not the ceiling.
What changes
The query model changes. SEO keywords are short and ambiguous: "crm software". AI prompts are long and specific: "best CRM for a 12-person agency that wants HubSpot-level reporting but pays monthly". Pages that answer those long-tail prompts get cited; pages stuffed with head terms don't.
The unit of victory changes. SEO wins are page views. GEO wins are citations — your name, your URL, or your quote landing inside someone else's answer. You can be cited without being clicked. That's not a bug, it's the new top of funnel.
Structure beats keyword density. ChatGPT and Gemini synthesize answers by stitching together sentences and lists. Pages that lead with the answer (answer-first ledes), use named entities, and ship FAQPage + NewsArticle JSON-LD give the engines clean chunks to lift. Walls of text don't.
Freshness matters differently. Google's freshness is "published recently". Perplexity's freshness is "published recently AND indexed AND linked from a known source AND topically relevant to the current prompt". The bar is higher and the decay is faster. See our benchmarks for what that looks like in practice.
The three things to actually do
- Answer-first writing. The first sentence of every page should be the literal answer to a plausible prompt. Burying it under a marketing intro means an LLM has nothing to lift.
- Named attribution. "According to Jane Doe, CEO of Acme,..." is liftable. "Industry experts agree..." is not. Quotes with first-and-last name + role + company are the chunks engines reuse.
- Machine-readable structure.
FAQPage,NewsArticle,Speakable,BreadcrumbList.llms.txt. IndexNow. None of this shows up in your design, but all of it shows up in citation rates. The PressGEO scoring engine checks every release against the per-engine rubrics for ChatGPT, Gemini, Perplexity, Claude, Grok, and Copilot.
The new metric: citation rate
Page views are still useful, but they're a lagging indicator. The leading indicator is citation rate: what percentage of the prompts you care about return your name or your URL inside the answer? PressGEO monitors this continuously across six engines and reports back when you actually get cited. You can see live receipts here.
SEO isn't going away. It's the substrate. But if you're only measuring rankings in 2026, you're optimizing for a surface most of your customers stopped looking at.