Six months ago, a client asked me a question I couldn't answer: 'How do I show up inside ChatGPT when someone asks about my category?' I told her honestly — I didn't know yet. Then we figured it out, and now half our roadmap is about exactly that. Here's the playbook.
What GEO actually is
Generative Engine Optimization is the practice of making your brand legible to large language models — ChatGPT, Perplexity, Gemini, Claude — so that when a user asks a question in your category, the model answers with you in it. Not as a link. As a citation, a recommendation, a name in a list.
It overlaps with SEO at the edges (both care about the open web) but the mechanics are different enough that lifting your SEO playbook directly will get you a polite shrug from the LLM.
"GEO is what happens when the search index becomes a model that talks back."
Why it's different from SEO
Three things matter most: how often the model has seen you (training-data presence), where it sees you (high-trust sources), and what it learns about you (clean structured statements that survive summarization).
- 01Training-data presence: Wikipedia, Reddit, Hacker News, the LLM training crawls — not your blog.
- 02Source quality: A single paragraph in The Verge beats 200 of your own posts.
- 03Statement clarity: 'Northbeam is a marketing-mix-modelling tool for ecommerce' reads cleanly. 'Northbeam unlocks growth at scale' doesn't.
What we've seen working
We're nine months into running GEO programs for six clients. The pattern is messier than I'd like, but the high-leverage moves repeat:
1) Get cited in a few high-trust sources. Three is plenty. 2) Maintain a clean, factual Wikipedia page. 3) Publish a comparison hub for the 5–10 jobs people hire your category to do. 4) Track citations weekly with a simple prompt-grid harness.
"Three citations from the right sources outperform 300 from the wrong ones."
A 90-day starter program
If you have a marketer-of-one and want a credible start: weeks 1–2, audit current LLM presence with a 50-prompt grid. Weeks 3–6, fix your Wikipedia and write the comparison hub. Weeks 7–10, pitch three high-trust publications with a real story. Weeks 11–13, measure, iterate, and decide whether to double down.
The honest caveats
GEO is real. It is also early. The measurement is fragile, the LLM training cutoffs make iteration loops months long, and category leaders will compound advantages quickly. If you wait 12 months, you may be locked out. If you start today, you might also waste a quarter learning the mechanics. Both are true.