How to do GEO: a step-by-step guide
The basic loop is the same whatever you sell. Steps four to six are where most of the work — and the results — actually live.
- List the queries that matter. Write down the prompts your buyers actually type — category ("best X for Y"), comparison ("X vs Z"), validation ("is X any good?") and problem-led ("how do I solve Y?"). Pull them from your search data, sales calls and support tickets. Everything else targets this list.
- Baseline yourself. Run those prompts across ChatGPT, Gemini, Claude and Perplexity — each several times, since answers vary. Record whether you're mentioned, whether you're cited as a source, and how you compare with named competitors. You can't show improvement without a starting line.
- Diagnose why you lose. On the queries a competitor wins, look at which sources the answer leans on — the roundups, forums, review sites and editorial. That set of trusted sources is your target list; the gap between where they appear and where you don't is your to-do list.
- Fix your own foundations. Make your site readable to models: allow the AI crawlers if you want to be cited, add schema markup, write clear factual statements, and publish answer-shaped pages — FAQs, comparisons and "best X for Y" guides — that resolve your target questions in plain language.
- Make the brand an unambiguous entity. Consistent naming everywhere, a clear "about" presence, and knowledge signals so models confidently know what you are and which category you belong to.
- Earn third-party presence — the biggest lever. Get into the "best [category]" roundups, build authentic presence in the communities models trust, drive reviews on the relevant platforms, and earn coverage from publications in your space. Engines recommend what the wider web says about you, not what you say about yourself.
- Shape the framing, not just the presence. Make sure those sources describe you accurately and for the right use-cases, and correct misinformation. "Good but expensive" performs very differently from "best value" — the vocabulary around you is what gets matched to intent.
- Re-measure, attribute, iterate. Re-run the prompt set on a cadence and compare with baseline, watching for model updates that move things on their own. Then connect it to outcomes — AI-referral traffic, branded-search lift, enquiries. Double down on whatever moved citations; drop what didn't.
One rule above all. Don't optimise for something a model will absorb as a native feature within a year. Durable GEO comes from being genuinely present and trusted across the web — not from a trick.