The multi-tenant SEO CMS platform I built now runs content for 12-plus client sites, and the questions coming from clients have shifted noticeably in the last year. It used to be “why aren’t we ranking.” Now it’s increasingly “why did an AI answer engine quote our competitor and not us.”
Those are different problems, and most marketing teams are still solving only the first one.
Two audiences now, not one
Every piece of content has to work for two different readers: the person who clicks through from a search results page, and the model that summarises your page into an answer without ever sending that person to your site. Optimising purely for the first reader, the way SEO has worked for twenty years, leaves the second one entirely unaddressed.
Generative engine optimisation, GEO, is the practice of writing for both. It’s not a separate discipline that replaces SEO, it’s an additional constraint layered on top of it.
What actually changes in the writing
Answer the question in the first two sentences, not the fifth paragraph. A model extracting an answer isn’t going to read your entire introduction to find it. Neither, honestly, is a human anymore.
Be citable, not just readable. A specific number, a named source, a clear claim someone could quote directly is far more likely to get pulled into a generated answer than three paragraphs of scene-setting.
Structure for extraction. Headings that actually describe the content under them, not clever wordplay. Lists where a list is the honest format. This isn’t about gaming a system, it’s about being unambiguous, which happens to help both readers at once.
What doesn’t change
Good content still needs a real point of view and real evidence behind it. GEO doesn’t reward generic, and if anything it punishes it harder, because generic content is exactly what a model will summarise and never bother to attribute. Specificity was always the moat. It’s a bigger one now.